The effect of trait anxiety on the generalisation of fear acquisition and extinction
Hon Ki Wong Bachelor of Psychology (Honours)
Submitted in partial fulfilment of the requirements for the degree of Doctor of Philosophy November 2018
School of Psychology
Faculty of Science Thesis/Dissertation Sheet
Surname/Family Name : Wong Given Name/s : Hon Ki (Alex) Abbreviation for degree as give in the University calendar : BCom / BSc(Hons) Faculty : Science School : Psychology The effect of trait anxiety on the generalisation of fear acquisition and Thesis Title : extinction
Abstract 350 words maximum: Fear generalisation refers to the spread of fear to novel situations. Recent evidence has suggested that over-generalisation of fear is a pathogenic marker of anxiety disorders. Given that trait anxiety has been widely accepted as a vulnerability factor for developing an anxiety disorder, the current thesis aimed to examine whether trait anxious individuals show over-generalisation of fear like their clinical counterparts. Using a continuous perceptual dimension, the first experiment (Chapter 5) identified various generalisation gradients, which aligned logically with participants’ reported rules. Trait anxious individuals showed over-generalisation of fear to the novel test stimuli, but this pattern was only observed among those who failed to identify a clear rule. The following experiments (Chapter 6) further examined fear generalisation to objects that were conceptually related to the threat cues. Trait anxious individuals did not show more fear to novel exemplars that had clear categorical membership and therefore clear threat value. However, they showed more fear to novel exemplars that could be classified in both threat and safe categories, that is, exemplars with ambiguous threat value. The results supported the notion of threat appraisal bias under ambiguous threat among trait anxious individuals. The experiments in Chapter 7 examined the effect of trait anxiety on the generalisation of extinction learning along a blue- green stimulus dimension. Participants who received a generalisation stimulus (GS) in extinction showed an increase in conditioned fear to the original conditioned stimulus (CS) or to another novel GS in test. Conversely, this pattern was not found in those who received standard extinction with the CS. No trait anxiety effect was observed in the generalisation of extinction learning, however, trait anxious individuals showed slower fear extinction to the CS, but not to a GS. In summary, the present work suggests that over-generalisation of fear and resistance in fear extinction may be a special case of the more general principle that trait anxiety is associated with excessive threat appraisal under conditions of ambiguity. It also highlights the importance of higher-order cognitive processes in human fear generalisation. The current findings have important clinical implications. Specifically, they suggest the importance of targeting cognitive reappraisal and strategies that reduce situational ambiguity in clinical treatments.
Declaration relating to disposition of project thesis/dissertation
I hereby grant to the University of New South Wales or its agents the right to archive and to make available my thesis or dissertation in whole or in part in the University libraries in all forms of media, now or here after known, subject to the provisions of the Copyright Act 1968. I retain all property rights, such as patent rights. I also retain the right to use in future works (such as articles or books) all or part of this thesis or dissertation.
I also authorise University Microfilms to use the 350 word abstract of my thesis in Dissertation Abstracts International (this is applicable to doctoral theses only).
…………………………………………………………… ……………………………………..……………… ……….……………………...…….… Signature Witness Signature Date The University recognises that there may be exceptional circumstances requiring restrictions on copying or conditions on use. Requests for restriction for a period of up to 2 years must be made in writing. Requests for a longer period of restriction may be considered in exceptional circumstances and require the approval of the Dean of Graduate Research.
FOR OFFICE USE ONLY Date of completion of requirements for Award:
ORIGINALITY STATEMENT
‘I hereby declare that this submission is my own work and to the best of my knowledge it contains no materials previously published or written by another person, or substantial proportions of material which have been accepted for the award of any other degree or diploma at UNSW or any other educational institution, except where due acknowledgement is made in the thesis. Any contribution made to the research by others, with whom I have worked at UNSW or elsewhere, is explicitly acknowledged in the thesis. I also declare that the intellectual content of this thesis is the product of my own work, except to the extent that assistance from others in the project's design and conception or in style, presentation and linguistic expression is acknowledged.’
Signed ……………………………………………......
Date ……………………………………………......
INCLUSION OF PUBLICATIONS STATEMENT
UNSW is supportive of candidates publishing their research results during their candidature as detailed in the UNSW Thesis Examination Procedure.
Publications can be used in their thesis in lieu of a Chapter if: • The student contributed greater than 50% of the content in the publication and is the “primary author”, ie. the student was responsible primarily for the planning, execution and preparation of the work for publication • The student has approval to include the publication in their thesis in lieu of a Chapter from their supervisor and Postgraduate Coordinator. • The publication is not subject to any obligations or contractual agreements with a third party that would constrain its inclusion in the thesis
Please indicate whether this thesis contains published material or not. ☐ This thesis contains no publications, either published or submitted for publication
Some of the work described in this thesis has been published and it has been ☐ documented in the relevant Chapters with acknowledgement
This thesis has publications (either published or submitted for publication) ☐ incorporated into it in lieu of a chapter and the details are presented below
CANDIDATE’S DECLARATION I declare that: • I have complied with the Thesis Examination Procedure • where I have used a publication in lieu of a Chapter, the listed publication(s) below meet(s) the requirements to be included in the thesis. Name Signature Date (dd/mm/yy)
COPYRIGHT STATEMENT
‘I hereby grant the University of New South Wales or its agents the right to archive and to make available my thesis or dissertation in whole or part in the University libraries in all forms of media, now or here after known, subject to the provisions of the Copyright Act 1968. I retain all proprietary rights, such as patent rights. I also retain the right to use in future works (such as articles or books) all or part of this thesis or dissertation. I also authorise University Microfilms to use the 350 word abstract of my thesis in Dissertation Abstract International (this is applicable to doctoral theses only). I have either used no substantial portions of copyright material in my thesis or I have obtained permission to use copyright material; where permission has not been granted I have applied/will apply for a partial restriction of the digital copy of my thesis or dissertation.'
Signed ……………………………………………......
Date ……………………………………………......
AUTHENTICITY STATEMENT
‘I certify that the Library deposit digital copy is a direct equivalent of the final officially approved version of my thesis. No emendation of content has occurred and if there are any minor variations in formatting, they are the result of the conversion to digital format.’
Signed ……………………………………………......
Date ……………………………………………......
i
Acknowledgements
First and foremost, I would like to express my deepest gratitude to my PhD advisor
Prof. Peter Lovibond. Peter, I still remember our first meeting, where you spent two hours introducing the literature of associative learning to me. It was your enthusiasm and unique insight into research that inspired me to pursue a career in research. During my PhD, you were never too busy to give me advice – whenever I popped my head into your office
(without an appointment), you always greeted me with a smile and took time to answer all my questions! I especially appreciate your effort to squeeze time out for us from your extremely busy schedule when you took on the role of Acting Dean. Your advice was always constructive - my knowledge in scientific research grew a bit more every time I left your office. Your concern for your students’ future career is also noteworthy – I particularly enjoyed the conversations when you explained how the academic world works. I have often heard that pursuing a PhD is a draining experience; however, my experience in the Lovibond lab was enjoyable, intellectually stimulating and fun.
I would also like to thank my secondary advisor Prof. Brett Hayes. Brett, you were always just an email away, and thank you for all the vital advices for my thesis!
I would also like to thank the various members in the Lovibond lab:
To Michelle Satkunarajah: You were the first lab member I met when I joined the
Lovibond lab. At first, you seemed like a quiet and shy girl – I was never so wrong. However, this makes you a great friend. Throughout the years, your friendliness and kindness may have treated my social anxiety. I enjoyed your company, which brightens up the lab especially during dreadful days of testing. Thank you for introducing me to fencing, which is a good excuse to stab others with a cool looking sword, and for bringing good coffee to the lab regularly! ii
To Jessica Lee: You are the role model in our lab! I have learnt a lot from you since you have joined the lab. You have shown all the characteristics of a good scientist – to name a few: highly motivated to learn, open-minded, attentive to details and well organized (though not as organized as my desktop). I appreciated the career advice you gave me and enjoyed our discussion on human learning - I hope you still remember our bet! I look forward to our future collaboration.
To Jessamine Chen and Anuja Ng: Thank you both for your company in the lab!
Jessamine, I appreciated your career advices, and Anuja, your positive attitude to everything. I am afraid I cannot spoil you with snacks anymore.
To all the Honours kids that have worked in the lab (Nic, Yvonne, Hannah and
David): It is enjoyable to see your enthusiasm in research. I can selfishly say that I am glad that some of you decided to pursue a research career. I also enjoyed the various topics of conversation over long days of testing, ranging from the meaning of life to which one is the best Pokemon. I wish you all a bright future!
To the UNSW human associative learning group: Mike, Tom, Oren, David, Adrian and Daniel, thank you for listening to the same research topic over the years and were still not jaded to offer critical feedback.
I would also like to extend my gratitude to Prof. Tom Beckers and the CLEP group in
Leuven, Belgium. Thank you for your hospitality during my lab visit! I have definitely learnt a lot more about research skills and beer after my stay in Leuven.
Finally, I would like to thank my family. To my father, Raymond, thank you for your unconditional support and never ask the forbidden question ‘When are you going to graduate?’. To my twin brother Alan, our intense academic competition has pushed me to iii always strive for more – so here I am submitting my thesis earlier than you. Nonetheless, I wish you much success in your research career.
iv
Abstract
Fear generalisation refers to the spread of fear to novel situations. Recent evidence has suggested that over-generalisation of fear is a pathogenic marker of anxiety disorders. Given that trait anxiety has been widely accepted as a vulnerability factor for developing an anxiety disorder, the current thesis aimed to examine whether trait anxious individuals show over-generalisation of fear like their clinical counterparts.
Using a continuous perceptual dimension, the first experiment (Chapter 5) identified various generalisation gradients, which aligned logically with participants’ reported rules.
Trait anxious individuals showed over-generalisation of fear to the novel test stimuli, but this pattern was only observed among those who failed to identify a clear rule. The following experiments (Chapter 6) further examined fear generalisation to objects that were conceptually related to the threat cues. Trait anxious individuals did not show more fear to novel exemplars that had clear categorical membership and therefore clear threat value.
However, they showed more fear to novel exemplars that could be classified in both threat and safe categories, that is, exemplars with ambiguous threat value. The results supported the notion of threat appraisal bias under ambiguous threat among trait anxious individuals.
The experiments in Chapter 7 examined the effect of trait anxiety on the generalisation of extinction learning along a blue-green stimulus dimension. Participants who received a generalisation stimulus (GS) in extinction showed an increase in conditioned fear to the original conditioned stimulus (CS) or to another novel GS in test. Conversely, this pattern was not found in those who received standard extinction with the CS. No trait anxiety effect was observed in the generalisation of extinction learning, however, trait anxious individuals showed slower fear extinction to the CS, but not to a GS.
In summary, the present work suggests that over-generalisation of fear and resistance in fear extinction may be a special case of the more general principle that trait anxiety is v associated with excessive threat appraisal under conditions of ambiguity. It also highlights the importance of higher-order cognitive processes in human fear generalisation. The current findings have important clinical implications. Specifically, they suggest the importance of targeting cognitive reappraisal and strategies that reduce situational ambiguity in clinical treatments.
vi
Table of Contents
Chapter 1: Anxiety disorders and fear learning……………………………………... 1 »›Anxiety disorders…………………………………………...... 2 »›Fear learning: The etiology of anxiety disorders…………………………………. 3 »›Fear learning in non-human animals………………………………………….. 4 »›Associative accounts of learning………………………………………………. 7 »›Fear learning in humans………………………………………………………. 11 »›Dual-system or single-system learning models?...... 14 »›Summary……………………………………………………………………….. 18
Chapter 2: Generalisation in animals and humans…………………………………... 20 »›Generalisation…………………………………………………………………. 21 »›Animal studies in generalisation………………………………………………. 22 »›Associative accounts for generalisation……………………………………….. 26 »›Human studies in generalisation………………………………………………. 31 »›Summary……………………………………………………………………….. 37
Chapter 3: Trait anxiety and anxiety disorders……………………………………… 39 »›Anxiety proness………………………………………………………………… 41 »›The measurement of trait anxiety……………………………………………… 48 »›Cognitive approach to anxiety………………………………………………… 50 »›Threat appraisal in fear learning……………………………………………… 57 »›Fear learning in trait anxiety………………………………………………….. 57 »›Continuity of trait anxiety……………………………………………………… 58 »›Summary……………………………………………………………………….. 61
Chapter 4: The effect of clinical and trait anxiety on fear generalisation………….. 62 »›Clinical anxiety and fear generalisation………………………………………. 63 »›Trait anxiety and fear generalisation………………………………………….. 65 »›Strong and weak situations…………………………………………………….. 67 »›The effect of trait anxiety on fear learning in the presence of ambiguity……… 68 »›Ambiguity and uncertainty……………………………………………………... 71 »›Summary and research aims…………………………………………………… 72
Chapter 5: The effect of trait anxiety on fear generalisation across a perceptual continnum………………………………………………………………….. 74 vii
»›Experiment 1 Method…………………………………………………………………... 78 Results…………………………………………………………………… 84 Discussion……………………………………………………………….. 96 »›Chapter summary……………………………………………………………… 105
Chapter 6: The effect of trait anxiety on fear generalisation to conceptually related objects…………………………………………………………….. 106 »›Validation Questionnaires Method…………………………………………………………………… 117 Results...... 120 Discussion………………………………………………………………... 123 »›Experiment 2 Method…………………………………………………………………… 124 Results…………………………………………………………………..... 131 Discussion………………………………………………………………... 136 »›Experiment 3 Method……………………………………………………………………. 137 Results……………………………………………………………………. 138 Discussion………………………………………………………………… 147 »›General discussion…………………………………………………………….. 149 »›Chapter Summary………………………………………………………………. 157
Chapter 7: The effect of trait anxiety on the generalisation of extinction………….. 158 »›Experiment 4 Method…………………………………………………………………… 170 Results……………………………………………………………………. 175 Discussion………………………………………………………………… 182 »›Experiment 5 Method……………………………………………………………………. 186 Results……………………………………………………………………. 188 Discussion……………………………………………………………….. 195 »›General discussion…………………………………………………………….. 196 »›Chapter summary……………………………………………………………… 203 viii
Chapter 8: General discussion………………………………………………………… 204 »›Key empirical findings………………………………………………………… 205 »›Trait anxiety……………………………………………………………………. 208 »›Strong and weak situations…………………………………………………….. 217 »›Cognitive and associative accounts of human generalisation………………… 218 »›Dual-system or single-system learning models?...... 221 »›Clinical implications…………………………………………………………… 225 »›Limitations and future directions……………………………………………… 228 »›Conclusions…………………………………………………………………….. 233
References………………………………………………………………………………. 234
Appendices……………………………………………………………………………… 311 »›Appendix A……………………………………………………………………. 311 »›Appendix B……………………………………………………………………. 338 »›Appendix C……………………………………………………………………. 371
ix
List of Tables and Figures
Table 1. Design of Experiment 1……………….……………………………………..... 80
Table 2. Examples of actual responses in questionnaires………………………………. 91
Table 3. Number of participants in each rule subgroup………………………………… 91
Table 4. Names of the pictures presented in the Typicality rating and Free
categorization groups…………………………………………………………. 118
Table 5. Design of Experiment 2……………………………………………………….. 125
Table 6. Design of Experiment 4……………………………………………………….. 172
Table 7. Design of Experiment 5……………………………………………………….. 188
Figure 1. Generalisation gradients in animal studies……………………………………. 24
Figure 2. Peak shifted gradients in pigeons……………………………………………… 26
Figure 3. The summation of the excitatory and inhibitory gradients……………………. 28
Figure 4. Model of personality………………………………………………………….. 44
Figure 5. Theory of appraisal and the four-factor theory of anxiety……………………. 52
Figure 6. Stimulus dimension from Lissek et al.’s generalisation paradigm……………. 63
Figure 7. Stimulus dimension used in Experiment 1……………………………………. 79
Figure 8. Mean expectancy and SCL data in Acquisition (Experiment 1)……………… 86
Figure 9. Mean expectancy and SCL gradients in test (Experiment 1)…………………. 88
Figure 10. Expectancy and SCL gradients for each rule subgroup (Experiment 1)……... 92
Figure 11. Comparison of generalisation gradients between anxiety groups within
each rule subgroup………………………………………………………….. 94
Figure 12. Typicality ratings for exemplars in Validation Questionaires……………… 121
Figure 13. Mean expectancy and SCL data in Acquisition (Experiment 2)……………. 132
Figure 14. Mean expectancy and SCL data in test (Experiment 2)…………………….. 134 x
Figure 15. Mean expectancy and SCL data in Acquisition (Experiment 3)……………. 140
Figure 16. Mean expectancy and SCL data in test (Experiment 3)……………………... 142
Figure 17. Mean expectancy and SCL data to the cross-classified exemplars according
to participants’ classification………………………………………………… 146
Figure 18. Stimuli used in Experiment 4………………………………………………… 171
Figure 19. Mean expectancy and SCL data in Experiment 4……………………………. 176
Figure 20. Stimuli used in Experiment 5………………………………………………… 187
Figure 21. Mean expectancy and SCL data in Experiment 5……………………………. 190
1
Chapter 1:
Anxiety disorders and fear learning
2
Anxiety disorders
Anxiety disorders, including posttraumatic stress disorder (PTSD), generalised anxiety disorders (GAD), panic disorder (PD), obsessive compulsive disorder (OCD) and specific phobias, are one of the most prevalent mental disorders (American Psychiatric Association,
2013). The lifetime prevalence rate of anxiety disorders is as high as 20% in Australia
(McEvoy, Grove & Slade, 2011), and up to 33.7% worldwide (Bandelow & Michaelis, 2015).
Patients suffering from anxiety disorders often have impaired physical activity, social and family relationships, daily functioning and work productivity as well as substance abuse
(DuPont et al., 1996; Mendlowicz & Stein, 2000; Olatunji, Cisler & Tolin, 2007). This, as one would expect, is detrimental to the quality of life, subjective life satisfaction and well-being.
Substantial economic loss is also incurred by the high prevalence of anxiety disorders. It was estimated that anxiety disorders have cost around $42-47 billion per annum in the United
States (DuPont et al., 1996; Greenberg et al., 1999). A more recent study estimated a cost of
$12.8 billion per annum by anxiety disorders in Australia (Lee et al., 2017). The economic loss is largely due to clinical treatment costs, and monetary cost due to decreased work productivity caused by morbidity and mortality (Kessler & Greenberg, 2002).
In light of the detrimental value of anxiety disorders, considerable research has been done to improve the effectiveness of clinical treatments. One line of research, experimental psychopathology, uses experimental means to understand the underlying mechanisms of anxiety disorders, constructs theoretical models and translates them into clinical trials. These experimental psychopathology models provide strong construct validity, good diagnostic validity and predictive validity (Scheveneels, Boddez, Vervliet & Hermans, 2016; Vervliet &
Raes, 2013), and have led to effective treatments such as exposure and response prevention
(Baum, 1970; Pittig et al., 2015, 2018). These models emphasize the etiology of the acquisition of maladaptive fear in anxiety disorders and the means to decrease it. 3
Fear learning: The etiology of anxiety disorders
Although both fear and anxiety are similar constructs and often occur together, they are slightly different from each other. Fear is a basic emotion that appears when an imminent threat is detected (Ekman, 1992), while anxiety refers to a more long lasting state of distress, usually prompted by an anticipation of a more distant threat (see Lang, Davis & Öhman,
2000). Fear can be expressed in three different response systems (Beckers, Krypotos, Boddez,
Effting & Kindt, 2013; Frijda, 1986; Lang, Bradley & Cuthbert, 1998). These response systems include physiological activity (e.g., skin conductance), overt behaviour (e.g., avoid or approach) and cognitive belief (e.g., threat expectancy), of which the last is generally assumed to be unique in humans. In most cases, fear serves adaptive purposes. For instance, evolutionary sculptured fear allows adaptive responses to potential threat, increasing the chances of survival (LeDoux, 2012). However, patients suffering from anxiety disorders often show maladaptive, excessive fear (American Psychiatric Association, 2013).
One way to further our understanding of fear is to investigate why some seemingly neutral objects or situations (e.g., a car) are capable of triggering a tremendous amount of fear in clinically anxious patients (e.g., PTSD patients who had experienced a traumatic car accident). Research in associative learning provides important insight in this regard.
Associative learning refers to a learning process where an associative link between two stimuli or between a stimulus and a behavioural response is formed in the environment.
Associative learning can be studied in a classical conditioning paradigm. A classic empirical demonstration of associative learning comes from Pavlov (1927). In this experiment, Pavlov paired a ringing bell with food delivery to his dogs. Upon repeated pairings, the dogs would salivate on hearing the sound of the bell even with the absence of food. The ringing bell here acts as the conditioned stimulus (CS) and the food acts as the natural biological unconditioned stimulus (US). Prior to the CS-US pairing, the presence of food (US) is able to trigger 4 salivation, the unconditioned response (UR), an unlearnt response that occurs naturally in reaction to the US. Upon repeated pairing of the CS and the US, salivation (UR) occurs when the bell is presented, referred to the conditioned response (CR). That is, the CR is triggered by the initially biologically neutral CS as a function of repeated CS-US pairing. Whilst using an appetitive US triggers an appetitive CR, such as salivation, the pairing of the CS and an aversive US triggers a conditioned fear reaction. This procedure is known as fear conditioning. It has long been suggested that the maladaptive fear component in anxiety disorders is acquired via conditioning (Mowrer, 1939; Watson & Morgan, 1917). Indeed, the process of conditioning is considered to be analogous to the etiology of anxiety disorders (see
Mineka & Zinbarg [2006] for a review). For instance, Dr. W develops PTSD after a traumatic car accident, and subsequently shows intense fear to all sorts of vehicles. The car, an object that was initially neutral, is analogous to the CS, while the traumatic accident itself acts as the aversive US. Having a car accident whilst riding in the car acts as the fear conditioning process, pairing the neutral CS and the aversive US together. As a result, Dr. W shows a great amount of conditioned fear with the presentation of CS (vehicles).
Fear learning in non-human animals
In light of exploring the pathogenesis of anxiety disorders experimentally, there was initial interest in investigating abnormal behaviours in animals using conditioning paradigms in the 1930s and 1940s (e.g., Anderson & Liddell, 1935; Gantt, 1942; Maier, 1940), inspired by Pavlov (1927). This approach was termed experimental neurosis and typically involved animals going through a discrimination task between a reinforced stimulus (CS+) and a non- reinforced stimulus (CS-). These discrimination tasks were thought to be demanding since the
CSs were highly similar and hence difficult to discriminate between them. The CS+ was usually paired with an aversive stimulus. Researchers were interested in how the conditioning process may induce erratic behaviours or abnormal physiological activities (i.e., CRs) in 5 animals. However, these studies were largely unsystematic and the independent variables were poorly controlled (see Mineka, 1985). For example, the CRs were investigated in an exploratory way and behavioural changes were interpreted as pathological in a post-hoc way.
Furthermore, due to the dominance of psychoanalysis in explaining the etiology of mental disorders at the time (Freud, 1920), work in experimental neurosis was overshadowed and hence largely dismissed. Nonetheless, the work in experimental neurosis laid the foundation of later fear conditioning studies in animals.
In the 1960s, interest in fear conditioning in non-human animals was re-ignited as fear had been seriously considered as a classically conditioned response – variables that affect CRs in non-fear conditioning paradigms were found to affect the magnitude of conditioned fear as well (Brown, 1961; McAllister & Mcallister, 1971). For example, the number of conditioning trials (e.g., Desiderato, 1964; Kamin, Brimer & Black, 1963), temporal interval between CS and US (e.g., Lyon, 1963), intensity of US (e.g., Annau & Kamin, 1961; Goldstein, 1960) and the rate of CS+ reinforcement (e.g., Brimer & Dockrill, 1966; Willis & Lundin, 1966) can affect the magnitude of conditioned fear. Furthermore, more sophisticated animal models had been proposed (see Mackintosh, 1974) and the parallel between conditioned fear in animals and anxiety syndromes had begun to be accepted (see Davis, 1990; Lal & Emmett-Oglesby,
1983; Shekhar et al., 2001 for reviews), emphasizing the translation value of animal studies.
These animal studies were often carried out in a classical conditioning paradigm or an operant/instrumental conditioning paradigm. The former paradigm was analogous to Pavlov’s
(1927) salivary paradigm, where a neutral CS, usually a tone (e.g., Rescorla, 1968) or a light
(e.g., Dyal & Goodman, 1966), was paired with an aversive US, typically an electric shock
(e.g., Rescorla, 1969a) or an unpleasant loud noise (e.g., Lolordo, 1967). In the following test phase, the CS was presented alone and animals typically showed an increase in magnitude in
CRs. These CRs were automatic, involuntary fear-related responses, including defensive 6 reactions, such as freezing (Weiss, Krieckhaus & Conte, 1968) and startle responses (e.g.,
Brown, Kalish & Farber., 1951; Kurtz & Siegel, 1966), or physiological responses, such as skin conductance (e.g., Anderson, Cole & McVaugh, 1968) and heart rate changes (e.g., Caul
& Miller, 1968) or conditioned suppression (e.g., Kamin, Brimer & Black, 1963; see also
Estes & Skinner, 1941). The typical findings were an increase in conditioned fear with the presence of the CS+, but not the non-reinforced CS (i.e., CS-).
In the latter operant/instrumental conditioning paradigm, the majority of the studies were carried out using appetitive conditioning. A set of stimuli (usually two) are presented, where performing the designated behavioural response to one led to the delivery of an appetitive outcome, whereas responding to the other stimulus led to the omission of the outcome respectively. The former, referred to as the discriminative stimulus (SD) or reinforced stimulus (S+), sets the occasion for responding, while the other, referred to as the
SΔ or S-, sets the occasion for not responding. Animals learns to respond only to S+ but not to
S- across training. Alternatively, responding could be manipulated by presenting or omitting an aversive stimulus. The majority of aversive learning studies carried out via operant/instrumental conditioning were done with active avoidance. In this procedure, if the animal performs the designated response in the presence of S+, the aversive shock would not be administered. For example, rats were first put in a two-compartment box, known as a shuttle box (Warner, 1932). The S+ was presented in one compartment to signal the forthcoming shock. If animals ran into the other compartment in time, the S+ would disappear and the shock would not be delivered. It was found that across trials of S+ presentations, animals learnt to actively avoid the shock by performing the designated behaviours, such as active avoidance and hurdle jumping (e.g., Bolles & Warren, 1965; McAllister & McAllister
1962; McNew & Thompson, 1966; Myers, 1964).
7
Associative accounts of learning
With numerous empirical demonstrations of fear conditioning in animals, early associative theories were developed attempting to account for conditioning in animal in general. Pavlov (1927, 1932) explained the mechanism on a neuronal level. He emphasized the role of temporary contiguity, stating that if the CS and US were delivered close together in time, a CS-US association would be formed and stored in the cerebral cortex. When the CS was presented again, it activated the cortical cells and the conditioned reflex (i.e., CR) would be triggered automatically. On the other hand, Thorndike (1898) proposed the Law of Effect, in an attempt to account for learning in operant/instrumental learning. In his theory, if a particular response (R) in the presence of a stimulus leads to a reward (S+), this response will increase in magnitude or frequency when S+ is presented. Conversely, if a response is followed by a negative stimulus (i.e., punishment), the magnitude of response will decrease in the presence of the stimulus. The Law of effect suggested that responses to a rewarded stimulus over trials would led to the formation of an S-R association. The stronger this association is, the larger the response would be with the presence of the stimulus, and vice versa if S+ is punished with a negative stimulus.
Building on this theory, Hull (1934a, 1934b) proposed the habit-family hierarchy model. According to Hull, an initial S+ presentation would led to the excitation of multiple distinct responses (e.g., lever pressing, nose poking). These distinct responses would compete with each other but only the response that led to reinforcement (i.e., reward) will eventually gain excitatory strength. Further stimulus reinforcement will strengthen the mental stimulus- response link (S-R link) and this link gradually becomes a ‘habit’. The response will then be evoked automatically when the S+ is presented. Applying the S-R theory to appetitive operant/instrumental conditioning studies, Hull (1943) suggested that the magnitude of response is directly correlated to the level of appetitive reinforcement (e.g., food pellets). In 8 other words, the S-R link becomes stronger over S+-reward presentations, and hence automatically triggers stronger behavioural responses in the presence of S+ (e.g., Brown &
Jenkins, 1968; Guttman, 1953; Morse & Skinner, 1958). In order to explain the motivation for behaviour, Hull and Spence also put forward the drive reduction theory (Hull, 1943), which proposed that a state of tension or arousal, namely drive, arises from biological or physiological needs, such as hunger and thirst. Given the unpleasantness caused by a drive, animals often perform behavioural responses that reduce these drives. For instance, the delivery of food pellets after pecking at a keylight (S+) not only strengthens the S-R link, but the consumption of the food pellets also reduces the drive of hunger. Hence, hunger motivates pecking and the reduction of hunger serves as a negative reinforcer of the designated pecking response. Additionally, it had been questioned that why only the designated stimulus gains associative strength and activates the responses, provided that many other stimuli may have been presented in the environment (e.g., background noise, scent of the cage) in training.
Whilst the importance of contiguity between CS and US in classical conditioning had been discussed (Pavlov, 1927), Spence (1945) further emphasized on this point by suggesting that
S+ gain high level of associative strength because of the high contiguity between the response and the reinforcer, while other irrelevant stimuli failed to do so because of their low contiguity with the response.
Despite numerous aversive operant/instrumental conditioning studies that have shown an increase in the designated avoidance behaviour with the presence of S+ (e.g., McAllister &
McAllister, 1962; Mowrer & Solomon, 1954; Solomon & Wyne, 1953), one theoretical problem was how the absence of the reinforcer (e.g., shock) increased the avoidance behaviour. Mowrer’s (1939, 1947, 1960) two factor S-R theory suggested that avoidance learning occurred via two different processes. First, in early training trials when animals had not yet learnt to perform avoidance response in the presentation of S+, they received shock; that is, Pavlovian pairings between S+ and the shock occurs during early training trials. Once 9 animals had learnt that S+ signaled aversive shock, S+ became a conditional driver (see Hull,
1943) that induced fear, and this was suggested to elicit the avoidance response, establishing the S-R link. In other words, the formation of the S-R link in avoidance learning is a combination of the Pavlovian conditioning between warning signal and shock and the pairing between the fearful S+ and responses. The two-factor theory then suggested that avoidance behaviour is reinforced in two ways. First, much alike the Hull-Spence drive reduction theory
(Hull, 1943), the reduction of the state of fear acts as a negative reinforcer of the avoidance behaviour. Secondly, the termination of the warning signal after performing the avoidance behaviour acts as a secondary reinforcer. Since the warning signal induces fear, the subsequent termination of it after avoidance brings out a state of relief that reinforces the avoidance behaviour.
Later associative accounts focused more on the stimulus-stimulus (S-S) connection, for instance, the influential Rescorla-Wagner Model (1972). The Rescorla-Wagner model suggested that prediction error plays an important part in determining associative strength between stimuli. Prediction error refers to the mismatch between prior expectations and reality. Large prediction error typically occurs in early training trials when animals do not expect an US following CS presentation, and the larger the prediction error, the more learning takes place and hence more associative strength between the CS and US. Therefore, there is diminishing increase in prediction error across training trials after animals have learnt that the
CS is a good predictor of the US. When prediction error reaches zero, no learning will occur and associative strength is at asymptote. The idea of prediction error suggests the number of
CS-US pairings relative to the base rate of the US, namely contingency, plays an important part in associative learning. In fact, studies have shown that contingency is the critical factor in the formation of the CS-US association; while contiguity is necessary, it is not sufficient for S-S learning to occur (see Papini & Bitterman, [1990] for a review). The S-S Rescorla-
Wagner model can also account for other associative learning phenomena that are not 10 predicted by S-R accounts, like blocking (Kamin, 1969), conditioned inhibition (Rescorla,
1969b) and some aspects of extinction (see Mackintosh, 1974). For example, blocking involves the presentation of a reinforced stimulus (i.e., A+), followed by the pairing of a compound consisting of the original stimulus and a novel stimulus, again followed by the outcome (i.e., AB+). The novel stimulus is then shown alone (i.e., B), and CRs to the novel cue are found to be low compared to a control cue (e.g., an elemental cue that has only been previously presented in compound with reinforcement). According to the Rescorla-Wagner model, prediction error when AB+ is presented is low since A is an established predictor of an outcome; therefore learning to B is blocked by A.
Nonetheless, these late associative accounts followed their predecessors and emphasized the link formation between representations, be it the link between stimulus and response (S-R), or between stimuli (S-S). These links are suggested to be triggered in an automatic manner; the presentation of a reinforced stimulus can automatically trigger the mental presentation of a stimulus or a reinforced behaviour via the associative link (see
McLaren et al., 1994; Rescorla & Wagner, 1972).
The presumed automatic nature of fear conditioning in animals is appealing for explaining the intense fear among anxiety patients. First, since fear is suggested to be triggered automatically by the presence of a threat cue, this potentially explains the uncontrollable fear in anxiety patients. Secondly, it may explain the etiology of anxiety disorders involving irrational fear, for instance, specific phobias (e.g., needle phobia). Once fear is acquired towards a specific object (e.g., syringe), fear is triggered by the presence of the object despite patients acknowledging the object being non-harmful. An automatic link- based mechanism provides a ready account of such irrational fear (see Öhman & Soares,
1993; Shanks, 2007). 11
Fear learning in humans
As previously mentioned, fear can be measured by three different response systems, namely physiology, behaviour and cognition (Beckers et al., 2013; Lang et al., 1998).
Although animal studies in fear conditioning provide great translation value, fear can only be measured in the first two constructs, whilst cognition is arguably unique to humans (see
Davey, 1987; MacLean, 2016). Nonetheless, many parallels have been found between fear conditioning studies in humans and animals. For instance, like animals, humans show an increase in conditioned fear after presentations of CS-US pairings. Similar to animal studies, the CRs in humans are measured in presumably low-level defensive reflexes, like eyelid reflex (e.g., Spence & Runquist, 1958; Spence & Taylor, 1951), physiological aspects such as skin conductance (e.g., Moeller, 1954; White & Schlosberg, 1952) and heart rate (Notterman,
Schoenfeld & Bersh, 1952; Zeaman, Deane & Wegner, 1954) or by behavioural means such as reaction time of response (e.g., Grim & White, 1965). One advantage of human studies over animals is the availability of direct measurement of cognitive processes during associative learning (see Davey, 1987). Coinciding with the shift in paradigm from behaviourism to cognitivism, there has been a sharp increase in human fear conditioning studies in the last few decades. Cognitive processes are measured by participants’ declarative knowledge of the CS-US contingencies, reflected in post-experimental questionnaires (e.g.,
Fuhrer & Baer, 1969; Lacey & Smith, 1954), but more often measured by self-reported online
US expectancy ratings (Dawson & Biferno, 1973; Furedy & Schiffman, 1973).
In fact, the concept of US expectancy in a broad sense has been discussed extensively in the animal literature, and was first put forward by Pavlov (1927). He suggested that the CS signals the occurrence of US, and the CR from animals is an anticipatory response. That is, the reflexive CR allows the organism to be ready for or expect the US (cf. Culler, 1938). A few years later, Tolman (1932) boldly proposed that CR to the CS is cognitively driven, that is, animals know to expect an US when the CS is presented, and the CR is derived from this 12 knowledge. Putting this into the context of fear learning, the conditioned fear can be viewed as an animal’s preparatory responses for the imminent threat. However, with the dominance of associative accounts at the time, this view was largely disregarded. Since then, it has been presumed that humans acquire fear via an automatic, link-based associative mechanism like animals are also assumed to do (e.g., Clark & Squire, 1998; see also Mineka & Öhman,
2002). However, recent empirical evidence has pointed to the importance of cognition in human fear learning, and associative learning in general.
First, empirical studies have found that awareness plays an important role in associative learning. Some studies used a masking task to minimize the probability of participants being aware of the CS-US contingency (e.g., Dawson, 1970; Dawson & Reardon,
1973), in which participants were asked to discriminate between two stimuli, and electric shock would be administered occasionally during the study. Unbeknown to the participants, one stimulus was consistently reinforced by the shock (i.e., CS+), and another stimulus was never followed by the shock (CS-). A substantial proportion of participants were not aware of the CS-US contingency, and showed no differential autonomic responses to CS+ and CS-.
However, differential conditioned responses were observed among those aware of the CS-US contingency. In other studies, participants were instructed to rate their US expectancy on a trial-by-trial basis (Biferno & Dawnson, 1977; Dawson & Biferno, 1973; Öhman, Ellstrom &
Bjorkstrand, 1976). Aligned with previous studies, only aware participants showed differential conditioning in skin conductance. Interestingly, the timing of the trend in CSs differentiation in autonomic responses aligned with the US expectancy ratings, suggesting the
CRs to the two CSs only started to diverge once participants became aware of the CS-US contingency. That is, participants did not demonstrate systematic differential responses in the early training phase when they were not aware of the relevant contingency. 13
Secondly, empirical studies have shown that informing participants verbally about the relationship between the CS and US is adequate to produce CRs. A study by Cook and Harris
(1937) demonstrated that after participants were informed verbally that a tone would be followed by a shock, they showed increased autonomic CRs towards the tone even though they had never experienced the direct pairing between the tone and shock. Similarly, after conditioning with a tone paired with a shock, verbal instructions that the tone would not be followed by a shock anymore reduced participants’ autonomic CRs to the CS (Colgan, 1970).
More recently, studies also demonstrated the impact of verbal instructions on fear generalisation in humans (Ahmed & Lovibond, 2015; Vervliet, Kindt, Vansteenwegen &
Hermans, 2010). In these studies, participants were trained with a coloured geometric figure
(e.g., a yellow triangle) paired with a shock. They were informed that a particular physical feature of the CS+ predicted the shock (e.g., the colour) either at the beginning of the study
(Vervliet et al., 2010) or after training (Ahmed & Lovibond, 2015). An increase in skin conductance was observed only to novel stimuli that contained the verbally instructed predictive feature (e.g., yellow square), but not to stimulus that shared the physical feature with the CS+ that was not verbally informed as a predictive feature (e.g., red triangle). As the semantic meaning of verbal instruction is believed to be processed in the cognitive system
(see Winograd, 1983), these studies provide strong evidence that cognitive processes are involved in fear learning in humans.
Thirdly, there is some evidence that associative learning and fear learning are modulated by reasoning, suggested by paradigms that involved cue competition. One of these paradigms, blocking, was pioneered by Kamin (1969). As previously mentioned, the blocking effect is accounted for by certain associative models (e.g., Rescorla & Wagner, 1972), and has been successfully replicated in animals since Kamin (e.g., Lovejoy & Russell, 1967;
Mackintosh & Honig, 1970; Willner, 1978; but see Maes et al., 2016). However, replication of the blocking effect was initially not as successful in humans (Davey & Singh, 1988; 14
Kimmel & Bevill, 1996; Lovibond, Siddle & Bond, 1988; Pellon, Montano & Sanchez,
1990). Mitchell and Lovibond (2002) suggested that in most experiments the outcome was at ceiling in the element phase (e.g., A+ trials) caused by the use of a binary outcome. In other words, because the outcome was at full strength on A+ trials, no increase in effect magnitude could be attributed to B when it was presented in compound with A. This rendered the casual status of B ambiguous, presumably resulting in the weak to no blocking effect. In order to get around the ceiling effect imposed to the element cue, Mitchell and Lovibond (2002) introduced the instruction of outcome-additivity in their autonomic conditioning studies.
Participants in the additivity group were instructed that if two elements predictive of shock
(e.g., X+, Y+) were presented together, the resulting compound would be followed by double shock (e.g., XY++). Such instructions were withheld from the control group. Responding to the blocked cue was significantly lower in the additivity group compared to the control group in the test phase. That means, the blocking effect was successfully replicated, but only in the additivity group. Mitchell and Lovibond (2002) attributed the results to the process of inferential reasoning. They suggested that after the additivity training, when the predictive element (X+) was presented with a novel element (A) in compound reinforced by a single shock (AX+), participants inferred that A was non-predictive; otherwise AX would had be followed by double shock (AX++). This resulted in lower responses to A and hence the blocking effect. Using similar experimental manipulations, the blocking effect in humans has been successfully replicated in other paradigms, for example, causal judgement studies (De
Houwer, Beckers & Glautier, 2002; Lovibond, Been, Mitchell, Bouton & Frohardt, 2003;
Mitchell, Lovibond & Condoleon, 2005).
Dual-system or single-system learning models?
In the extensive literature of animal conditioning, it has been shown that CRs gradually increased across trials of CS-US contingency exposure. Associative accounts 15 propose that across the acquisition period of CS-US contingency, an excitatory link is formed between the two stimuli. The presence of the CS activates this mental link, and triggers the
CR in an automatic fashion. Since humans and animals learn in a similar way in the same conditioning paradigm (e.g., both humans and animals showed increased CRs in a fear conditioning paradigm), this suggests they may both use a common automatic learning system. However, cognition, which is assumed to be unique to humans, is also found to be critically involved in human conditioning. Therefore, it has been proposed that human learning involves both the cognitive system and an automatic, reflexive system, forming the basis of dual-system model (e.g., Clark & Squire, 1998; McLaren et al., 2014). The dual- system models presume that the two systems contribute to learning but in distinctive ways.
The associative system, as discussed before, is said to involve the formation of links between mental representations of stimuli or events. The strength of these links increases with increasing experience of the relationship between stimuli or events, reflected by the increased magnitude of CR across trials of CS-US pairings. It is argued that the mental representation of links is formed in an automatic sense, regardless of whether the relationship between the two events is learnt on a conscious level (Shanks, 2007). The cognitive system is said to be responsible for the development of explicit propositional beliefs about the relationship between events/stimuli, which often involves processes of inductive and deductive reasoning.
Since learning is suggested to be governed by these two independent systems, dual-system models predict dissociations between the CRs controlled by the two systems.
The most striking evidence for dual-system models would be the Perruchet effect
(Perruchet, 1985). In a single-cue eyeblink conditioning paradigm, participants were told that a tone (CS) would be presented for every trial, and had a 50% chance to be followed by an airpuff US. They had to rate their expectancy of the airpuff occurring on the next trial. The trials were either reinforced or non-reinforced on a pseudorandom schedule. Under the gambler’s fallacy, a false belief that something will be less likely to occur in the future if it 16 has occurred frequently in a given period, it was predicted that participants might show lower expectation of the airpuff in the next trial after a series of reinforced trials, and vice versa. The expectancy measures did conform to Perruchet’s prediction; however, the eyeblink CRs showed the opposite pattern: eyeblink CRs increased after consecutive reinforced trials, attributed to the accumulative excitatory associative strength, and decreased after consecutive non-reinforced trials due to the accumulative inhibitory CS-US association. The results suggested that there was a dissociation between the expectancy and eyeblink measures; that is, a dissociation between the cognitive system responsible for expectancy measures, and a low-level, reflexive system responsible for eyeblink CRs, hence providing supportive evidence for the dual-system models. Furthermore, irrational beliefs that are often observed in patients with specific phobias have been regarded as supporting the dual-system models. For example, hemophobics show a great amount of fear to blood despite knowing the fact that blood per se is not harmful. Dual-system accounts may explain this irrational fear as the result of link-based mechanism being triggered automatically, whilst the cognitive system holds the belief that blood is not threatening.
As an alternative to the dual-system models, the propositional model or single system account suggests that associative learning in humans is only accounted for by the cognitive system (e.g., De Houwer, 2009; Mitchell, De Houwer & Lovibond, 2009). It is suggested that the increase in CRs to the CS is derived from the increasing strength of the propositional belief of the CS-US contingency. Proponents of propositional model argue that little evidence exists for a separate associative system in human learning, and they deny the existence of an automatic associative learning process. Indeed, there is considerable evidence that challenges the existence of the automatic learning system (e.g., Dawson & Biferno, 1973; Ross &
Nelson, 1973; Weidemann & Antees, 2012; Weidemann, Satkunarajah & Lovibond, 2016).
Some empirical studies added a secondary task to a differential conditioning paradigm. These secondary tasks were usually presented at the end of CS presentation and aimed to divert 17 participants’ awareness away from the CS-US contingency. For instance, Ross and Nelson
(1973) asked participants to perform a ‘time estimation task’ at the end of each CS trial.
Participants had to press the right button after a certain amount of time had passed since the
CS offset. This kind of experimental manipulation successfully increased the number of participants who were unaware of the CS-US contingencies. However, the results in these studies consistently showed that when participants were unaware of the CS-US contingency, no differential conditioning was observed in either propositional knowledge (e.g., expectancy ratings) or CRs (e.g., eyeblink reflex). These findings violated dual-system accounts’ prediction, as these accounts would predict differential responses in the CRs regardless of participants’ awareness of differential contingencies, since the CRs were believed to be controlled by an independent low level, automatic system. Furthermore, the establishment of
CRs by verbal instructions described earlier (e.g., Colgan, 1970; Cook & Harris, 1937) also questions the existence of an automatic learning system. Associative accounts assert that the development of the mental link between representations requires direct experience with the pairing between CS and US. Therefore, it is hard to see how these models can account for the increase in CRs to the CS by mere verbal instructions of the CS-US contingency. Similarly, without direct experiencing CS-US pairings, participants showed an increase in autonomic
CRs to the CS after merely observing a confederate getting shocked with the presence of the
CS+. Findings like this pose serious challenges to the dual-system models (Olsson, Nearing &
Phelps, 2007; Olsson & Phelps, 2004).
Summing up the human learning literature, it appears that some findings support the dual-system models while others support the single system model. Proponents of the single system view argue that, findings that seemingly support dual-system models are also opened to alternative explanations. For instance, the dissociation between US expectancy and eyeblink CRs reported by Perruchet (1985) can be potentially explained by non-associative effects. Weidemann and Lovibond (2016) suggested that the increase in CRs after consecutive 18 reinforced trials can be attributed to sensitization to the US. The irrational fear commonly seen in phobias can be accounted for by genetic factors (Kendler et al., 1995, 2001) and faulty reasoning (see Lovibond, 2011). In fact, quite a few empirical dissociations in learning, reasoning and decision-making that are commonly interpreted within the dual-system are potentially amenable to a propositional account (Chater, 2003; Kinder & Shanks, 2001;
Newell & Dunn, 2008). Therefore, one substantial advantage of the propositional account over the dual-system models is parsimony.
In the context of anxiety disorders, the propositional model can account for the abnormal, specific threat beliefs commonly seen in anxiety disorders. For example, while acrophobics (i.e., patients afraid of height) overemphasize the risk of falling from high places
(Menzies & Clark, 1995), social phobics have exaggerate beliefs in being rejected and appraised negatively in social situations (Rapee, 1995). These specific threat beliefs are suggested to subsequently influence the physiology of anxiety (see Britton et al., 2011). These threat beliefs also led patients to engage in what clinicians call ‘safety behaviours’. For example, panic disorder patients often carry medication and social phobics will avoid eye contact with people in order to avoid being negatively evaluated. However, these safety behaviours interfere with exposure-based therapies because patients frequently attribute the absence of catastrophic outcomes to the safety behaviours, rather than reducing their maladaptive threat beliefs accordingly.
Summary
The fear conditioning paradigm serves as an excellent laboratory tool to investigate the etiology of anxiety disorders. Fear learning has been largely examined in animals, but also increasingly in humans. Recent empirical studies have provided evidence for the importance of cognitive processes in human fear learning. While dual-system models propose that learning occurs in separate cognitive and associative systems that run independently from 19 each other, the propositional model argues that humans learn via a single, higher-order, cognitive system. The latter suggests a focus on the cognitive threat beliefs of anxious patients, and how these threat beliefs lead to maladaptive physiology, threat processing and behaviours. However, more evidence is required to see whether all aspects of fear learning are driven by cognitive processes.
Generalisation refers to the transfer of knowledge across situations. In the context of associative learning, generalisation refers to the transfer of CRs to novel stimuli that are similar to the CS+. Recent evidence suggests that excessive generalisation of conditioned fear is a pathogenic marker of anxiety disorders (Kaczkurkin et al., 2017; Lissek et al., 2010,
2014). As such, the following chapter will present a review of generalisation, including both animal and human studies. It will also include a theoretical review on how both dual-system- and propositional-models approach human generalisation.
20
Chapter 2:
Generalisation in animals and humans
21
Generalisation
Once we have learnt the association between two events and later encounter a novel, but similar event, we are likely to extract what we have learnt and transfer that knowledge to the novel event, without needing to learn anew. This process is referred to as generalisation, which plays an important role in real life situations. In an ethological approach, animals generalise their exploratory or avoidance behaviour to novel stimuli or situations; this allows rapid and adaptive responses to novelty and increases the chance of survival (Smith, 1993).
The ability to generalise knowledge is also found to be critical in children’s cognitive development. If children failed to generalise what they learnt in one situation to other novel situations, their development would be much slower and less efficient. One of the reasons why autistic individuals suffer from learning deficiency is thought to be their inability to generalise knowledge (Molesworth, Bowler & Hampton, 2005; Plaisted, O’Riordan & Baron-
Cohen, 1998). These examples suggest that the process of generalisation allows organisms to adaptively respond to the ever-changing environment.
However, if one excessively generalises from one stimulus to other novel stimuli, particular in the context of fear, generalisation becomes maladaptive. In other words, over- generalisation of fear involves predicting threat from threat neutral stimuli that only slightly resemble the threat cue. This will provide an inaccurate prediction of threat, and these repeated false alarms will lead to a constant state of anxiety. This may eventually interfere with one’s daily functioning. In fact, a series of laboratory-based, fear conditioning studies found over-generalisation of conditioned fear in patients suffering from panic disorder (Lissek et al., 2010), generalised anxiety disorder (Lissek et al., 2014) and posttraumatic stress disorder (Kaczkurkin et al., 2017). Empirical evidence like this suggests over-generalisation of fear is a pathogenic marker of anxiety disorders, and highlights the clinical importance of better understanding the mechanisms underlying generalisation. 22
Animal studies in generalisation
Numerous empirical studies in generalisation have been done in the last century, mostly in animals. The majority of these studies used an appetitive conditioning paradigm, where the US is an intrinsically rewarding reinforcement (e.g., food pellets), motivating animals to respond to the S+. There were relatively few studies that involved aversive conditioning. Nonetheless, it will be beneficial to review generalisation in animals in general, allowing a more comprehensive insight into the mechanisms of generalisation. One of the earliest empirical demonstrations of generalisation was observed in the classical Pavlovian conditioning paradigm. Pavlov (1927) first paired a specific tone CS with food US, and measured the saliva produced (CR) by the dogs when the tone CS was presented alone. The dogs not only salivated to the CS, but also to other tones that had never been directly paired with the US. Pavlov also reported that the strength of the CR (i.e., amount of saliva produced) progressively declined as the tone became more different to the CS. In other words, the CR was generalised to similar tones even though they had never been paired with the US.
Therefore, Pavlov suggested stimuli that were quantitatively different from the trained CS could also elicit the same CR but to a lesser extent. Following Pavlov’s (1927) study, numerous animal studies have been carried out to investigate stimulus generalisation (see
Mackintosh, 1974).
In a series of experiments using a single-cue, operant conditioning paradigm, Guttman and Kalish (1956) trained pigeons to peck at an illuminated light with a particular wavelength
(e.g., 550nm). In a subsequent test, they presented a range of novel lights differing in wavelength. The pigeons showed less pecking when different lights were presented, and the pecking responses were proportionally weaker to wavelengths further away from the trained value along the dimension. This response pattern formed a peaked gradient with the maximum response to the trained value and a gradual decrement in responses to stimuli more 23 dissimilar to it. The decrease in response strength or frequency to novel stimuli dissimilar to the reinforced cue is referred as generalisation decrement. This peaked generalisation gradient has been replicated throughout the literature (e.g., Brennan & Riccio, 1973; Hearst &
Koresko, 1968; Moore, 1972; Siegel, Hearst, George & O’Neal, 1968; Thomas & Switalski,
1966; see Fig. 1).
In light of early suggestive evidence, Lashley and Wade (1946) argued that true generalisation stems from the failure to distinguish novel stimuli from the reinforced cue, and that generalisation occurs because the animal has had too little experience with the stimuli involved to be able to discriminate among them. In fact, this notion was supported by some early studies that observed flat generalisation gradients (e.g., Blackwell & Schlosberg, 1943).
The Lashley-Wade theory argued that the apparent generalisation decrement observed in other studies was due to animals’ prior experience with the stimuli that were similar to those used in testing (e.g., pigeons that learnt to respond to a yellow keylight but may have pecked less to other wavelengths because they had already learnt to discriminate colour before they were trained in the experiment). If animals were prevented from having any experience with a certain kind of stimulus, such as colour, a flat generalisation gradient would have been observed when the colour dimension was used. However, numerous animal studies have since shown that animals are still able to discriminate between stimuli after being reared in environments deprived from those stimuli (Ganz & Riesen, 1962; Mecke, 1983; Peterson,
1962; but see Jenkins & Harrison, [1960] for supportive evidence), which conflicted with the
Lashley-Wade theory. On the other hand, Hull (1952) suggested that the flat generalisation gradient was accounted for by competition between the training stimulus and incidental stimuli. Incidental stimuli refer to any set of stimuli, other than the training and test set, which may come to control the subject’s behaviour as a consequence of acquisition training. In other words, other irrelevant stimuli (e.g., the background of the training cage, external noise) that 24
A B
CS+
C D
CS+ M ean response speed (s)
Figure 1. The generalisation gradients of visual and auditory conditioning in pigeons, rats and rabbits. (A) From Guttman & Kalish, 1956; (B) From Hearst & Koresko, 1968; (C) From Siegel et al., 1968; (D) From Brennan & Riccio, 1973. were presented in training may have gained excitatory value. Since these incidental stimuli were consistently presented for each test stimulus, this would increase animals’ responses and hence resulting in the apparent flat gradient. Hull (1952) suggested that discrimination training would eliminate any control exerted by the incidental stimuli as it entails the non- reinforcement or random reinforcement of the incidental stimuli. This would allow control 25 acquired by the stimulus of interested to be revealed. Hull’s (1952) suggestion led to an extensive body of research in generalisation using the differential conditioning paradigm.
Hanson (1959) trained pigeons to discriminate between a reinforced light of 550nm and a non-reinforced light of 560nm. As predicted, a sharper peaked gradient was observed, presumably because learning to incidental stimuli was suppressed due to discrimination training. However, the more interesting finding was that the responses no longer peaked at the trained value (550nm). Instead, the responses peaked at 540 nm, which had not been reinforced in the prior training phase (see Fig. 2). This phenomenon was coined ‘peak shift’, as the response peak shifted beyond the original trained value in the direction away from the non-reinforced stimulus (i.e., the light of 560nm). Hanson (1959) further varied the non- reinforced stimuli from 555 nm to 590 nm in different groups of pigeons, and found that the smaller the difference between the non-reinforced and the reinforced stimuli along the dimension, the further away from the trained value the response peaked at. This peak shift effect has been replicated throughout the animal literature (e.g., Ellis, 1970; Friedman &
Guttman, 1965; Honig, Thomas & Guttman, 1959; Terrace, 1964).
However, the peaked generalisation gradient was not always observed in animals.
Sometimes, an increasing linear gradient was observed with the response peak at the end of the stimulus dimension in the direction opposite of CS- (e.g., Huff, Sherman & Cohn, 1975;
Razran, 1949). A review from Ghirlanda and Enquist (2003) revealed that most studies that found a linear gradient used an intensity stimulus dimension. An intensity dimension refers to the increasing magnitude of the stimulus along the dimension - for example, an increase in decibels along a tone dimension (Huff et al., 1975), or an increase in brightness of a light dimension (Razran, 1949). Evidence like this suggests that the usage of an intensity stimulus dimension combined with a differential conditioning procedure encourages intensity generalisation, rather than the peaked gradient with the peaked shifted that was commonly observed in animal studies. 26
Associative accounts for generalisation
Hull-Spence account: With numerous empirical demonstrations of stimulus generalisation in animals, theoretical models were developed to account for the generalisation process. The associative models, as discussed in the previous chapter, were the dominant
Figure 2. Response gradients from pigeons in different groups. All groups were trained to respond to S+ of 550mn, and differed in S-. The control group were trained with S+ with the absence of S- (Hanson, 1959).
accounts for animal learning, and also for generalisation. The habit-family hierarchy model proposed by Hull (1934a, 1934b) suggested that an S-R link would be formed when responses to a stimulus have been learnt, and gradually become a ‘habit’. That is, the presence of S+ will activate the S-R link and trigger the response in an automatic sense. The habit strength can be spread to stimuli that are physically similar to S+, and the magnitude of responses is a function of physical similarity between S+ and the novel stimuli. 27
Building on a similar idea, Spence (1937) developed the transposition theory. This theory proposed that training with S+ will result in an excitatory response tendency to S+, and also other stimuli that are similar to it. Conversely, an inhibitory gradient will be formed when trained with an S-. The theoretical excitatory gradient has been supported by the aforementioned empirical studies that showed peaked gradients using a single-cue, operant paradigm (e.g., Guttman & Kalish, 1956). The theoretical inhibitory gradient has also been supported empirically. Honig, Boneau, Burstein and Pennypacker (1963) trained pigeons with a reinforced blank keylight (S+) and a non-reinforced blank keylight with a vertical line (S-).
The pigeons were then presented with novel stimuli with lines that tilted to varying degrees.
Pigeons showed the least pecking to S-, and pecking responses gradually increased when stimuli became more dissimilar to S-. Spence’s (1937) transposition theory also proposed that when S+ and S- were on the same stimulus dimension, the excitatory gradient with its peak at the S+ would interact with the inhibitory gradient with its peak at the S-, resulting in an excitatory-inhibitory summation. The inhibitory gradient of S- might suppress excitation to stimuli around the S+, especially stimuli on the side of the dimension close to the S- (see Fig.
3). In other words, responding to S+ decreases due to the inhibitory gradient of S-, and stimuli beyond S+ in the direction away from S- are less affected by this inhibitory gradient. The resulting net gradient is assumed to have its response peak beyond S+ in the direction further away from S-.
Spence’s account also predicted the magnitude of peak displacement as a function of the distance between S+ and S-. The closer S- is to S+, the more stimuli around S+ will be suppressed by the inhibitory gradient, hence resulting in a larger peak displacement.
Impressively, Spence’s transposition theory accurately predicted the later peak shift phenomenon (Hanson, 1959).
28
Figure 3. Adapted from Podlesnik and Miranda-Dukoski (2015). The summation of the excitatory and inhibitory gradients (solid and dashed lines, respectively) predicted by Spence’s (1937) transposition theory. The resulting gradient (dotted lines) has it peak shifted away from S+ opposite of S-.
Elemental accounts: Blough (1975) further explored the transposition theory on the basis of elemental analysis. His neural network model suggests that there are numerous input nodes (i.e., sensory neurons) that are connected to one output node (i.e., motor system for behavioural response). Importantly, Blough’s (1975) model assumes that each stimulus consists of a set of hypothetical elements. Each input node is responsible for encoding the corresponding individual element via the sensory process. If the stimulus is reinforced (CS+), all the input nodes corresponding to the elements of CS+ would activate the output node, triggering the behavioural response. In other words, each individual element of the CS+ acquires associative strength and the combined activity of the elements determines the degree of activation of the conditioned response. Stimuli close to the reinforced value along the stimulus dimension are assumed to share more common elements, hence more input nodes are activated and trigger a conditioned response with higher magnitude. This account was empirically supported by the gradual decrease in responding to stimuli further along the stimulus dimension from the trained stimulus (e.g., Friedman & Guttman, 1965; Honig et al., 29
1959; Terrace, 1964). The elemental representation model put forward by McLaren and
Mackintosh (2002) proposed a similar idea. This account suggested that every stimulus is learnt as a set of abstract, individual elements. McLaren and Mackintosh (2002) suggested that similarity of stimuli is based on the proportion of overlapping elements between stimuli.
In other words, response strength to a stimulus is based on which elements are activated and how many of these are shared with the S+. In support of the elemental representational model, studies (e.g., Bennett, Wills, Wells & Mackintosh, 1994; Mackintosh, Kaye & Bennett, 1991) have shown that when rats were conditioned to stimulus A, they showed little to no generalisation to B. However, when conditioned to a compound stimuli AX, the rats showed greater generalisation to BX because both compound stimuli shared the common element X.
These findings provide support to the notion that the strength of generalisation to novel stimuli depends on the common elements shared between the trained and novel stimuli.
Both Blough’s (1975) and McLaren and Mackintosh’s (2002) accounts are able to explain the peak shift effect. After discrimination training, elements in CS+ acquire excitatory strength while those in CS- acquire inhibitory strength. Since CS+ and CS- lie along the same perceptual dimension, it is assumed that they have some overlapping elements. That is, some shared elements in CS+ trials will gain inhibitory associative strength, and vice versa. Peak shift occurs because of the change in net associative strength in stimuli due to the interaction between the overlapping elements. In other words, associative strength peaks at a stimulus that is more different from CS- (i.e., less shared inhibitory elements), yet sharing the highest proportion of overlapping excitatory elements with CS+. Therefore, this explains why responding peaked at the stimulus beyond CS+ but in the direction away from the CS- (e.g.,
Friedman & Guttman, 1965; Hanson, 1959; but see Thomas [1993] & Thomas et al. [1991] for an alternative explanation). 30
Configural account: In contrast to the elemental accounts, Pearce (1987, 1994, 2002) proposed a configural theory, emphasizing the unitary nature of the representation of stimuli.
This theory suggests that animals possess a perceptual buffer, and in the case of CS+ training, the whole representation of CS+ will be stored in this buffer. Upon test trials, the novel stimuli will enter the buffer and be compared against the representation of CS+.
Generalisation occurs because of the physical similarity between the CS+ and test stimuli.
However the differences between these stimuli will change this buffer to a certain extent, which in turn decreases the associative strength of the test stimulus. The configural theory is also able to account for a variety of learning phenomena, such as blocking, overshadowing, summation and discriminative learning. However, the configural theory has been criticized for not being able to account for the peak shift effect (Livesey & McLaren, 2018).
All the aforementioned associative accounts proposed that response strength is a function of the physical similarity of novel stimuli to the CS+, hence always predicting a peaked generalisation gradient. However, as previously mentioned, an increasing linear gradient has sometimes been observed in animal studies (e.g., Huff et al., 1975; Razran,
1949). One common feature of these findings was the usage of intensity dimension. The associative accounts attribute such linear gradients to response bias to stimuli of higher intensity due to the energizing effect of intensity (Hull, 1949). That is, stimuli of high intensity are more likely to capture attention and enhance learning and performance.
However, such an explanation cannot account for response bias favour to stimuli of low intensity following from discrimination training of an intense CS- and a faint CS+ (e.g.,
Pierrel & Sherman, 1960; Zielinski & Jakubowska, 1977).
Although the associative accounts conceptualize generalisation in different ways, they predict the gradients of generalisation in a comparable way. That is, they predict a peaked gradient with maximum responses at the trained value and a gradual decrease in responding to 31 stimuli further away from the trained value following single-cue conditioning, and a shift in peak responding following discrimination training. The associative accounts also imply that the generalisation process is controlled by a low-level, automatic system of link formation.
The spread of associative strength and hence the response strength to novel stimuli is controlled by the physical similarity of stimuli to the reinforced cue. This leads to a potential gap when translating the findings in animals to humans, as there is no doubt that humans are able to utilize cognitive processes in associative learning, and hence affect generalisation in humans. Bearing this in mind, studies in human generalisation will be discussed in the following section.
Human studies in generalisation
One of the earliest empirical demonstration of generalisation in humans was the infamous ‘Little Albert’ experiment (Watson & Rayner, 1920). In this study, an infant named
Albert, was presented with a white rat (CS). Seconds later, a loud sound (US) was produced to startle Albert. After trials of the CS-US pairings, Albert not only developed fear responses to rats, but also expressed fear to other white furry objects like rabbits, fur coat, and even a
Santa-Claus mask with a white beard. The results were interpreted as the conditioned fear to rats being generalised to other similar objects.
However, Watson and Rayner’s (1920) study has been criticized for not controlling for other variables systematically (e.g., the lack of control group), and the novel stimuli used were qualitatively different from each other, preventing a quantitative analysis of generalisation. In light of this, subsequent studies employed a continuous stimulus dimension in a single-cue, fear conditioning paradigm (Hovland, 1937; Wickens, Schroder & Snide,
1954), similar to animal studies. Human participants were presented with a tone of a certain frequency paired with a shock, and subsequently tested with tones along the frequency dimension. The results partially aligned with the findings in animals, as participants showed 32 peaked skin conductance responses to the trained value, and a gradual decrease in skin conductance to stimuli further along the dimension. However, it remained unknown whether generalisation in humans would result in a symmetrical peaked gradients like animals, since these studies located the CS+ at the extreme end of the stimulus dimension, and hence only allowed the generalisation decrement to be measured to one side of the stimulus dimension.
Furthermore, other studies that used a similar paradigm failed to replicate the decrement in gradients. Instead, flat gradients were found (e.g., Burstein, Epstein & Smith, 1967;
Humphrey, 1939; Littman, 1949).
Concerned that the flat gradient observed was due to participants’ inability to discriminate between stimuli (see Lashley & Wade, 1946), researchers started to employ differential conditioning paradigms. These studies used stimuli such as tones (Baron, 1973;
Cross & Lane, 1962), brightness (Hebert, 1970), line orientation (Nicholson & Gray, 1971;
Thomas, Lusky & Morrison, 1992) and spatial location (LaBerge, 1961). Some of these studies found peak shift in humans after discrimination training (e.g., Baron, 1973; Cross &
Lane, 1962; Nicholson & Gray, 1971; Thomas et al., 1992). Evidence like this was taken to support the notion that humans learn in a similar way to animals, and that associative learning in humans is governed by a similar low level, associative system. However, the gradients found in these studies were often highly asymmetric, with a lack of decline in responding towards the end of stimulus dimension in the direction away from CS-. The lack of symmetry in these gradients is hard to reconcile with associative accounts, as these accounts predict a similar drop off in responses on both sides of CS+ along the dimension, even if the peak is shifted (but see Rilling [1977] for an alternative explanation – area shift). Some studies that failed to find peak shift in humans also showed an interesting pattern (e.g., LaBerge, 1961).
The gradients showed an increasing linear pattern from CS- towards the direction of CS+, and eventually peaked at the extreme end of the dimension. Although one may argue that linear 33 gradients can be explained by Hull’s (1949) energizing accounts, the stimuli used in these human studies were not along an intensity dimension (e.g., LaBerge, 1961).
Given the incongruent findings in generalisation between humans and animals, it appears that some other factors are influencing generalisation in humans, and the higher-order cognitive processes that are strongly developed in humans provide a likely candidate. Thomas and Mitchell (1962) were among the first to attribute the uniqueness in human generalisation to cognitive processes such as labelling. In the pre-conditioning stage, participants were shown an aqua colour (light of 550nm). In training, a number of different lights were presented and participants were instructed to only make a finger-lift response when the original light (CS+) was presented. In the subsequent test phase, the full range of lights of different wavelengths were presented and the responses to each stimulus were measured. As expected, participants who were instructed to label the CS+ as ‘green’ showed response biases to ‘greener’ light (an upward shift in wavelengths) while those not instructed to label the CS+ showed no response biases. Follow-up studies in the same laboratory (Doll & Thomas, 1967;
Thomas & DeCapito, 1966) further supported the ‘labelling’ hypothesis as response biases tended to be consistent with the label (more responding to greener stimuli when CS+ was labelled as green and more responding to bluer stimuli when CS+ was labelled as blue).
Relative to the labelling hypothsis, Shanks and Darby (1998) made explicit how relational rules substantially influence generalisation in humans. Although their study did not follow the traditional approach of using dimensional stimuli, it elegantly showed that human generalisation can be based on either similarity or other abstract rules. Using a causal judgement paradigm and a within-subject design, elemental stimuli were presented individually and followed by an outcome, while compound stimuli made up by such elements were not followed by an outcome (e.g., A+, B+, AB-) and vice versa for other stimuli (e.g., C-
, D-, CD+) in the training stage. Participants also received presentations of other stimuli 34 during training (e.g., E+, F+, GH-) to be used in the test phase. Associative accounts developed to explain animal conditioning would predict high responding to an EF compound during test because its elements were both reinforced in training, and low responding to elemental cues G and H. However, instead many participants appeared to infer a rule that the outcome of the compound stimulus is the reverse of the outcome of the individual elements that make up the compound, and the outcome of individual element is the reverse of the outcome of their compound. Participants who learned such a ‘reversal outcome’ rule during acquisition were able to apply and generalise this rule to novel stimuli in the test phase. By contrast, those who did not learn the reversal rule showed high levels of responding to compound stimuli made from reinforced elemental stimuli and vice versa, as predicted by the animal associative accounts. Therefore, Shanks and Darby’s (1998) study not only showed that inferred rules greatly influence generalisation in humans, but that when such verbal rules were not learnt, humans tended to generalise in a similar manner to animal studies.
Recently, Livesey and Mclaren (2009) employed a blue-green stimulus dimension, and manipulated stimuli to differ in hue, making it difficult to verbalize the difference between stimuli relative to other stimulus dimension such as luminance or orientation. Using this dimension in a causal judgement task, they found a peak shift effect early in the test phase.
However, the generalisation gradient gradually became monotonic with the peak responding at the end of the stimulus dimension over the course of testing. Livesey and McLaren (2009) explained the occurrence of the initial peak shift as a result of participants not being able to verbalize the difference between stimuli. Therefore, they were learning and generalising via a low-level, associative system, that led to the peak shift pattern as predicted by the associative accounts. Across the testing trials, participants gained more experience with the set of stimuli.
As a result, participants were more likely to verbalise the difference between stimuli and hence more likely to infer a relational linear rule, resulting in the increasingly linear gradient. 35
To interpret empirical evidence like this, most researchers have attributed the differences in animal and human generalisation to a higher-order, cognitive process uniquely possessed by humans. The results align with the dual-system learning models, showing that when sufficient task-relevant information is gathered, generalisation in humans will be primarily driven by the cognitive system. But when participants are not able to learn on a cognitive level, generalisation will instead be driven by the low level, associative system.
However, more recent studies have cast doubt on the necessity of proposing two independent systems to account for generalisation of associative learning in humans. In a fear conditioning paradigm, Ahmed and Lovibond (2018) and Lee, Hayes and Lovibond (2018, experiment 2) trained participants to discriminate between a CS+ and CS-, and then presented a wide range of stimuli across the stimulus dimension in test. Results from both studies showed a peaked gradient, with tendency for responding to peak at the stimuli adjacent to
CS+ away from CS- (i.e., peak shift) as predicted by the associative acocunts. Participants reported inferring different rules in the post-experimental questionnaires, and the overall gradient was then decomposed accordingly to the rules participants reported using. Some participants reported adopting a similarity rule, whereby they expected that stimuli perceptually similar to CS+ would be more likely to predict shock, while stimuli perceptually dissimilar to CS+ would be less likely to predict shock. Other participants reported inferring a linear rule that if the stimulus became more unlike CS-, the more likely it would predict shock. After grouping the data from participants according to their reported rules, the gradients were strikingly consistent with the rules. A symmetrical peaked gradient was observed in those who reported a similarity rule, while a linear gradient with the peak responding at the extreme end of the stimulus dimension away from CS- was found in those who reported a linear rule. Furthermore, the apparent peak shift observed in the overall gradient seemed to be a combination of similarity and linear rules, that is, peak shift may be a result of combining the data from the rule subgroups. This potentially explains the 36 asymmetric peaked gradients often observed in human studies – the lack of decline in responses to stimuli beyond CS+ could be attributed to the linear gradients, as the response biases in these gradients were consistently in the direction opposite to CS- along the dimension. This suggests that the overall gradient may not be informative, or may even be misleading.
In a single-cue fear conditioning study, Wong and Lovibond (2017) found a similar alignment between reported rules and generalisation gradients. Besides the similarity and linear rules, some participants who did not identify any clear rules were classified into a No rule subgroup. These participants showed a peaked but relatively flat gradient. Importantly, the skin conductance gradients were consistent with the reported rules and the corresponding shock expectancy data. The findings of these studies are more consistent with the propositional model. First, the apparent peak shift observed that is normally explained in terms of associative links (within a dual-system model), is open to an alternative interpretation in terms of a combination of different rules, under the propositional model.
Secondly, some strict dual-system accounts (e.g., Clark & Squire, 1998) would predict a dissociation between cognitively controlled expectancy ratings and the associatively controlled skin conductance whenever participants infer a non-similarity rule. For instance, the dual-system accounts predict a sharp peaked gradient in skin conductance even for those participants who came up with no rules, as gradients learnt via the associative system should be influenced by the associative strength between stimuli but not by cognitive factors.
However, a strong correspondence was observed between the expectancy and skin conductance gradients, with both measures showing a flat, peaked gradient. This is more consistent with the propositional model, in which CRs are a product of propositional beliefs.
The propositional model also provides alternative explanations for other findings that appear to favor the interpretation of dual-system models. For instance, in Shanks and Darby’s 37
(1998) study, the similarity-based generalisation pattern could be a result of participants using an explicit similarity rule, as it arguably exerts less cognitive load on participants compared to the learning and usage of a reversal rule. In Livesey and McLaren’s (2009) study, the peak shift effect observed in the early test phase could be an artifact of participants not being willing to extrapolate a linear rule. The increasingly linear gradient observed in late test phase could be the result of increased willingness of the extrapolation of a linear rule.
Besides labelling and relational rules, verbal instructions also greatly influence generalisation in humans. Vervliet et al. (2010) and Ahmed and Lovibond (2015) paired a coloured geometric figure (e.g., blue triangle) with an electric shock in the training phase.
Participants were informed that a particular feature of the CS+ predicted the shock (e.g., colour). This instruction was given either at the beginning of the study (Vervliet et al., 2010) or after training (Ahmed & Lovibond, 2015). An increase in both shock expectancy ratings and skin conductance were observed to stimuli that contained the verbally instructed predictive feature (e.g., blue circle), but not to stimuli that shared the physical feature with the
CS+ that was not verbally informed as a predictive feature (e.g., yellow triangle). This suggests that fear can be selectively generalised to novel stimuli by instructions. Both dual- system and propositional models are able to account for by the above findings, since both learning models attribute this bias in generalisation to participants’ belief that only stimuli containing the predictive feature led to shock. However, the dual-system models do not make a clear prediction as to when generalisation would be guided associatively or cognitively. In contrast, the propositional model provides a more parsimonious explanation.
Summary
Generalisation has been studied extensively in both animals and humans in the past century. Most findings in animals aligned with the predictions of associative accounts, revealing a sharp peaked gradient, with magnitude of responses proportional to perceptual 38 similarity between test stimuli and the trained value. The peak shift effect, often observed in animals after discrimination training, can also be accounted for by the associative theories.
However, generalisation in humans is less consistent compared to animal studies. Different shapes of gradients are observed in human studies that cannot be predicted by the associative accounts. It is clear that the variability observed in human generalisation can be attributed to the operation of higher-order cognitive processes, and empirical studies have pointed to the importance of cognition in human generalisation (e.g., Shanks & Darby, 1998; Thomas &
Mitchell, 1962). Recent evidence has also questioned the necessity of an associative system in humans (e.g., Ahmed & Lovibond, 2018; Wong & Lovibond, 2017), and the findings have been argued to favour a more parsimonious single system model (Mitchell, De Houwer &
Lovibond, 2009). Nonetheless, the question of whether single or dual-system models provide a better account of human learning and generalisation remains hotly debated (e.g., McLaren et al., 2014).
So far, we have covered the theoretical and empirical aspects of fear learning and generalisation, both of which are regarded as important aspects for the study of psychopathology of anxiety disorders (e.g., Lissek et al., 2008b; Mineka & Zinbarg, 2006). In the next chapter, trait anxiety, a factor that has been found to heavily influence fear learning, will be discussed.
39
Chapter 3:
Trait anxiety and anxiety disorders
40
As mentioned in Chapter 1, aversive experiences have been suggested as a mechanism for the etiology of anxiety disorders, and the fear conditioning procedure provides a great laboratory tool to investigate this issue (Mineka & Zinbarg, 2006). However, not everyone who has been exposed to traumatic event ends up developing an anxiety disorder (Mineka &
Sutton, 2006; Rachman, 1990). For instance, the prevalence rate of chronic PTSD of concentration camp survivors and prisoners of wars is approximately 50% (Goldstein, van
Kammen, Shelly, Miller & van Kammon, 1987; Yehuda et al., 1995), and only 4% of the victims who experienced natural disasters subsequently developed PTSD (Shore, Vollmer &
Tatum, 1989). Associative learning theories provide some potential explanations for these apparent individual differences in fear acquisition. First, it can be explained by a learning phenomenon known as latent inhibition (Lubow, 1973). Latent inhibition refers to the pre- exposure of a CS in the absence of the US that leads to a reduction in CR magnitude in the subsequent conditioning phase. Specifically, if an individual has prior safe experience with a
CS before an aversive conditioning experience, the amount of conditioned fear acquired will be reduced, decreasing the chance of developing a full blown anxiety disorder. This idea was supported by empirical evidence that individuals who had been exposed to more benign dental treatments were less likely to develop dental phobia after a traumatic dental experience, compared to those who had fewer prior dental experiences (Davey, 1989; De Jongh, Muris, ter Horst & Duyx, 1995). Secondly, the transition from conditioned fear into clinically diagnosed fear may be attributed to a phenomenon called US revaluation. Some animal studies paired the CS with a relatively mild US, and then delivered a stronger US without the presence of CS. When the rats were presented with the CS again, they showed a greater degree of conditioned fear to it, and this CR was in line with the unconditioned response (UR) to the stronger US (Rescorla, 1974; Sherman, 1978). Similar examples of this ‘inflation effect’ have also been observed in humans (Davey, 1997). This process is suggested to involve a reevaluation of the aversiveness of the CS after experiencing a stronger US 41 subsequent to the original conditioning experience. For instance, one may not develop PTSD immediately after experiencing a mild traumatic experience (CS-US pairing). However, being physical assaulted (a stronger US) sometime after the accident may lead to the reevaluation of the aversiveness of the stimuli associated with previous traumatic experience. That is, the revaluation of the CS may lead to more intense fear to objects that resemble the original traumatic experience, and lead to a full blown PTSD.
However, there are also numerous factors other than learning per se that may contribute to the individual differences in the development of anxiety disorders, such as biological and environmental factors. Biological factors include genetic factors, where monozygotic twin studies suggest that anxiety disorders are partly hereditary (e.g., Skre,
Onstad, Torgersen, Lygren & Kringlen, 1993; Torgersen, 1983), and gender differences, where evidence indicates that females are more likely to develop anxiety disorders (e.g.,
Donner & Lowry, 2013; McLean, Asnaani, Litz & Hofmann, 2011). Environmental factors such as lower levels of socioeconomic status has also been found to be associated with a higher prevalence rate of anxiety disorders (e.g., McLaughlin, Costello, LeBlanc, Sampson &
Kessler, 2012; Mwinyi et al., 2017). Among these other factors, a personality trait known as anxiety proneness has established as a vulnerability factor for the development of anxiety disorders.
Anxiety proneness
Anxiety proneness refers to a stable predisposition to show negative emotional responses across situations. H. J. Eysenck (1947) was among the earliest researchers to conceptualize anxiety proneness as a personality trait, and proposed a neurobiological model of personality traits.
Neurobiological models of anxiety: Eysenck (1947, 1967) identified two major personality dimensions, extraversion-introversion (E-I) and neuroticism-stability (N-S) in his 42 biological model of personality. The E-I dimension measures the degree to which an individual enjoys social interaction and the N-S dimension measures emotional stability – the tendency to experience negative feelings like anxiety, worry, fear and frustration (Eysenck,
1947). According to Eysenck (1967), the two personality dimensions are determined by two distinctive neural mechanisms. The degree of introversion is regulated by the ascending reticular activating system (ARAS) located in the brain stem, which controls the level of cortical arousal (Magoun, 1963; Moruzzi & Magoun, 1949). Counterintuitively, introverts are thought to have a constant, higher level of ARAS arousal (or lower ARAS threshold), hence they can easily become overstimulated. To avoid excessive stimulation, introverts usually choose to actively inhibit their behaviours. In contrast, extraverts have a lower resting level of
ARAS activity, and are therefore always seeking excitement in order to increase the arousal to an ideal level.
On the other hand, the degree of neuroticism is said to reflect the level of reactivity of the autonomic nervous system (Eysenck, 1957, 1967; Eysenck & Eysenck, 1975). That is, individuals high in neuroticism are thought to have a tendency to overreact to negative stimuli, reflected in automatic activation. The notion that introverts are highly aroused and neurotic individuals are overreacting to negative stimulation led to the idea that individuals high in both introversion and neuroticism will have excessive sensitivity to stressful situations. That is, neurotic introverts are suggested to react with excessive anxiety when faced with negative stimuli, and anxiety proneness is suggested to be the result of hyperactivity of both the ARAS and autonomic nervous system. There is mixed evidence in the literature to support Eysenck’s biological model of anxiety proneness. For instance, in the classic lemon juice experiment, Corcoran (1964) placed several drops of lemon juice on participants’ tongues, and found that introverts salivated almost as twice as extraverts did, supporting the notion that introverts have a higher level of ARAS arousal. By the same token,
Lader and Wing (1964) found that dysthymics (i.e., individuals high in both introversion and 43 neuroticism, as defined by Eysenck, 1959) showed more skin conductance responding to an auditory stimulus. However, such findings are inconsistent in the literature (e.g., Bronzaft,
Hayes, Welch & Koltuv, 1960; McReynolds, Acker & Brackbill, 1966).
It was observed in animal studies that removal of the anterior cingulate cortex (ACC) resulted in a loss of anxiety and fear responses (e.g., Glees, Cole, Whitty & Cairns, 1950).
Subsequent clinical observations in anxious patients revealed that lesions of ACC greatly reduced anxiety symptoms (Whitty, 1955), suggesting a link between AAC and anxiety.
Powell (1979) reviewed lesion studies (e.g., Kelly, Richardson & Mitchell-Heggs, 1973;
Mitchell-Heggs, Kelly & Richardson, 1976) in humans and found that the removal of ACC reduced neuroticism more than introversion. Based on the above evidence, Gray (1981; 1982;
Gray & McNaughton, 2000) suggested a simpler model with only one dimension, namely anxiety, which determines anxiety proneness. Compared to Eysenck’s model, the anxiety dimension lies at a 30-degree angle to neuroticism, suggesting that it has a higher correlation with neuroticism than with introversion (see Fig. 4). The reinforcement sensitivity theory proposed by Gray (1982) suggests that there is a behavioural inhibition system (BIS) located in the septohippocampal system. The BIS is suggested to be activated to an optimal level when aversive stimuli that involve punishment and uncertainty are presented, which in turn activates the fight/flight system and allows the organism to respond to the aversive stimuli adaptively. However, individuals who score high in the anxiety dimension are suggested to have hyperactivity in the BIS, which leads to maladaptive, excessive avoidance. The hyperactivity in BIS also results in a heightened sensitivity to punishment and uncertainty, which in turn leads to an increased level of anxiety. Therefore, increased sensitivity in BIS was proposed to be the neurobiological basis for anxiety proneness.
Physiological studies of anxiety: As proposed by Eysenck (1957, 1967) and Gray
(1981, 1982), individuals who are prone to anxiety more easily experience physiological 44
Figure 4. H. J. Eysenck’s (1957, 1967) model of personality. The x-axis is the Extraversion- Introversion dimension, while the y-axis is the Neuroticism-Stability dimension. Eysenck proposed that anxiety-prone individuals have a combination of a low level of extraversion and a high level of neuroticism. The superimposed diagonal line is the anxiety dimension proposed by Gray (1981, 1982). It is a more parsimonious dimension for measuring anxiety proneness, while highly correlated with neuroticism, it is just slightly correlated with introversion.
arousal when faced with a negative and stressful environment, hence resulting in heightened anxiety. This suggests that an increase in physiological activity increases the anxiety one would experience. In light of these proposals, studies have been conducted to examine if individuals who differ in anxiety proneness would show different level of physiological responses. First, some studies found that anxiety-prone individuals showed an increased level of skin conductance (e.g., Nielsen & Petersen, 1976) and heart rate (e.g., Santibáñez-H &
Schroeder, 1988) to aversive stimuli. However, the majority of the literature did not find any differences in physiological responses between high and low anxious individuals (e.g.,
McReynolds et al., 1966; Neary & Zuckerman, 1976; O'Gorman, 1977). In a series of well- controlled studies, Fahrenberg and colleagues (1982, 1983) compared multiple physiological 45 measures (e.g., skin conductance, heart rate, EEG, EMG, respiratory irregularity) between anxiety-prone and control individuals. All physiological measures consistently failed to show any anxiety differences.
Secondly, if increased anxiety is due to over-sensitivity of the neurobiological system, it would suggest less habituation of physiological responses in anxiety-prone individuals.
Habituation refers to the diminishing physiological responses to an external stimulus as participants adjust to the stimulus across trials (Sokolov, 1960). Some studies found that anxiety-prone individuals show little to no habituation to the aversive stimulus (e.g., Hart,
1974), while other studies failed to find significant differences (Arena & Hobbs, 1995;
Martin-Soelch, Stocklin, Dammann, Opwis & Seifritz, 2006).
Thirdly, it had been suggested that anxiety-prone individuals have a higher baseline level of physiological responses. Although studies measuring resting heart rate provided some evidence for this notion (e.g., Hodges & Speielberger, 1966; Kelley, Brown & Shaffer, 1970), such a difference was not found in skin conductance (Knyazev, Slobodskaya & Wilson, 2002;
Martin-Soelch et al., 2006).
It can be seen that the empirical evidence is difficult to reconcile with Eysenck’s
(1957, 1967) and Gray’s (1981, 1982) neurobiological models for anxiety proneness.
However, Weinberger, Schwartz and Davidson (1979) suggested that the null differences in physiological responses between anxiety prone and non-anxiety prone individuals were due to the failure to identify repressors. Repressors are individuals who score low on anxiety proneness, but high on measures of defensiveness of self-image and self-ego. Weinberger et al. (1979) suggested that repressors actively suppress the negative thoughts induced by aversive stimuli in order to protect their self-images, but this active coping style inevitably leads to an increase in physiological activity. This notion is supported by some empirical evidence, showing an increase in autonomic reactivity when participants actively repressed 46 their negative thoughts (e.g., Dunn, Billotti, Murphy & Dalgleish, 2009; Schwartz, 1995).
However, it is problematic to identify a true repressor. If repressors have a strong desire to protect their self-image and self-ego, they may not respond honestly in the social desirability questionnaire, in order to defend their self-ego. In other words, in order to defend their image and ego and also increase their social desirability, true repressors may deliberately respond to the questionnaires as if they were low anxious individuals (low in self-image and self-ego defensiveness), therefore making the identification of true repressors extremely difficult.
Trait and state anxiety: Cattell and Scheier (1958, 1961) were among the earliest investigators to further elaborate anxiety other than as a stable personality trait, like neuroticism (Eysenck, 1947, 1967). In their factor analytic studies, Cattell and Scheier (1958,
1961) suggested that anxiety can be categorized into state anxiety and trait anxiety, a distinction which Spielberger (1966, 1975a, 1975b) furthered elaborated. State anxiety refers to the transitory emotional state of one’s subjective feeling of apprehension, anxious expectation and activation of autonomic activity. On the other hand, trait anxiety refers to a stable individual trait reflecting anxiety proneness. Much like neuroticism, individuals high in trait anxiety are more likely to respond negatively to aversive stimuli or situations. Given the importance of the distinction between trait and state anxiety, Spielberger and colleagues
(1970, 1975a, 1983) put forward the State-Trait Anxiety Inventory (STAI) to separate assessment of trait anxiety from state anxiety. Trait anxiety measured by STAI closely resembles neuroticism, correlating approximately 0.7 with it (Watson & Clark, 1984), but is only slightly correlated with extraversion (r=-0.3). This suggests that STAI provides good construct validity for measuring anxiety proneness, and also aligns with Gray’s (1981, 1982) contention that trait anxiety is closer to neuroticism than to introversion. Spielberger (1975b) also proposed the state-trait process model, which posited the interaction between trait and state anxiety. When a threatening stimulus is presented, state anxiety will be initiated, and the level of state anxiety is determined by one’s trait anxiety. Individuals high in trait anxiety are 47 suggested to appraise the situation as more dangerous, which in turn activates a higher level of state anxiety with a longer duration of intense apprehension. Individuals high in trait anxiety may also experience a chronic state of heightened anxiety, which is suggested to be the basis for their proneness of developing anxiety disorders after traumatic experiences
(Spielberger, 1985).
Anxiety as negative affectivity: Watson and colleagues (1984, 1985) proposed two orthogonal mood dispositional dimensions that are relevant to anxiety proneness, namely negative and positive affectivity. Negative affectivity (NA) refers to the predisposition to experience negative emotions such as anxiety, fear and sadness. On the other hand, positive affectivity (PA) refers to how likely one is going to express positive emotion such as enthusiasm, interest, joy and determination. Watson and Clark (1984) and Watson and
Tellegen (1985) emphasized that these two dimensions are not correlated with each other. In other words, individuals scoring high in NA do not necessarily score low in PA, and vice versa. An extensive review by Watson and Clark (1984) concluded that state anxiety was highly correlated with NA, under both stress and baseline conditions, and the majority of the reviewed studies found that state anxiety had no significant correlation with PA. Follow-up evidence found that depression, similar to anxiety, was positively correlated with NA
(Tellegen, 1985; Watson & Tellegen, 1985), but also had a negative correlation with PA. That is, among individuals high in the NA dimension, PA scores can be used to differentiate depression from anxiety. Given the comorbidity between anxiety and depression, Watson and colleagues (1984, 1985, 1988) argued that one advantage of measuring anxiety proneness via affectivity is its ability to discriminate between anxiety and depression.
Clark and Watson (1991) later developed the tripartite model, to distinguish between clinical anxiety and depression. Similar to their previous review (Watson & Clark, 1984), they suggested the NA dimension to be nonspecific to anxiety and depressive disorders, since both 48 constructs are highly correlated with NA. Anhedonia is an important component that defines depression since trait PA is significantly negatively correlated with depression, while anxiety has no significant correlation with PA. The tripartite model also suggests the differentiation of anxiety from depression by examining one’s physiological arousal. Despite the literature described earlier which showed an inconsistent pattern in an increased physiological activity
(e.g., skin conductance, heart rate) among trait anxious individuals, Clark and Watson (1991) argued that both trait anxious individuals and anxious patients did reliably show symptoms of physiological hyperarousal when state anxiety is induced, with symptoms such as trembling, shortness of breath and sweaty palms (e.g., Gencoz, Gencoz & Joiner, 2000; Joiner et al.,
1999). The tripartite model has received empirical support on how the three factors (NA, PA and physiological arousal) serve to distinguish between anxiety and depression (see Joiner,
Catanzaro & Laurent, 1996; Watson et al., 1995a, 1995b).
The aforementioned constructs (neuroticism, trait anxiety and NA) are moderately correlated and can all be used to some extent to assess anxiety proneness. However, according to the tripartite model (Clark & Watson, 1991), neuroticism is strongly associated with NA, which is common to anxiety and depression (see also Clark, Watson & Mineka, 1994; Bishop
& Forster, 2013). Neither neuroticism or NA assess physiological symptoms specific to anxiety. By contrast, trait anxiety has been found to broadly correlate with NA but not with
PA, while being able to assess physiological symptoms specific to anxiety (Watson & Clark,
1984). Therefore, trait anxiety is arguably the most suitable construct to measure anxiety proneness specifically.
The measurement of trait anxiety
Numerous psychometric questionnaires have been designed to measure trait anxiety, for example, the neuroticism scale from the Eysenck Personality Questionnaire (Eysenck &
Eysenck, 1975), the STAI (Spielberger et al., 1970, 1975a, 1983), the Beck Anxiety Inventory 49
(BAI; Beck, Eüsteom, Brown & Steer, 1988) and the Hospital Anxiety and Depression Scale
(HADS; Zigmond & Snaith, 1983). Among these questionnaires, STAI is the most widely used measurement for trait anxiety. However it is strongly correlated with depression. For instance, the STAI has been found to be moderately to strongly correlated with the Beck
Depression Inventory (BDI; Beck, Ward, Mendelson, Mock & Erbaugh, 1961) and the revised BDI-II (Beck et al., 1996), ranging from r=0.69 to r=0.81 (e.g., Harris & D’Eon,
2008; Pallesen, Nordhus, Carlstedt, Thayer & Johsen, 2006; Storch, Roberti & Roth, 2004).
Therefore, the STAI has been criticized for measuring both aspects of anxiety and depression despite the fact it was designed to measure anxiety purely (e.g., Caci, Baylem, Dossios,
Robert & Boyer, 2003). It has also been suggested that STAI measures general negative affect
(Andrade, Gorenstein, Vieira Filho, Tung & Artes, 2001) or general psychopathology
(Kennedy, Schwab, Morris & Beldia, 2001) instead of trait anxiety.
In light of overcoming the criticisms of the STAI, Lovibond and Lovibond (1995a) developed the Depression Anxiety Stress Scale (DASS). The DASS consists of three self- reported scales designed to measure depression, anxiety and stress/tension. While the
Depression scale assesses anhedonia, hopelessness and devaluation of life, the Stress/tension scale assesses persistent non-specific arousal such as nervous arousal, irritability and difficulty in relaxing. Similar to the tripartite model (Clark & Watson, 1991), the Anxiety scale assesses the symptoms of autonomic arousal specific to anxiety, such as the awareness of dryness of the mouth, pounding of the heart and breathing difficulties. The Anxiety scale also assesses situational anxiety and subjective experience of anxious effect. The Anxiety
Scale in DASS has been found to only moderately correlated with depression (BDI, r=0.58,
DASS-Depression, r=0.54; Lovibond & Lovibond, 1995b). The DASS Anxiety scale was also found to be positively correlated with NA, ranging from r=0.57 to r=0.63, but not to PA (r=-
0.18; Brown, Chorpita, Korotitsch & Barlow, 1997; Gloster et al., 2008), consistent with the idea that PA can be used to differentiate anxiety from depression (Tellegen, 1985; Watson & 50
Tellegen, 1985). In fact, the Anxiety and Depresssion scales are capable of differentiating between anxiety and depressive patients (Antony, Bieling, Cox, Enns & Swinson, 1998;
Brown et al., 1997).
The DASS is normally a state measure, as it asks the respondent to rate the degree to which each symptom was experienced over the past week. However, the DASS can be modified to a trait measure, by asking the respondent how likely each symptom is experienced in a typical week in the past year, or in general. Lovibond (1998) administered the DASS to a large sample of university students (Time 1), and then re-administered the DASS to the same sample after 3 to 8 years (Time 2). The scales remained stable overtime, and most importantly, the scales in DASS showed selective stability. That is, each scale obtained at
Time 1 was the best predictor of the same scale at Time 2. Furthermore, at Time 1 some students received a trait version of DASS, with the questions concerning how likely the symptoms were experienced during ‘a typical week in the past 12 months’, while some students received the state version of DASS, assessing symptoms experienced ‘during the past week’. At Time 2, all participants received the trait version. Lovibond (1998) found no evidence that the trait and state versions affect how the predictability of the scales from Time
1 to Time 2. This suggested that the state version of DASS is able to provide a reliable measure of trait anxiety.
Cognitive approach to anxiety
Threat appraisal: The inconsistency in the physiological findings between trait anxiety, combined with the notion that anxiety can be experienced as a transient state of emotion (e.g., Spielberger, 1970; Watson & Clark, 1984), has further suggested that trait anxiety can modulate anxiety on the cognitive level. It has been hinted in Spielberger’s
(1975b) state-trait process model that trait anxiety modulates one’s appraisal of threat, and in turn determines the magnitude of state anxiety. This process of threat appraisal was suggested 51 to occur on a cognitive level (Lazarus, 1966, 1981, 1991). According to Lazarus, cognitive appraisal of threat can be categorized into three major stages. First, there is primary appraisal, in which individuals evaluate the perceived level of threat. Second, there is secondary appraisal, where individuals evaluate the cognitive and other resources available to cope with the situation, in order to avert the threat. Third, there is reappraisal that helps modifying the primary or secondary appraisals if the level of perceived threat is not successfully reduced
(see Smith and Lazarus [1993] for more details). Lazarus’ (1991) theory emphasized the role of the cognitive process of appraisal (see Fig. 5). According to this theory, when an aversive stimulus is presented, individuals will go through primary appraisal, and make an initial evaluation of the perceived level of threat of the situation. This appraisal process will then directly determine the emotional experience (e.g., anxiety level), the level of physiological activity (e.g., skin conductance) and the action tendencies (e.g., avoidance behaviour; cf. Lang et al., 1998). The process of the primary appraisal will determine how individuals cope with the situation via secondary appraisal, which may feed back to reduce anxiety. Similar to the state-trait process model (Spielberger, 1975b), the theory of appraisal suggested that trait anxiety can greatly influence how one would cognitively appraise the threat level, and hence the level of anxiety one would experience. The notion that greater threat appraisal leads to higher expectancy of harm is highly consistent with the cognitive accounts in fear conditioning mentioned in Chapter 1 (Mitchell et al., 2009; Reiss, 1980). These accounts proposed that the level of anxiety (i.e., the CRs in fear conditioning studies) is the product of the propositional belief of threat expectancy.
Influenced by Beck’s schema theory for anxiety disorders (Beck & Clark, 1988; Beck
& Emery, 1985), Parkinson (1994, 1995) and M. W. Eysenck (1997) modified the theory of appraisal and proposed the four-factor theory of anxiety (see Fig. 5). This theory posited two major changes from Lazarus’ original theory of appraisal. First, it proposed that cognitive schemas, which are representations of prior knowledge and experiences, receive feedback 52
Figure 5. The white text boxes represent a simplified version of the theory of appraisal (Lazarus, 1991). The addition of the green text boxes represent a simplified four-factor theory of anxiety (Parkinson, 1994, 1995; M. W. Eysenck, 1997).
from emotional experiences. In other words, the proposed cognitive schema will be tuned to become more attentive to threatening information due to the feedback mechanism of state anxiety. This schema is then suggested to induce cognitive bias that increases the level of threat appraisal, which in turn increases the state anxiety experienced, hence forming a vicious cycle. Secondly, similar to Lazarus’ theory of appraisal, the four-factor theory also proposed that the appraisal process will directly determine the level of physiological activity.
However, it also emphasizes on how one’s cognitive schema would bias the perception of their own physiological activity. That is, having a cognitive schema that is tuned to threat, trait anxious individuals will have a higher level of perceived physiological activity regardless 53 of the actual physiological activity. This heightened level of perceived physiological activity is said to increase the magnitude of anxiety one will experience. In an attempt to provide supportive evidence for the four-factor theory, Derakshan and Eysenck (1997) ran a series of studies and found a significant relationship between self-reported anxiety and the perceived appraisal outcomes (e.g., physiological activity, action tendencies), but also a significant correlation between self-reported anxiety and the actual physiological activity. That is, supportive evidence was found for both the theory of appraisal (Lazarus, 1991) and the four- factor theory of anxiety (Eysenck, 1997; Parkinson, 1994, 1995). The finding that trait anxious individuals showed a significantly higher level of perceived physiological activity and behavioural tendency suggested that trait anxious individuals show a greater degree of threat appraisal. Numerous empirical studies have also found different cognitive biases toward threat among trait anxious individuals, which will be discussed below.
Attentional bias: In Beck’s (1976) schema theory, it was suggested that anxious individuals have a cognitive schema that is biased towards threat. This suggested that information processing will be heavily oriented to threat-related information, including early processes such as attention allocation. In a dichotic listening task (Mathews & MacLeod,
1986), stories were delivered through the attended channel via a headphone, while participants were required to read back aloud the story, In the unattended channel, either affective neutral or threatening words (e.g., assault, tragedy) were delivered. Simultaneous with the listening task, participants were instructed to perform a reaction time task, which was synchronized with the words presentation in the unattended channel. Trait anxious individuals were found to react slower when a threatening word was delivered via the unattended channel relative to the low anxious control group, and showed similar response latency in the task when a neutral word was delivered via the unattended channel. Similar results were found in an analogue visual task. In a modified Stroop task (Stroop, 1935), namely the emotional
Stroop task, words written in different colours were visually presented to participants. Trait 54 anxious individuals were found to name the colour of threatening words more slowly than individuals with low trait anxiety (e.g., Dawkins & Furnham, 1989; Mathews & MacLeod,
1985; Richards, French, Johnson, Naparstek & Williams, 1992). The above results suggested that trait anxious individuals allocate their attention to the threatening words which then interferes with their task performance, supporting the notion that trait anxious individuals have their attentional resources biased towards threat.
However, this interpretation has been criticized as the impaired reaction time to threatening words could also be interpreted as an effortful avoidance of threat cues rather than direct attentional bias to these cues (De Ruiter & Brosschot, 1994; MacLeod, Mathews &
Tata, 1986). In an attempt to overcome this problem, MacLeod et al. (1986) created the dot- probe paradigm. In this task, two cues, one affectively threatening and the other one affectively neutral, were presented on the screen concurrently. After the offset of these cues, a dot probe would replace either one of the cues, and participants had to respond to the dot probe as fast as possible. Response latency to the dot probe was suggested to reveal the amount of attention allocated to the cues. In other words, faster responses to the dot probe that replaced the threatening cue suggested more attention captured by the threat cue. This pattern was consistently observed in trait anxious individuals, favoring the idea that trait anxious individuals had their attention biased towards threat, rather than actively avoiding it (e.g.,
Broadbent & Broadbent, 1988; MacLeod & Mathews, 1988; MacLeod et al., 1986).
Attentional bias to threat in trait anxious individuals has been consistently found using similar visual tasks, like emotional spatial cueing and visual search paradigms (see Bar-Haim, Lamy,
Pergamin, Bakermans-Kranenburg & van IJzendoorn [2007] for a review). Furthermore, trait anxious individuals showed higher response latency to the target probe that was located differently from a previous threatening cue (e.g., Fox, Russo, Bowles & Dutton, 2001; Koster,
Crombez, Verschuere, van Damme & Wiersema, 2006; Mogg, Holmes, Garner & Bradley,
2008). This suggested that trait anxious individuals have greater difficulty in shifting their 55 attention away from threat cues. In other words, anxious individuals find it hard to disengage their attention from threat.
The attentional bias towards threat combined with the difficulty in disengaging from threat was suggested to lead to attentional narrowing, preventing individuals from attending to other safety cues in the surroundings (Eysenck, Derakshan, Santos & Calvo, 2007). This attention narrowing to threat will undoubtedly lead to an increase in threat appraisal, and therefore induces a heightened state of anxiety.
Interpretation bias: A large body of literature has shown that trait anxious individuals tend to interpret ambiguous cues in a negative way. Haney (1973) first showed participants a sentence, and then asked them to choose the meaning of the sentence out of two choices, one affectively neutral and one threatening. For example, the sentence “The index finger was placed on the tray” was first shown, followed by two words, “pointing” and “amputation”.
Results showed that trait anxious individuals were significantly more likely to choose the threatening interpretation (i.e., amputation) of the ambiguous sentences. Similarly, using a modified, self-paced RSVP paradigm, MacLeod and Cohen (1993) found that after the presentation of an ambiguous sentence (e.g., “The doctor examined little Emily’s growth”), trait anxious individuals showed faster response latency to a negative interpretation of the sentence (e.g., “Her tumor had changed little since last visit”) relative to a neutral interpretation (e.g., “Her height had changed little since last visit”). It was hypothesized that the comprehension latency would increase if the text shown was inconsistent with one’s initial interpretation of the sentence (e.g., Haberlandt & Bingham, 1978; Keenan, Bailet & Brown,
1984). Therefore, MacLeod and Cohen’s (1993) findings suggested that trait anxious individuals tend to interpret ambiguity in a negative way (see also Calvo & Eysenck, 1995;
MacLeod, 1990). Some studies (Byrne & Eysenck, 1993; Eysenck, MacLeod & Matthews,
1987) presented homophones to participants auditorily (e.g., /dʌɪ/), and found that trait 56 anxious individuals were more likely to interpret them in a threatening way (e.g., die) rather than in a neutral way (e.g., dye; see also Hadwin, Frost, French & Richards, 1997). These findings showed that when a cue could be interpreted in alternative ways, trait anxious individuals were likely to interpret it in a threatening way. The findings were consistent with the idea that trait anxious individuals have a cognitive schema that is biased to threat (Beck,
1976; Beck & Emery, 1985), and this schema would induce biases on the process of primary threat appraisal (Eysenck, 1997).
Threat estimation bias: The literature also suggests that trait anxious individuals show biased judgements of the probability or cost of uncertain negative events, leading to negative expectations for future events. Butler & Mathews (1987) found that trait anxious university students showed a significantly higher level of subjective estimations of failing an upcoming examination. Interestingly, this overestimation of subjective risk ratings was specific to themselves, in other words, when trait anxious individuals were to judge how likely the negative events would occur to others, they showed no inflation in their subjective ratings.
Since the outcome of the examination was unknown, the results supported the notion that trait anxious individuals have an inflated subjective threat estimation under ambiguity, and this estimation bias seems to be confined to themselves. In a text completion task, Stöber (1997) gave participants texts of various scenarios, and instructed them to complete the texts by choosing the most plausible risk description. It was found that trait anxious individuals showed an inflated subjective estimation of risk specific to negative events. That is, when asked to judge about the likelihood of a positive event, no probability biases were found in trait anxious individuals. Mitte (2007) found similar results, in addition to the finding that trait anxious individuals had a heightened perception of subjective cost of negative events. These findings suggest that trait anxious individuals show biased estimation of the probability of an imminent threatening event, and also an increase in perceived cost of uncertain negative events. 57
Threat appraisal in fear learning
As previously discussed, threat appraisal reflects the anticipatory anxiety experienced depending on how one perceives the level of threat of an imminent event. Carr (1974) proposed that this perception of threat is composed of two elements, the perceived probability of a threatening event and the perceived cost of the event. This notion is supported by the findings in threat and cost estimation biases among trait anxious and clinically anxious individuals (e.g., Butler & Mathews, 1983, 1987; Mitte, 2007). Similarly, threat expectancy has been suggested to comprise of the same two constructs that make up threat appraisal.
Paterson and Neufeld (1987) proposed that anticipatory anxiety (i.e., threat expectancy) not only reflects the perceived likelihood of a negative event to occur, but also the perceived cost of the event. They argued that even if an imminent event is very likely to happen, if it is not perceived to be highly undesirable, it will not provoke any anticipatory anxiety. In contrast, an event that is perceived to be associated with a high cost will more likely trigger anticipatory anxiety, and hence an increase in threat appraisal. In parallel to fear conditioning, the probability of a threatening event corresponds to the reinforcement rate of the aversive US.
That is, US expectancy increases along with an increase in the reinforcement rate of an aversive US, and vice versa, similar to an increase in threat appraisal when there is a high perceived probability of a threatening event. On the other hand, the perceived cost corresponds to US intensity. As such, threat expectancy not only measures the conditioned fear in human studies (Reiss, 1980), but also reflects the cognitive threat appraisal to an imminent threat cue.
Fear learning in trait anxiety
As discussed in Chapter 1, fear learning has been suggested as a primary mechanism in the etiology of anxiety disorders. Therefore, if trait anxious individuals are at risk of developing an anxiety disorder after a traumatic experience, one may suggest that fear 58 acquisition among anxious individuals would be different from the low anxious population. It has long been suggested that trait anxious individuals would show stronger and more rapid fear acquisition to the CS+ in an aversive conditioning paradigm (see Eysenck, 1967; Levey
& Martin, 1991), that is, individuals high in trait anxiety are suggested to show enhanced conditionability to an aversive CS+. Some empirical studies showed supportive evidence for this notion, with neurotic individuals (e.g., Santibáñez-H, & Schroeder, 1988) and trait anxious individuals (e.g., Zinbarg & Revelle, 1989) showing increased physiological responses to the CS+ when compared to non-neurotic/low anxious individuals. Similarly, some neuroimaging studies also showed heightened amygdala reactivity to the aversive cue in trait anxious individuals relative to the control group (e.g., Indovina, Robbins, Nunez-
Elizalde, Dunn & Bishop, 2011). However, the findings on enhanced conditionability in trait anxious individuals have not been consistently found in the literature; such findings failed to be replicated in a series of fear conditioning studies (e.g., Chan & Lovibond, 1996;
Gazendam, Kamphuis & Kindt, 2013; Haddad, Pritchett, Lissek & Lau, 2012; Otto et al.,
2007; Sehlmeyer et al., 2011). Instead, some differential conditioning studies found that trait anxious individuals showed elevated self-reported threat expectancies and fear ratings (e.g.,
Gazendam et al., 2013; Kindt & Soeter, 2014; Torrents-Rodas et al., 2013) and physiological responses (e.g., Gazendam et al., 2013; Grillon & Ameli, 2001) to the safety cue (i.e., CS-).
Other similar studies have found that individuals high in trait anxiety fail to inhibit their fear responses to the CS+ in the presence of an inhibitor (Haaker et al., 2015).
Continuity of trait anxiety
Until now, trait anxiety has been discussed as a predispositional factor for the development of anxiety disorders, and evidence has been reviewed on how trait anxious individuals differ from the normal population, especially on the cognitive aspect of threat appraisal and the acquisition of conditioned fear. It has long been assumed that trait anxiety is 59 a continuous dimension, and clinically anxious patients sit at the extreme end of it. In fact, there are numerous findings that support the notion of continuity between anxiety in high trait anxious individuals and clinically anxious patients. First, arguably the most supportive evidence is that individuals high in trait anxiety, neuroticism and negative affect, while controlling other factors, are more likely to develop anxiety disorders and severe anxiety symptoms in longitudinal studies (e.g., Chambers, Power & Durham, 2004; Gershuny & Sher,
1998; Jorm et al., 2000). Secondly, negative affectivity, a personality trait closely associated with trait anxiety (see Watson & Clark, 1984), and trait anxiety itself, are higher in individuals with diagnoses of anxiety disorders, for example, specific phobias (Amies, Gelder
& Shaw, 1983; Watson, Clark & Carey, 1988), PTSD (Eberly, Harkness & Engdahl, 1991) and OCD (Hirschfeld & Kleman, 1979; Watson et al., 1988; also see Clark, Watson &
Mineka [1994] for a review). Thirdly, anxiety patients show similar cognitive biases to trait anxious individuals.
However, while trait anxious individuals show cognitive bias to threat in general, anxious patients showed greater cognitive biases specifically to threats that are relevant to their disorders. For instance, studies that used the modified Stroop task found that panic disorder patients showed greater latency to name panic-relevant words (e.g., catastrophe and bodily sensation words) than to neutral words (McNally, Riemann, Louro, Lukach & Kim,
1990; McNally, Riemann & Kim, 1992) and to social threat words (Hope, Rapee, Heimberg
& Dombeck, 1990). Panic disorder patients also showed shorter comprehension latency to the negative interpretation in a text comprehension task, but this effect was specific to interpretations involving bodily sensation (Clark et al., 1988). This indicates that patients suffering from panic disorders showed interpretation bias to ambiguous cues that were specific to panic-relevant interpretations. Similarly, social phobics showed attentional biases to social-relevant threat cues in the modified Stroop task (e.g., Hope et al., 1990; Mattia,
Heimberg & Hope, 1993) and the dot-probe task (e.g., Amir, Elias, Klumpp & Przeworski, 60
2003; Asmundson & Stein, 1994). They also showed negative interpretation towards their own behaviours (e.g., Rapee & Lim, 1992; Stopa & Clark, 1993). That is, social phobics perceived their own behaviour to be more negative, but they did not show such bias when observing others performing the same action. Anxious patients also showed biases in threat estimation like their trait anxious counterpart, with empirical studies showing patients overestimating the probability of risk (Butler & Mathews, 1983; Nesse & Klaas, 1994).
Last but not least, both trait anxious individuals and clinically anxious patients show abnormal fear acquisition. As mentioned previously, some findings indicate that trait anxious individuals show enhanced conditionability to a threatening CS+, in terms of more rapid and stronger CR formation (e.g., Santibáñez-H, & Schroeder, 1988; Zinbarg & Revelle, 1989), and enhanced amygdala activity (e.g., Indovina et al.,2011). Similarly, anxious patients showed increased conditioned fear to the CS+ relative to the control group. For example, social phobics showed larger EMG startle responses to facial stimuli followed by negative criticism (CS+; Lissek et al., 2008a), spider phobics showed enhanced skin conductance to cues that predicted spider pictures (Schweckendiek et al., 2011), while PTSD (Orr et al.,
2000) and GAD patients (Thayer, Friedman, Borkovec, Johnsen & Molina, 2000) showed increased physiological responses to innocuous cues followed by shock or threat words respectively. Neuroimaging studies have also found heightened amygdala activation in anxiety patients when the threatening CS+ was presented (e.g., Birbaumer et al., 1998; Rauch et al., 2000; Schneider et al., 1999). However, similar to the findings in the trait anxiety literature, results supporting the idea of enhanced conditionability to CS+ in patients have been mixed (e.g., Grillon & Morgan III, 1999; Hermann, Ziegler, Birbaumer & Flor, 2002;
Lissek et al., 2009). Instead, these studies tended to find elevated fear responses to the safety cue (i.e., CS-), supporting the hypothesis that failure to inhibit fear responses to safety is a biomarker of anxiety disorders (Davis, Falls & Gewirtz, 2000). Furthermore, studies have 61 found elevated fear responses to the CS+ in the presence of an inhibitor (Jovanovic et al.,
2009, 2010), consistent with Davis et al.’s (2000) theory.
In summary, a large body of literature provides evidence for the hypothesized continuum in anxiety between normal and clinical populations. Both trait anxious individuals and anxious patients report high level of negative affectivity and anxiety in self-administered questionnaires, show similar cognitive biases to threat and abnormal fear acquisition in fear conditioning studies.
Summary
Not every individual develops anxiety disorders after experiencing traumatic events.
Trait anxiety has been suggested to be a predispositional factor underlying the proneness to developing anxiety disorders. Neurobiological and physiological models have been developed to account for trait anxiety, but have received limited support. On the other hand, a substantial amount of evidence suggests that trait anxiety affects the etiology of anxiety disorders via mechanisms of cognitive biases to threat and abnormal fear acquisition. Evidence like this also provides support for the idea that anxiety is a continuum, and clinically anxious patients sit at the extreme end of it.
So far, we have covered generalisation in humans and how trait anxiety affects fear learning. In the next chapter, these topics will be combined. The role of fear generalisation in anxiety disorders will be discussed. Additionally, the literature of trait anxiety effect on fear generalisation will be reviewed. Research questions arise from the literature will then be addressed.
62
Chapter 4:
The effect of clinical and trait anxiety on fear generalisation
63
Clinical anxiety and fear generalisation
Empirical evidence has been found to support the notion of over-generalisation of fear as a pathogenic marker of anxiety disorders (Kaczkurkin et al., 2017; Lissek et al., 2010,
2014). In these studies, participants received differential training, with a large circle followed by an electrical shock (CS+), and a small circle (CS-) paired with no shock, counterbalanced across groups. In the subsequent test phase, generalisation stimuli (GSs) intermediate between the two CSs were presented (see Fig. 6). Compared to the non-clinical control group, PD
(Lissek et al., 2010), GAD (Lissek et al., 2014) and PTSD patients (Kaczkurkin et al., 2017) showed a greater degree of fear responding averaged across all GSs, and significantly flatter generalisation gradients in behavioural risk ratings and physiological responses (Lissek et al.,
2010, 2014) and neuro-imaging responses (Kaczkurkin et al., 2017). The flatter generalisation gradients were characterized by two features. First, anxious patients showed more fear responding to the GSs that were more similar to the safety cue (e.g., C1; see Fig. 6), and to the safety cue itself (Kaczkurkin et al., 2017; Lissek et al., 2010). Secondly, anxious patients showed no differences in fear responding to CS+ and the GSs that mostly resembled the threat cue (e.g., C4; see Fig. 6). Given that the patients did not show heightened fear responses to
CS+ on any measure, the authors interpreted that patients exhibited over-generalisation of fear rather than a general elevation of responses to all stimuli. The two features of the flat gradients were interpreted as markers of over-generalisation of fear: the elevation of fear
Figure 6. Adapted from Lissek et al. (2010). Stimulus dimension with increasing circle sizes from CS- to CS+ (counterbalanced across groups). C1, C2, C3 and C4 are generalisation stimuli intermediate between the two CSs that were presented in test.
64 responding to stimuli towards the direction of CS- was interpreted as excessive generalisation of fear from CS+, while the non-differential responding to CS+ and the GSs most perceptually similar to it was due to anxious patients having a lower threat detection threshold. However, these patterns are susceptible to alternative explanations. Given that empirical evidence has shown that anxious patients exhibit a deficit in safety learning (e.g.,
Grillon & Morgan III, 1999; Peri, Ben-Shakhar, Orr & Shalev, 2000), it is possible that the elevation of fear responding to CS- and the GSs similar to it is a consequence of insufficient generalisation of safety learning. Furthermore, given that the CS+ and CS- lay on the same dimension differing in size, Lissek’s experimental procedure is unable to distinguish whether the elevation in fear responses to safe cues in anxious patients is due to over-generalisation of fear from CS+ or deficient safety learning to CS-, or a combination of both.
Nonetheless, some past studies that found increased fear responding to safety cues among anxious patients can be explained alternatively as a result of over-generalisation of fear. For instance, Grillon and Morgan III (1999) used a blue circle as the CS+ and a green circle as the CS-. It was observed that PTSD patients showed a higher level of EMG eyeblink startle to the CS- when compared to the healthy control group. The increase in responding to the safety cues in patients could be attributed to an over-generalisation of fear along the blue- green continuum. Similarly, after discrimination training between a bowl (CS+) and a mug
(CS-), Lissek et al. (2009) found that PD patients showed more shock expectancies and EMG startle responses to the CS-. This increase in fear responding to CS- in the clinical group could also be attributed to over-generalisation of fear, since both CSs belong to the same category
(i.e., tableware), promoting fear generalisation from CS+ to CS- (see Dunsmoor & Paz, 2015;
Dymond, Dunsmoor, Vervliet, Roche & Hermans, 2015).
65
Trait anxiety and fear generalisation
Despite increasing evidence that suggests a connection between over-generalisation of conditioned fear and anxiety pathology (e.g., Kaczkurkin et al., 2017; Lissek et al., 2010,
2014), it cannot be distinguished whether excessive fear generalisation is a consequence of anxiety disorders, or whether it is a vulnerability factor for their development. In addition, clinical samples introduce a great deal of comorbidity as well as sequellae of their clinical condition. Therefore, it is important to study individuals at risk of developing anxiety disorders, for instance, individuals high in trait anxiety, and examine if their pattern of fear generalisation would be similar to that of the clinical samples. Surprisingly, there is little research on how trait anxiety affects fear generalisation, and mixed evidence has been found.
Haddad et al. (2012) examined how trait anxiety affects fear responses to safety cues.
Participants were trained with one CS+ and two CSs-, where one CS- was perceptually similar to CS+ (i.e., similar CS-) and the other one not (i.e., dissimilar CS-). A higher level of
EMG eyeblink startle to the similar CS- was observed among highly anxious individuals, but not to the dissimilar CS-. The results provided some evidence that anxious individuals show greater fear generalisation from CS+ to a similar CS-. The authors argued that such results could not be explained by a general elevated fear response to safety cues, otherwise an increase in responding should have been observed to both safety cues. In another study carried out by Lommen, Engelhard and van den Hout (2010), participants underwent differential training between a black (CS+) and white circle (CS-). In the subsequent test phase, they were shown circles with shades of grey along the black-white continuum, and participants were given the chance to avoid a potential electric shock by pressing the spacebar during stimulus presentation. They found that when given a relatively long latency to respond, individuals high in neuroticism had a tendency to avoid more stimuli, including GSs that were more similar to CS-. Although the results can be interpreted as neurotic individuals showing over- 66 generalisation of fear based on more avoidance behaviours to stimuli more similar to CS-, the authors did not plot the full gradient of avoidance across the stimulus dimension. Instead, the mean avoidance responses were analyzed averaged across all stimuli, hence making the interpretation of over-generalisation of fear difficult. Using a similar differential conditioning paradigm with stimuli differing along a size dimension, Kaczkurkin and Lissek (2013) found that participants high in obsessive-compulsive symptoms showed more conditioned fear to novel stimuli that resemble CS+ the most, but not to stimuli that are similar to CS-. However, this effect was only found in participants high in threat estimation in a post-hoc analysis, a subscale in the Obsessive Beliefs Questionnaire (OCCWG, 2005).
There have also been other studies that did not find any effect of trait anxiety on fear generalisation that can be interpreted more straightforwardly. Using the same paradigm developed by the Lissek lab (2008b, 2010, 2014), Torrents-Rodas et al. (2013) trained participants to discriminate between a large circle followed by shock (CS+) and a small circle paired with no shock (CS-). In test, participants were presented with stimuli differing in size, intermediate between both CSs along the stimulus dimension (see Fig. 6). Although the high anxious group showed significantly higher risk ratings to stimuli most similar to the safety cue compared to the low anxious control group, there were no significant differences in the shape of the generalisation gradients between anxiety groups, and all groups showed similar ratings to stimuli most similar to CS+. Furthermore, no differences were found in the psychophysiological responses to the generalisation cues. The authors therefore concluded that their study failed to find evidence that trait anxiety has any effect on fear generalisation.
Using a similar paradigm, Arnaudova, Krypotos, Effting, Kindt and Beckers (2017) presented stimuli along a black-white continuum after differential conditioning, and found no differences in fear generalisation across measures between high- and low-neurotic individuals.
67
Strong and weak situations
The investigation of the effect of anxiety on fear generalisation is still in its early phase, and a discrepancy between the effect of clinical and trait anxiety on fear generalisation has been observed. It seems that evidence for over-generalisation of fear in clinical studies is mostly positive, while some studies examining individuals at risk of developing anxiety disorder found a null trait anxiety effect on the generalisation of conditioned fear (Arnaudova et al., 2017; Torrents-Rodas et al., 2013). There are several possible reasons accounting for this discrepancy in findings. First, no strong conclusion can be made for the findings that anxious patients exhibit excessive fear generalisation, since there are only a few studies demonstrating the effect of clinical anxiety on fear generalisation. Secondly, it is possible that the magnitude of any trait anxiety effect on fear generalisation is not as strong as clinical anxiety. Thirdly, it is possible that the experimental configurations were not ambiguous enough for the effect of trait anxiety to be observed. It has been argued that the typical differential fear conditioning paradigm represents a ‘strong situation’, consisting of cues with clear threat values (Lissek, Pine & Grillon, 2006). In this case, most participants would show adaptive fear to the cues, that is, more fear responding to CS+ and less fear responding to CS-.
This makes it difficult for any potential individual differences to be observed in fear acquisition (Beckers et al., 2013). In contrast, a ‘weak’ situation comprises a more ambiguous experimental configuration. For example, Lissek et al. (2006) argued that partial reinforcement of the CS+, where the CS+ does not predict threat all the time, may induce a
‘weak’ situation. This ambiguity of threat information may allow trait anxiety to manifest its effect on fear acquisition (Beckers et al., 2013). Although Torrents-Rodas et al. (2013) used a partial reinforcement schedule for CS+ in an attempt to create a ‘weak situation’, two features of their paradigm potentially rendered this manipulation less effective. First, the usage of differential conditioning paradigm inevitably introduced a safety cue. Although the process of generalisation is intrinsically ambiguous since the threat value of the novel GSs is unknown, 68 having been trained with a CS- provides a reference point for participants to extrapolate that some novel cues can be safe.
Secondly, the nature of the paradigm being used may contribute to the null anxiety effect. Torrents-Rodas et al. (2013) argued that stimuli intermediate between CS+ and CS- were ambiguous since their threat value was unknown. However, the test stimuli differed from each other in a linear fashion along the dimension (see Fig. 6), rendering it straightforward for participants to infer the threat value of each test stimulus based on their similarity to CS+ or CS-. Furthermore, an intensity stimulus dimension was used. For example, circles at one end of the dimension are larger in size than those at the other end in the dimension used in Torrents-Rodas et al.’s (2013) study. These intensity stimulus dimensions may further encourage participants to respond accordingly to the intensity of stimulus (e.g., larger circle means more likely to get shock, and vice versa for smaller circles).
In fact, a phenomenon called intensity generalisation has sometimes been observed in the animal literature (as discussed in Chapter 2). When an intensity dimension was used, animals did not show the typical generalisation decrement to stimuli dissimilar to the reinforced cue.
Instead, animals showed increased responding to novel stimuli that were higher in intensity relative to the reinforced cue (e.g., Huff et al., 1975; Lawrence, 1973). These factors can arguably disambiguate the generalisation task and turn the experimental configuration into a
‘strong situation’, reducing the opportunity to detect any potential trait anxiety differences in fear generalisation.
The effect of trait anxiety on fear learning in the presence of ambiguity
Some fear conditioning studies that found trait anxiety effect on fear learning contain a certain degree of ambiguity, which arises either from experimental manipulation or from the nature of the paradigm. In Chan and Lovibond’s (1996) study, participants were trained in a conditioned inhibition procedure, where one stimulus was followed by an electric shock (A+), 69 except when it was presented in compound with another stimulus (AB-). In the overall data, the high anxious group showed slower differentiation in shock expectancy ratings to A+ and
AB- when compared to the low anxious group. The high anxious group showed higher overall expectancy ratings than the low anxious group primarily due to the slow learning of safety to stimulus B. Nonetheless, no differences were observed in the skin conductance data to both cues between groups. Interestingly, the authors further categorized participants into two groups, those who were aware of the CS-US contingencies (i.e., A led to shock, AB led to no shock) and those who were not. Among the high anxious participants, those who were aware of the CS-US contingency showed clear discrimination between A+ and AB- that was comparable to the low anxious group; however, the unaware high anxious individuals showed no differential responding to the cues. Instead they gave high shock expectancy ratings to both cues. That is, they responded to the safe compound cue (AB-) as if it was a threat cue (A+).
Since the unaware participants did not know which cue predicted electric shock, all cues in the task effectively became ambiguous. In fact, the overall higher expectancy ratings observed at the group level was mainly driven by trait anxious individuals who were not aware of the
CS-US contingency. Therefore, the study provided strong evidence that trait anxious individuals show a bias in threat expectancy, but only when they perceived the situation as ambiguous.
Recently, Boddez et al. (2012) studied the effect of trait anxiety on fear learning using a blocking procedure in an aversive conditioning paradigm. Participants received additivity training before the conditioning task, which has been shown to enhance the blocking effect
(Lovibond et al., 2003; Mitchell & Lovibond, 2002). The results showed a positive correlation between trait anxiety and shock expectancy to the blocked stimulus. In other words, trait anxious individuals showed elevated threat appraisal to the blocked stimulus which usually receive low level of responding. Since the blocked stimulus had never been presented by itself, its predictiveness of shock was unknown, arguably rendering it ambiguous. 70
Furthermore, trait anxious individuals did not show the same elevation in threat appraisal to other control cues (e.g., reinforced cue that had not been blocked), hence this bias in threat appraisal among anxious individuals is suggested to be specific to the ambiguous blocked cue.
One may argue that since participants had received additivity training, they were presumably able to infer that the blocked stimulus was not predictive of shock. Therefore, the results could be interpreted alternatively as trait anxious individuals failing to inhibit fear responses to a safety cue. However, participants only received one trial of additivity training without further additivity instructions, making it questionable if participants were actually able to attribute the non-causal status of the compound stimuli (AB) to the blocked stimulus (B).
Hence, the results favor the interpretation of threat appraisal bias to ambiguity among trait anxious individuals.
More recently, Chen and Lovibond (2016) tested whether individuals high in intolerance of uncertainty (IU) would show elevated threat appraisal to ambiguous cues. IU has been argued to be a vulnerability factor for the development of anxiety disorders, in particular GAD (Dugas, Gagnon, Ladouceur & Freeston, 1998; Dugas & Robichaud, 2007) but also PTSD (Fetzner, Horswill, Boelen & Carleton, 2013) and social anxiety disorder
(SAD; Boelen & Reijntjes, 2008). Prior to the conditioning task, participants were instructed how the majority of the cues were related to an aversive outcome. One cue was a perfect predictor of the aversive outcome (CS+), one cue was never followed by an outcome (CS-), and one cue predicted the outcome 50% of the time (uncertain cue). However, the experimenters surprised participants by presenting a fourth cue in the conditioning task that also predicted the outcome at 50% of the time, but this had not been mentioned in the instructions (labelled as ambiguous cue). The results showed no differences in conditioned fear to either of the cues that had clear threat values (CS+ and CS-) between individuals of high and low IU. Interestingly, high IU individuals showed increased threat appraisal to the ambiguous cue but not to the uncertain cue when compared to the low IU individuals. This 71 suggests that anxiety-prone individuals show threat estimation bias to ambiguity but not to uncertainty.
Ambiguity and uncertainty
Although both ambiguity and uncertainty involve imperfect prediction of an outcome, there is a substantial conceptual difference between the two constructs. In an uncertain condition, the relative likelihood of an outcome (i.e., probability of the outcome) is often known or can be estimated; therefore individuals can make their decision based on the information provided. In contrast, an ambiguous condition refers to the complete lack of information regarding outcome probability (see Ellsberg, 2001). Ellsberg’s (1961) two-urn choice theory elegantly demonstrates the difference between ambiguity and uncertainty, despite originally being used to differentiate ambiguity aversion from risk aversion (Ellsberg,
1961). In this example, Urn 1 contains 50 red balls and 50 black balls, while Urn 2 contains
100 balls but the ratio of red to black balls is unknown. Urn 1 is classified as an uncertain condition, because although it is still unknown if the next ball drawn from the urn would be red or black, participants can still accurately adjust their judgement based on the outcome probability provided. On the other hand, Urn 2 is classified as an ambiguous condition.
Similar to Urn 1, it is still unknown whether the ball drawn would be red or black, but participants now have no clear basis to calibrate their outcome expectancy. The ambiguous situation described here is similar to the aforementioned fear conditioning studies in trait anxiety, where there was a lack of information regarding the causal status of the cues (Boddez et al., 2012; Chan & Lovibond, 1996; Chen & Lovibond, 2016), while the uncertain situation corresponds to the uncertain cue in Chen and Lovibond’s (2016) study. The finding that threat estimation bias was observed among anxious individuals only under ambiguity but not uncertainty suggests how threat is appraised differently in given circumstances. Under uncertainty, anxious individuals can still calibrate their threat appraisal based on the objective 72 outcome probability provided; however, when there is a complete lack of information regarding threat probability, anxious individuals will then make threat expectancy based on their cognitive schemas or personal beliefs, which are presumably biased towards threat (Beck
& Clark, 1988; Beck & Emery, 1985; see also Alloy & Tabachnik, 1984).
Summary and research aims
As reviewed, fear learning is proposed as a major factor in the etiology of anxiety disorders. The fear conditioning paradigm is a well-established procedure for studying the acquisition of fear and also fear generalisation, with over-generalisation believed to be a pathogenic marker of anxiety disorders. Empirical studies have also pointed to the effect of cognitive processes on the learning and generalisation of conditioned fear, for example, the high consistency between threat beliefs and conditioned fear.
Given the importance of furthering our understanding of fear generalisation, it is still unclear whether over-generalisation of fear is a consequence of anxiety disorders or a vulnerability factor that expresses itself in individuals at risk of developing anxiety disorders, for instance trait anxiety. There have been only a few studies that have investigated the effect of trait anxiety on fear generalisation, and mixed results were found. However, these studies used paradigms that may constitute a ‘strong situation’, which lowers the level of ambiguity of the situation and reduces the opportunity to observe any trait anxiety differences. Empirical studies that found increased threat appraisal to ambiguous threat among trait anxious individuals further support the idea that the effect of trait anxiety manifests in the presence of ambiguity. Therefore, the current project aims to address the following research questions in the first empirical chapter:
- Is over-generalisation of fear a characteristic of trait anxious individuals? 73
- Will trait anxious individuals be more likely to show over-generalisation of fear when
a ‘weak’ situation is induced, that is, when the level of ambiguous threat in the
situation is high?
- Will trait anxious individuals differ in inferred rules in fear generalisation when an
arbitrary stimulus dimension is used, and if so will such rules modulate differences in
the observed generalisation?
74
Chapter 5:
The effect of trait anxiety on fear generalisation across a
perceptual stimulus continuum
Published as:
Wong, A. H. K., & Lovibond, P. F. (2018). Excessive generalisation of conditioned fear in
trait anxious individuals under ambiguity. Behaviour Research and Therapy, 107, 53-
63. https://doi.org/10.1016/j.brat.2018.05.012 75
In traditional animal studies of generalisation, the generalisation gradients were plotted across stimuli on the same continuum that differ from each other perceptually in a quantitative way. The typical finding in such studies was a peaked gradient, with the peak responding to CS+ and a gradual decrease in response magnitude to stimuli that were more dissimilar to the reinforced value. However, as reviewed in Chapter 2, this generalisation pattern could not be readily replicated in humans. Thomas and Mitchell (1962) suggested that a higher-order process such as labelling, may help the formation of a relational rule that is responsible for the unique generalisation patterns observed in humans. More recently, studies in our lab have been designed specifically to distinguish the unique relational rules made during the generalisation task (Ahmed & Lovibond, 2018; Lee, Hayes & Lovibond, 2018;
Wong & Lovibond, 2017). These studies assessed individual relational rules made during the conditioning task via the administration of post-experimental questionnaires, and examined how generalisation gradients were influenced by the inferred rules. After categorizing participants into different rule subgroups, it was consistently found that inferred rules greatly affect the shape of the generalisation gradients. Furthermore, gradients formed from different rules were distinctly different from each other, which suggests that the overall generalisation gradient may be potentially misleading, as it may comprise of distinctive gradients formed from different relational rules. Accordingly, the current study addressed the importance of cognitive mediation in human fear generalisation, and used similar methods to identify how relational rules affect fear generalisation. Furthermore, the current study examined the interplay between trait anxiety and cognitive mediation, and how this may affect fear generalisation.
According to Eysenck & Calvo (1992), the processing efficiency of working memory among trait anxious individuals is impaired in two ways. First, when facing a stressful situation, trait anxious individuals would have excessive worrisome thoughts that consume the limited capacity of working memory. Secondly, trait anxious individuals experience more 76 state anxiety in stressful situations (Beck & Emery, 1985; Speilberger, 1975). In order to minimize the anxiety state, individuals may use extra cognitive effort and strategies to reduce the anxiety they experience. This process consumes cognitive resources and again uses up the limited capacity of the working memory. This idea has been supported by some empirical studies. For instance, some studies presented a series of digits or letters and participants were requested to memorize them as accurate as possible (e.g., Calvo & Ramos, 1989; Eysenck,
1989a, 1989b). Participants were informed that task performance determines intelligence level, therefore imposing ego-threat on participants and creating a stressful situation. The researchers introduced a random probe at the time when participants were encoding and processing the digit or letter spans, and participants were required to respond to the probe as fast as possible. It was found that trait anxious individuals had a significantly higher latency to respond to the probe, and this was not due to anxious individuals having slower reaction time in general. Similarly, some studies introduced a secondary loading task on top of a difficult primary task (e.g., Calvo & Ramos, 1989; Derakshan & Eysenck 1998; MacLeod &
Donnellan, 1993). Trait anxious individuals were found to perform significantly worse in the primary task when the concurrent secondary task was cognitively demanding. These findings suggest that anxious individuals’ processing capacity was impaired, presumably by the excessive worrisome thoughts of performing the task badly, or the usage of cognitive strategies to reduce the anxiety state. In light of this, it was hypothesized that fewer trait anxious individuals in the current study would come up with relational rules, presumably because their cognitive resource was preoccupied by the anticipation of getting shocked.
As reviewed, previous studies (Arnaudova et al., 2017; Torrents-Rodas et al., 2013) that did not find a trait anxiety effect on fear generalisation may have unintentionally introduced an unambiguous situation by using a ‘strong’ paradigm, potentially hiding any individual differences in fear acquisition and generalisation. Therefore, the current study used a single-cue conditioning procedure, aiming to introduce a high level of ambiguity by using 77 this ‘weak’ paradigm (Beckers et al., 2013; Lissek et al., 2006). The single-cue conditioning procedure arguably induces a higher level of ambiguity compared to the differential conditioning procedure. In a single-cue conditioning procedure, participants only receive CS+ trials; therefore there is less information available to guide generalisation because of the lack of a safety cue (see Homa, Sterling & Trepel, 1981). This notion is supported by a meta- analysis that found a greater clinical anxiety effect on fear acquisition in single-cue conditioning procedures (Lissek et al., 2005; but see Duits et al., 2015). An additional advantage of using a single-cue conditioning procedure is its ability to focus on how fear is generalised from CS+. As discussed in Chapter 4, it is unclear whether higher responding to stimuli intermediate between CS+ and CS- in a differential conditioning design is due to excessive fear generalisation of CS+, inadequate generalisation of safety learning, or a combination of both. Without the interference of a CS-, the current study was able to solely focus on how trait anxiety may affect the generalisation of fear from CS+. Furthermore, the current study used a non-intensity stimulus dimension that aimed to maintain the level of ambiguity of the generalisation process. As discussed previously, the usage of an intensity stimulus dimension may disambiguate the generalisation task. Preliminary evidence in our lab
(Ahmed, 2014) has consistently observed an increasing linear generalisation pattern in single- cue conditioning studies using intensity stimulus dimensions (e.g., amount of chemical, brightness, circle size). This suggests that participants may infer the outcome probability by the intensity or magnitude of the novel stimulus relative to the CS+, which arguably decreases the level of ambiguity of the generalisation task.
Therefore, the current study aimed to examine the effect of trait anxiety on fear generalisation by inducing an ambiguous situation via the utilization of a single-cue conditioning procedure and a non-intensity stimulus dimension. In addition, it sought to examine how cognitive mediation affects fear generalisation. The effect of trait anxiety on both overall generalisation gradients and gradients for individual rule subgroups were 78 examined in order to investigate whether trait anxiety may have different effects on generalisation in different rule subgroups.
Method
Participants
Undergraduate students were recruited as participants who received course credit or
AUD $15 for participation. Participants were pre-screened with the DASS-21 (Lovibond &
Lovibond, 1995a) via the recruiting SONA system, which is a shorter version of the original
DASS that has been shown to have similar psychometric properties to the DASS (Antony et al., 1998; Henry & Crawford, 2005). Participants with a DASS-anxiety score of 4 or below were recruited to the low anxious (LA) control group, while those with a DASS-anxiety score of 18 or above were assigned to the high anxious (HA) group. Participants were re- administered the DASS-21 at the time of testing, and only those whose DASS anxiety scores remained consistent with the recruitment criteria were included. Forty participants were recruited in each group. The recruitment strategy was to continue recruiting until there were
40 participants in each group who met inclusion criteria (see Results for more detail). This led to a total recruitment of 113 participants, with 33 excluded. The final sample comprised 80 participants (43 females) with a mean age of 21.1 years (SD = 3.8).
Apparatus and materials
Participants were tested individually in an experimental room. A 64-cm DELL® LCD computer monitor was used to present the experimental instructions and stimuli. A computer equipped with Matlab software (with Pschophysics Toolbox extensions: Brainard, 1997; The
MathWorks Inc., 2014) was located outside the experimental room, which generated the stimuli presented to the participants and recorded the expectancy ratings, while another 79 computer controlled AD instruments equipment to record the skin conductance data via
GRASS® silver disc electrodes at a sampling rate of 1000/s throughout the experiment.
A symmetrical spatial stimulus dimension was used to minimize any intensity biases
(Ahmed, 2014; Wong & Lovibond, 2017). The stimuli were yellow squares [5.5 x 5.5 cm] with a black dot varying horizontally from left to right (Fig. 7). The location of the dot was manipulated by an equal distance of 0.5 cm from one stimulus to the next. Stimulus E, with the dot in the center, served as the CS+. A red lightning bolt served as the symbolic shock. All stimuli and the symbolic shock were presented in the center of a white background on the computer screen.
Figure 7. Stimulus dimension; stimulus E served as the CS+; all other stimuli served as GSs
Procedure
After signing the consent form, participants were asked to fill in the DASS-21. Shock electrodes were then attached to participants’ fingers, and they were led through a work-up procedure in which they selected a level of shock that was ‘definitely uncomfortable but not painful’. Participants were then taken into the experimental room and had the skin conductance electrodes attached. As shown in Table 1, the study consisted of an acquisition phase and a test phase, with each phase divided into two stages, similar to previous fear conditioning studies in our lab (Lee et al., 2018; Wong & Lovibond, 2017).
The acquisition phase consisted of two stages, Acquisition1 and Acquisition2. The shock electrodes were disconnected in the former and reconnected in the latter. The reason for 80 administering electric shock only in Acquisition2 was to minimize habituation to the shock.
The purpose of Acquisition1 was to increase the number of training trials, in order to facilitate learning of the CS-US association.
Table 1. Design of current study
Phase Acquisition1 Acquisition2 Test1 Test2
CS+ (6) / CS* (2) CS+ (3) / CS* (1) CS* (1) CS* (1)
GS- (8) GS- (3)
Note. + indicates shock presentation; - indicates shock omission; * indicates non-reinforced CS+; GS refers to generalisation stimuli; numbers in brackets indicate the number of trials of that type in each phase. Shock electrodes were disconnected during Acquiistion1 and Test1 and reconnected during Acquisition2 and Test2.
Acquisition1 (shock electrodes disconnected). This stage consisted of 8 trials of CS+, reinforced at 75% (6 out of 8 trials were followed by the symbolic shock). CS+ was not fully reinforced for three reasons. First, it avoids a ceiling effect and hence allows room to examine if HA individuals would show elevated fear responding to CS+ (i.e., enhanced conditionability). Secondly, it also allowed room for increased responding to stimuli beyond
CS+ to be observed, for example a rule-based linear gradient (Lee et al., 2018; Wong &
Lovibond, 2017). Thirdly, partial reinforcement of CS+ slows down extinction in the latter test phase (Partial reinforcement extinction effect, see Mackintosh, 1974), which is important because rule-based generalisation is sensitive to extinction (cf. Ahmed & Lovibond, 2018).
The presentation order was pseudo-randomized, so that non-reinforced CS+ trials did not occur twice in a row, and the first and last trials were always reinforced.
Before the conditioning task started, participants were informed that figures would be presented on the computer screen, which may or may not be followed by a shock; they were 81 asked to learn the relationship between the figures and the shock. Participants were then instructed to use the dial to indicate their shock expectancy during stimulus presentation.
Shock expectancy ratings were recorded at the end of the stimulus; hence participants could freely adjust their ratings during stimulus presentation. They were also instructed not to focus on the order of stimulus presentation. This was to minimize superstitious learning arising from the partial reinforcement schedule. Participants were then informed that due to ethical restrictions, the number of shocks were limited, hence setting up the cover story for disconnecting the shock electrodes. They were told that when the shock electrodes were disconnected, only the symbolic shock would appear on the monitor. The trial structure was made up by a 10-s baseline period, a 10-s stimulus presentation, followed by a 2-s period where feedback (symbolic shock) was either presented or not presented. The inter-trial interval (ITI) varied between 10 and 21s, starting from stimulus offset to the onset of the baseline period.
Acquisition2 (shock electrodes connected). When Acquisition1 ended, the experimenter paused the program and went into the experimental room to reconnect the shock electrodes. Participants were informed that they would now be receiving the physical shock along with the symbolic shock. Acquisition2 consisted of 4 CS+ trials, which were identical to Acquistion1 in terms of trial structure, except that the electric shock was delivered in the last 0.5 s of the symbolic shock presentation. The 75% partial reinforcement schedule was maintained in Acquisition2 (3 of the 4 trials were followed by both the physical and the symbolic shock). The presentation order was again pseudo-randomized, so that the first and last trial were always reinforced.
Similar to acquisition, the test phase was divided into two stages, Test1 and Test2, with the shock electrodes disconnected in the former and reconnected in the latter (see Table
1). 82
Test1 (shock electrodes disconnected). The experimenter went into the experimental room again and informed the participants that the shock electrodes would be disconnected again due to ethical reasons. They were also told that neither symbolic nor physical shock would be administered, but they were asked to continue making their expectancy ratings, assuming hypothetically that it was still possible for them to receive a shock. This is conceptually equivalent to the ‘missing data procedure’, which is used to minimize the impact of extinction during testing in causal judgement and prediction tasks (e.g., Shanks & Darby,
1998). This procedure also avoids the confusion that participants may experience when stimuli they expect to be followed by shock are presented alone, potentially prompting them to modify their response strategy. In this stage, all 9 stimuli along the dimension were presented in a randomized order. In other words, the CS+ and 8 GSs of varying degrees of similarity were presented in this phase.
Test2 (shock electrodes reconnected. The experimenter reconnected the shock electrodes and participants were told that it was again possible to receive physical shock (but not the symbolic shock) in this stage, so that skin conductance data for the test stimuli could be collected. In fact, no electric shocks were presented. In addition to CS+, only 3 selected
GSs (C, G and I) were presented in Test2, in a randomized order, in order to minimize extinction. The right-most stimulus was included to maximize sensitivity to positively sloped linear gradients, since our previous research had shown that all participants who reported a linear rule expected shock to follow stimuli to the right of CS+ (Wong & Lovibond, 2017).
When the conditioning task was completed, participants were asked to fill in a 2-page questionnaire (see Appendix A) On the first page, the experimenter wrote down the expectancy ratings that the individual participants had made to the stimuli at the opposite ends of the stimulus dimension (i.e., stimuli A & I) during Test1. Participants were asked to explain why they made these ratings, and to write down in detail any rules/strategies of 83 responding they used. The second page was administered only after the first page was completed, and consisted of 5 statements. Each statement described the relationship between stimuli and shock in terms of different rules (similarity, linear left, linear right, no rule and other). Participants were asked to indicate how much they considered the statement to be true on a 1-100 scale, with 1 being false and 100 being true. Participants were told that if none of the statements described their rule-based responding, they should write down their own description in the ‘Other’ section.
Scoring and analysis
Although expectancy ratings were recorded in both Test1 and Test2, only those in Test
1 were used for data analyses as they covered the whole stimulus dimension. Using a similar paradigm, our previous research has found that expectancy ratings made in Test2 were highly similar to those made in Test1, supporting the validity of expectancy data in Test1 (see Wong
& Lovibond, 2017). For the skin conductance measure, analysis was based on the data collected when the shock electrodes were attached (Acquisition2 and Test2), since this was when physical shock could be delivered and anticipatory anxiety was expected to occur. A low-pass digital filter was applied to cut off any skin conductance activity higher than 50Hz, in order to avoid aliasing. The raw skin conductance data were then log transformed to minimize individual differences. Skin conductance scores for each trial were calculated as the difference between the mean of log skin conductance level (SCL) during the 10-s stimulus presentation and log mean SCL during the 10-s baseline period for that trial.
Planned contrasts were used to compare groups and to assess acquisition and generalisation gradients. For the acquisition data, learning of the CS-shock contingency across trials was analyzed with a linear trend repeated measures contrast. For the generalisation test data, a linear contrast was used to capture any linear gradients across the stimulus dimension, while a quadratic contrast was used to capture peaked, unimodal gradients. Group contrasts 84 were used to compare HA with LA participants, and to compare pairs of rule subgroups identified in the post-experimental questionnaire. Finally, all interactions between the group and repeated measures contrasts were tested to evaluate group differences in trends.
Results
Exclusion of participants
Statistical analyses were applied to participants who satisfied the acquisition criterion, that is, expectancy ratings to CS+ needed to be above 50 averaged across all 4 trials in
Acquisition2. Eleven and eight participants in the HA and LA group were excluded respectively based on the acquisition criterion, suggesting no substantial group differences in fear learning. Participants who did not provide shock expectancy ratings for two or more stimuli during test (i.e. missing data) were also excluded (2 in HA group and 3 in LA group).
Additionally, participants who reported responding based on the presentation order of stimuli in the post-experimental questionnaire and those who misinterpreted the instructions, were also excluded (4 in HA group and 5 in LA group). Altogether, 17 and 16 participants in the
HA and LA group were excluded respectively, leaving 40 participants in each group.
Missing data
Shock expectancy ratings were considered ‘missing’ if participants left the dial in the
Off position during stimulus presentation. This occurred on 0.28% and 0.12% of all the trials in the HA and LA group respectively. Most of the missing data occurred in early acquisition trials, when participants were still learning the experimental requirements. When participants failed to make expectancy ratings in early trials, the experimenter went into the experimental room during the ITI, and reminded them to respond by turning the dial during stimulus presentations. Missing data during acquisition were replaced with the average ratings made during that particular trial across all participants within the group. Missing test data were 85 replaced with the average ratings made during that trial type across all participants within the participants’ rule subgroup.
Anxiety groups
The mean DASS-anxiety scores were 20.0 and 2.2 for the HA and LA groups respectively. The mean shock intensities for both groups were 2.3mA, indicating no group difference in the tolerance of electric shock, F(1,78) = 0.004, p=0.95, n.s.
Acquisition
Figure 8A shows the mean shock expectancy ratings during the acquisition phase for the HA and LA groups. Both groups showed a steady increase in expectancy ratings to CS+, confirmed by a significant main effect of linear trend across trials, F(1,78) = 101.6, p<0.01,
2 ηp = 0.57. There was no overall difference between groups, F(1,78) = 0.03, p=0.86, n.s., and nor was there an interaction between linear trend and groups, F(1,78) = 0.7, p=0.41 n.s., suggesting that there were no differences in acquisition between the anxiety groups.
Figure 8B shows the mean change in log SCL during the last 4 acquisition trials
(Acquisition2) in the HA and LA groups. Skin conductance responding to CS+ decreased over trials in both groups, resulting in a significant linear trend averaged across groups,
2 F(1,78) = 17.0, p<0.01, ηp = 0.17. The skin conductance data did not directly align with the expectancy data, as the level of responding to CS+ decreased across trials for both groups.
This pattern is consistent with previous studies in our lab (Lee et al., 2018; Wong &
Lovibond, 2017), and may be due to several possible factors. First, by the beginning of
Acquisition2, participants had already had the opportunity to learn the association between
CS+ and shock, so there was limited scope for additional learning. In other words, the associative strength may have already reached ceiling by the time of the first reinforced trial.
Second, habituation of skin conductance to the CS and US could be responsible for the 86 decrease in responding across trials. Finally, participants may have become anxious when they were told they were about to receive the first shock but quickly re-adjusted their
Figure 8. Mean shock expectancy ratings (Top panel) and skin conductance level (SCL; Bottom panel) across acquisition trials. HA = High Anxious; LA = Low Anxious. The skin conductance data were collected during Acquisition2, when the shock electrodes were connected.
87 appraisal, resulting in a heightened SCL on the first reinforced trial, and then a decrease in
SCL across the remaining acquisition trials. Similar to the expectancy data, neither the main effect for group nor the interaction between linear trend and groups were significant (highest
F=0.03, p=0.86), suggesting that there was no difference in fear acquisition between groups as measured by skin conductance.
Test Phase
Figure 9A depicts the mean generalisation gradients for the shock expectancy ratings in the HA and LA groups. Both groups showed a peaked gradient with the highest ratings to
CS+ and lower ratings to the test stimuli that were more dissimilar to CS+. This gradient shape resulted in a main effect of quadratic trend across the stimulus dimension, F(1,78) =
2 49.0, p<0.01, ηp = 0.39. The gradient in the HA group was relatively flatter than the one in the LA group, with greater generalisation to the test stimuli. This pattern was supported by two statistical effects. First, HA participants had higher ratings averaged across stimuli, as shown by a significant main effect for the contrast comparing the two groups, F(1,78) = 11.0,
2 p<0.01, ηp = 0.12. Second, quadratic trend was stronger in the LA participants, leading to a
2 significant interaction between quadratic trend and groups, F(1,78) = 8.3, p<0.01, ηp = 0.10.
A significant main effect of linear trend across group was also observed, F(1,78) = 23.8,
2 p<0.01, ηp = 0.23, presumably due to the slightly higher responding to stimuli right of CS+.
However, the interaction between linear trend and group was non-significant, F=0.4, p=0.53 n.s., suggesting no group differences in the linear component of the generalisation gradients.
Interestingly, the HA and LA groups gave very similar shock expectancy ratings to
CS+, which were both very close to the actual reinforcement rate of 75%. To further explore the degree of generalisation in the two groups, an additional exploratory contrast was tested to directly compare CS+ to the test stimuli. Averaged across groups, shock expectancy to CS+ was significantly higher than to the average of the remaining stimuli, F(1,78) = 119.4, p<0.01, 88
Figure 9. Mean overall shock expectancy ratings (Top panel) and skin conductance level (Bottom panel) in the test phases. HA = High Anxious; LA = Low Anxious. The skin conductance data were collected during Test2, when the shock electrodes were connected. 89
2 ηp = 0.61, confirming generalisation decrement to the test stimuli. This comparison also
2 interacted with groups, F(1,78) = 5.9, p=0.02, ηp = 0.07, directly demonstrating greater generalisation decrement in the LA group – in other words, greater generalisation to the test stimuli in the HA group.
Figure 9B shows the overall generalisation gradients for the skin conductance data in the HA and LA groups. The SCL gradients were broadly consistent with the expectancy data, showing a peaked gradient with the peak responding at CS+, supported by a significant main
2 effect of quadratic trend across stimuli, F(1,78) = 10.3, p<0.01, ηp = 0.13. A significant main
2 effect of linear trend across groups was also observed, F(1,78) = 5.2, p=0.03, ηp = 0.06, presumably due to the drop off in responding from CS+ to stimulus I. However, no interactions were found to be significant (highest F=2.6, p=0.11), suggesting there were no reliable differences in the SCL gradients between groups. The contrast comparing CS+ with
2 the generalisation stimuli was significant, F(1,78) = 13.5, p<0.01, ηp = 0.14, confirming generalisation decrement to the generalisation stimuli. Although this decrement appeared to be greater in the LA participants, the interaction with group was not significant for the skin conductance measure, F(1,78) = 1.4, p=0.24, n.s.
Post-experimental questionnaire
Since our previous work had shown that the rules inferred by participants strongly influence their generalisation gradients (Ahmed & Lovibond, 2018; Wong & Lovibond,
2017), we analyzed the questionnaire data to categorize participants into subgroups according to the rules they reported. Two raters, who were blind to the expectancy and skin conductance data, categorized participants into different subgroups based on their questionnaire responses.
This was done by first classifying participants’ self-reported rules from the open-ended question on the first page of the questionnaire. If the reported rule was ambiguous, that participant was categorized according to the rule they endorsed most strongly in the second 90 section of the questionnaire. A high level of consensus was observed between the two raters using Cohen’s Kappa (k = 0.82, p<0.01). Discrepancies between raters’ categorizations were resolved via discussion.
In the HA group, 10 participants stated that they adopted a similarity rule, whereby they expected that stimuli perceptually similar to CS+ would be more likely to predict shock, whereas those perceptually dissimilar to CS+ would be less likely to predict shock (Similarity subgroup). Eight participants reported inferring a linear rule that if the dot was more to the right, the more likely the stimulus would predict shock (Linear subgroup). The remaining 22 participants reported not identifying a clear rule (No rule subgroup). In the LA group, 17 participants were categorized into the Similarity subgroup, 4 participants reported a linear rule and the remaining 19 participants reported not identifying a clear rule (see Table 2 for examples of the actual responses in the questionnaires and Table 3 for an overview of the number of participants in each rule subgroup). Across the linear rule subgroups, no participant reported the alternative left-based linear rule, that is, the more the dot was to the left, the more likely shock occured. The number of participants in the HA group who came up with rules did not substantially differ from those in the LA group (χ2 [1] = 0.45, p=0.50, n.s.).
Statistical analysis was first conducted to characterize differences in the generalisation gradients between rule subgroups, collapsed over anxiety (see Fig. 10). Subgroups were then compared across HA and LA participants to examine whether the effects of the anxiety factor differed between rule subgroups (Fig. 11).
Comparison between rule subgroups
Figure 10 shows the expectancy and SCL gradients in each rule subgroup. The expectancy gradient in the Linear subgroup showed a stronger positive linear trend (higher responding to stimuli on the right) than in the Similarity subgroup, leading to a significant 91
Table 2. Examples of actual responses in questionnaires
Rule subgroups Examples
Similarity ‘I expected the shock most when the black dot was in the middle. So I followed a strategy were the further away the black dot was from the middle, the less likely I would be shocked.’
Linear ‘I just had the idea that maybe when the dots were towards the left, there was going to be no shock and when it was towards the right there was a high possibility of shock.’
No rule ‘Honestly I could not deduce any pattern or correlation between shocks and the position of the dot in the square. Hence I played safe and just went with 50/50 on all displays.’
Table 3. Number of participants in each rule subgroup
Rule subgroup/ Similarity Linear No rule Anxiety group
HA 10 8 22
LA 17 4 19
interaction between linear trend and the Linear vs Similarity comparison, F(1,37) = 41.4,
2 p<0.01, ηp = 0.53. Conversely, the expectancy gradient in the Similarity subgroup was more peaked than that in the Linear subgroup, leading to an interaction with quadratic trend,
2 F(1,37) = 65.0, p<0.01, ηp = 0.64. A similar pattern was observed in the skin conductance data, with the Similarity subgroup having a more peaked gradient than the Linear subgroup, confirmed by a significant interaction between quadratic trend and the Linear vs Similarity 92
Figure 10. Mean shock expectancy ratings (Top panel) and skin conductance level (Bottom panel) during test phases for each rule subgroup, collapsed across anxiety groups.
93
2 comparison, F(1,37) = 8.3, p<0.01, ηp = 0.18. However the interaction with linear trend did not quite reach significance, F(1,37) = 3.9, p=0.056, n.s.
Both rule subgroups were compared to the No rule subgroup. The expectancy gradient in the Similarity subgroup was more peaked than that in the No rule subgroup, F(1,66) = 89.3,
2 p<0.01, ηp = 0.58. A similar interaction was observed in the skin conductance data, F(1,66) =
2 6.1, p<0.02, ηp = 0.08. No other interaction effect was observed for these two subgroups
(highest F = 0.1, p=0.75). The Linear subgroup had a stronger positive linear expectancy gradient than the No rule subgroup, confirmed by a significant interaction between linear
2 trend and the Linear vs No rule comparison, F(1,51) = 53.0, p<0.01, ηp = 0.51. This
2 interaction was also observed in the skin conductance data, F(1,51) = 5.7, p=0.02, ηp = 0.11.
No other interactions were significant (highest F = 2.8, p=0.10).
Comparison between High and Low anxious participants as a function of rules
The initial group comparisons indicated that there was broader generalisation in HA participants compared to LA participants, supported by significant interactions in the case of the expectancy measure. The rule subgroup analyses in turn suggested that generalisation was heavily modulated by the rules participants induced. Accordingly, an overall analysis was carried out to examine the interaction between cognitive mediation and trait anxiety on fear generalisation, by comparing high and low anxious participants within the different rule subgroups.
Figure 11 shows the comparison of generalisation gradients between HA and LA participants within each rule subgroup. For the expectancy data, a significant interaction was found between the anxiety groups and the contrast comparing the Similarity and the No rule
2 subgroups, averaged across the 9 test stimuli, F(1,74) = 4.6, p=0.04, ηp = 0.06, suggesting that the trait anxiety difference was greater in the No rule subgroup than in the Similarity subgroup. Although the interaction between the anxiety groups and the contrast comparing the 94
Figure 11. Comparison of the High Anxious (HA) and Low Anxious (LA) participants within each rule subgroup for expectancy (Left panel) and skin conductance (Right panel). Top row: Similarity subgroup; Middle row: Linear subgroup; Bottom row: No rule subgroup. 95
Linear and No rule subgroups did not quite reach significance, F(1,74) = 2.9, p=0.09, n.s., there was a significant 3-way interaction with the contrast comparing CS+ with the other 8
2 test stimuli, F(1,74) = 5.7, p=0.02, ηp = 0.07. This suggested that the HA and LA groups showed similar ratings to the CS+, but the HA groups showed a higher degree of fear generalisation in the No rule subgroup but not in the Linear subgroup. No significant interactions were found between anxiety groups and the contrast comparing the Similarity and the Linear subgroups averaged across the 9 test stimuli, F(1,74) = 0.01, p=0.9, n.s., nor with the contrast comparing CS+ with the other 8 test stimuli, F(1,74) = 1.9, p=0.2, n.s. No significant interactions were observed in the skin conductance data (highest F = 1.4, p=0.2).
Overall, the above analysis suggested that when comparing participants who came up with relational rules (Similarity and Linear), there was no evidence of trait anxiety on fear generalisation. However, when comparing participants who failed to identify any rules to those who had come up with relational rules (Similarity or Linear), a significant trait anxiety effect was observed. Therefore, the following analyses focused on verifying these effects by examining trait anxiety effects in each rule subgroup.
The effect of trait anxiety within individual rule subgroups
Figure 11A-D show the gradients for both measures in the Similarity and Linear rule subgroups. For these two rule subgroups, there were no significant interactions between the contrast comparing HA with LA participants and any of the stimulus contrasts (highest F =
3.1, p=0.09). That is, there was no evidence that trait anxiety had any effect on fear generalisation in the two rule subgroups (Similarity and Linear). In fact, HA and LA participants showed very similar gradients on both measures. These analyses further suggest that trait anxiety had little effect on the degree of fear generalisation when participants came up with a relational rule, regardless of whether it was similarity-based or linear.
96
No rule subgroups
Figure 11E and F show the expectancy and skin conductance gradients respectively for
HA and LA participants in the No rule subgroup. The HA-LA differences here are similar to those seen in the overall analysis involving all participants, but more pronounced. For the expectancy data, the HA participants gave higher overall ratings than the LA participants,
2 resulting in a main effect for anxiety group, F(1,39) = 14.2, p < 0.01, ηp = 0.27. However, as in the full group data, the difference was isolated to the test stimuli as the HA and LA groups gave very similar shock expectancy ratings to CS+. This pattern is supported by a significant
2 interaction between quadratic trend and anxiety group, F(1,39) = 4.4, p=0.04, ηp = 0.44. A significant interaction was also observed between anxiety groups and the comparison between
2 CS+ and the generalisation stimuli, F(1,39) = 6.0, p=0.02, ηp = 0.13, further confirming a specific pattern of greater generalisation in the HA group rather than an overall elevation in ratings to all stimuli including CS+. No other effects reached significance on the expectancy measure (highest F = 0.04, p=0.84). Although the skin conductance measure showed a somewhat similar pattern to expectancy, no interactions involving trait anxiety group reached significance (highest F = 0.8, p=0.38).
Discussion
The current study aimed to investigate the effect of trait anxiety on fear generalisation, while increasing the level of ambiguity of the generalisation task by using a single-cue conditioning procedure. The study also took advantage of recent developments in the literature to examine whether explicit reasoning processes during generalisation task would interact with trait anxiety, which may in turn affect fear generalisation. Looking first at the results collapsed across anxiety groups, both expectancy and skin conductance showed a relatively broad but nonetheless peaked gradient, with the highest responding to CS+. 97
However, qualitatively different generalisation patterns were found when participants were categorized into subgroups according to their reported inferred rules. Consistent with previous generalisation studies in our lab (e.g., Lee et al., 2018; Wong & Lovibond, 2017), three major subgroups were found across anxiety groups: the Similarity subgroup, the Linear subgroup and the No rule subgroup. The Similarity subgroups showed a sharper generalisation gradient than the mean gradient across all participants. They showed highest expectancy of shock and physiological responses to CS+, and a gradual decrease in responding to stimuli further away from CS+ along the dimension. The Linear subgroups revealed a gradual increase in responding from stimulus A, resulting in the highest responding at stimulus I in the expectancy measure. The skin conductance gradient was somewhat similar to the expectancy ratings, with the highest fear responding to stimuli right of CS+. For those in the No rule subgroups, a relatively flat generalisation gradient was observed as participants had similar expectancy ratings on chance across all the novel test stimuli. This suggests that participants were merely guessing the shock predictivity of the novel stimuli. A similar flat generalisation pattern was also observed in the skin conductance measures, suggesting that participants had similar amount of fear to all the generalisation stimuli, which corresponded to participants’ shock expectancy beliefs.
Further examining the generalisation patterns in each rule subgroup revealed that trait anxiety has different effects on fear generalisation. For those who endorsed a clear rule, either similarity or linear, trait anxiety did not affect the shape of generalisation patterns.
Furthermore, high anxious individuals did not show over-generalisation of fear, as they gave similar overall shock expectancy ratings and skin conductance responses to the low anxious individuals. However, by contrast to the rule subgroups, high anxious individuals who reported not identifying a clear rule (No rule subgroup) showed higher expectancy ratings to all the test stimuli except CS+, suggesting that they were over-generalising their fear from 98
CS+ to the novel generalisation stimuli. In fact, the difference observed in the full group data was driven almost entirely by the anxiety difference in the No rule subgroups.
The following section will discuss the theoretical aspects of cognitive mediation in human fear generalisation and the findings of over-generalisation of fear among high anxious individuals who failed to endorse any rules.
Associative and cognitive accounts
In the full group data collapsing across anxiety groups, a similarity-based gradient was observed in both expectancy and skin conductance measures. These peaked gradients were consistent with the prediction of associative accounts. According to such accounts, the fact that the highest responding fell at CS+ reflects a strong associative link between CS+ and US
(Rescorla & Wagner, 1972), while the declining responses to stimuli towards the extreme ends of dimension reflects progressively lower associative strength as stimuli share fewer overlapping elements with CS+ (Blough, 1975; McLaren & Mackintosh, 2002). However, the associative accounts have difficulty explaining any gradients that are not similarity-based
(i.e., linear).
Instead, the findings of non-similarity based gradients suggest the operation of rules.
The formation of inferred rules is thought to involve inferential reasoning, which is a higher- level, controlled cognitive process that derives a conclusion from the available information
(Premack, 1995). In other words, different rules could be formed, depending on factors such as the amount of information available during training and the prior learning history of participants. Only around half of the participants were able to infer a relational rule in the current study, presumably because they were only exposed to one CS+ during training, and hence had relatively little information to identify a rule. Previous studies in our lab have shown that participants are more likely to infer a rule in a differential conditioning procedure than a single-cue conditioning procedure (Lee et al., 2018; Wong & Lovibond, 2017). This 99 finding also suggests that the single-cue conditioning procedure did successfully induce a sufficient level of ambiguity, reflected by the number of participants who successfully came up with an inferred rule. Recent evidence has suggested that the presence of ambiguous information takes up extra cognitive resources (Pushkarskaya, Liu, Smithson & Joseph, 2010;
Shou & Smithson, 2015), hence resulting in fewer participants coming up with rules.
Generalisation is similar to a reasoning process called inductive reasoning. Inductive reasoning is distinct from deductive reasoning, where a definitive conclusion can be made. In the case of inductive reasoning, the conclusions are uncertain (Johnson-Laird, 2000; Rips,
1999). Generalisation is similar to inductive reasoning, since it involves novel stimuli with unknown predictiveness of shock, where the learner needs to infer their shock predictiveness based on extrapolation from the limited evidence given (CS+). Since no single conclusion was required by the evidence, participants were free to generate a variety of different rules.
According to the inductive reasoning account (e.g., Murphy, 2002), participants formed different threat beliefs which determined how they responded to the novel generalisation stimuli, hence generating the distinct gradients observed in the current study.
Because both dual-system and propositional models allow for inferential reasoning, they are both competent with the current results. However, some of the findings were more consistent with propositional model than the dual-system models. Firstly, the distinctive gradients in different rule subgroups were consistent with the participants’ self-reported rules.
For instance, participants who reported responding based on perceptual similarity formed peaked gradients, while those who reported inferring a linear rule showed their highest responding to one end of the dimension. Even expectancy ratings to the novel test stimuli were approximately at chance level in the No rule subgroups when collapsed across anxiety groups, which was consistent with participants’ report of merely guessing the shock predictiveness of test stimuli. By contrast, dual-system learning models (e.g., Clark & Squire, 100
1998) do not strongly anticipate the congruence between reported rules and generalisation gradients, since they are attributed to different systems that are expected to run in parallel and independently. Secondly, most of the SCL gradients aligned with their corresponding reported rules and expectancy patterns. Though one may argue that the correspondence between the expectancy and skin conductance gradients in the Similarity subgroup could be the result of parallel but independent associatively-driven skin conductance responses and cognitively- driven expectancy ratings, the correspondence between both measurements in the Linear subgroup suggests otherwise. Some strict dual-system accounts (e.g., Clark & Squire, 1998) would predict a dissociation between expectancy ratings and skin conductance measurements whenever participants infer a non-similarity rule such as a linear rule. Instead, the skin conductance data in the Linear subgroup showed a higher level of responding to stimuli right of CS+, corresponding to the expectancy data, as predicted by the propositional model.
Similarly, the dual-system account would predict a peaked SCL gradient even when participants failed to identify any rules; however, participants in the No rule subgroups showed a relatively flat SCL generalisation pattern, which aligned with the expectancy data.
Although the congruence between self-reported rules and generalisation patterns does not contradict the predictions of the dual-system account, it diminishes the potential role of a separate associative system. Furthermore, the alignment between skin conductance gradients and reported rules provides positive evidence for the propositional account. As none of the current findings require the postulation of a separate associative system, parsimony favors the propositional model.
Trait anxiety and ambiguity
The current results showed that trait anxiety had no effect on fear generalisation when a clear rule was identified, but induced excessive generalisation of fear when no rule could be inferred. This suggests that trait anxiety has no effect when the situation is unambiguous, but 101 leads to over-generalisation of fear when ambiguity is perceived in the situation. From an associative point of view, the process of generalisation may be unambiguous, since conditioned response is suggested to be activated in an automatic sense and responding to the novel stimuli merely depends on their perceptual similarity to CS+. On the other hand, the process of generalisation is ambiguous from the propositional point of view, because other higher-order level factors may affect the perceived threat value of the novel stimuli, for example, participants’ personal belief of threat or their subjective ignorance of the experimental configuration (see Chumbley et al., 2012). One possible way to disambiguate the situation is to use one’s prior experience to infer the probability of getting shocked when the novel stimuli were presented. In this case, relational rules were constructed based on prior experience with CS+. The threat value of the novel stimuli depends on the dot location relative to CS+. The constructed relational rules created a temporary schema to anchor threat assessment, guiding the generalisation process since participants could accurately judge the threat value of various stimuli based on their inferred rules. In contrast, not being able to identify a clear relational rule may have helped maintain the level of ambiguity of the generalisation task, since participants had no basis to infer the threat value of the generalisation stimuli. Considering that trait anxious individuals are suggested to have a cognitive schema that is biased to threat (Beck & Clark, 1988; Beck & Emery, 1985), therefore they showed elevated threat appraisal to the ambiguous novel test stimuli, resulting in over-generalisation.
The current finding is consistent with past research that showed high anxious individuals displaying expectancy bias to ambiguous threat. For instance, trait anxious participants who were not aware of the CS-US contingency showed heightened shock expectancy to both the threat and safety cues in a conditioned inhibition procedure (Chan &
Lovibond, 1996). Since the threat value of the cues was unknown to the unaware participants, all cues in the task effectively became ambiguous. Similarly, Boddez et al. (2012) found a 102 positive correlation between trait anxiety and shock expectancy to the blocked stimulus. Since the blocked stimulus had never been presented by itself, its causal status of shock was unknown, arguably rendering it ambiguous. A more recent study has also found that individuals high in IU showed increased threat expectancy to cues that had ambiguous threat value (Chen & Lovibond, 2016).
The current results are also consistent with findings in the broader cognitive literature that trait anxious individuals show interpretation biases to ambiguity (Calvo & Eysenck,
1995; MacLeod & Cohen, 1993). These studies typically showed ambiguous cues that had more than one meaning, usually one affectively neutral and one affectively negative, and trait anxious individuals were more likely to interpret them in a threatening way. In the current study, the novel test stimuli were ambiguous in a similar way: they either predicted an electric shock or not. The elevation in shock expectancy to these ambiguous stimuli among trait anxious individuals is consistent with the idea of negative interpretation to ambiguity found in anxious individuals. Furthermore, the current results are consistent with threat estimation biases among trait anxious individuals (e.g., Butler & Mathews, 1987). Overestimating the probability of a negative event is parallel to the increased shock expectancy to novel stimuli among trait anxious individuals in the current study.
Alternative explanations for past studies
The current results also provide a potential explanation for previous studies
(Arnaudova et al., 2017; Torrents-Rodas et al., 2013) that failed to find any trait anxiety effect in fear generalisation. These studies used a differential conditioning paradigm with the CS+ and CS- located at the extreme ends of the stimulus dimension, with all the generalisation stimuli situated between the two CSs. As argued by Lissek et al. (2006) and Beckers et al.
(2013), a ‘strong’ situation’ is less likely to reveal any individual differences in fear learning.
A differential conditioning paradigm arguably provides a ‘strong situation’ which is relatively 103 unambiguous, since it provides extra information (i.e., CS-) to guide generalisation (see Homa et al., 1981). Our previous work also suggested that the differential conditioning precedure is more likely to induce rule formation, compared to the single-cue conditioning procedure (Lee et al., 2018; Wong & Lovibond, 2017). Furthermore, the studies by Torrent-Rodas et al.
(2013) and Arnaudova et al. (2017) also used intensity stimulus dimension. As reviewed before, intensity stimulus dimensions may further encourage participants to respond accordingly to the intensity of stimulus, forming an intensity rule. Since the CSs were located at the extreme end of the dimension, an intensity rule would be similar to the Linear rule that was found in the current study. The combination of using a differential conditioning procedure and an intensity dimension may have strongly encouraged participants to identify a clear rule. Therefore, it is possible that most participants in the aforementioned studies
(Arnaudova et al., 2017; Torrents-Rodas et al., 2013) were able to come up with a rule, disambiguating the generalisation task and hence attenuating any effects of trait anxiety.
Alternative mechanism
An alternative explanation for the apparent over-generalisation of fear in the No rule subgroup among trait anxious individuals would be that these individuals failed to inhibit their fear responses (Davis et al., 2000; Gazendam et al., 2013). However, this explanation would also predict higher responding to the CS+ among high anxious individuals, which was not observed in the current results. Given the CS+ was partially reinforced, the null anxiety difference in responding to CS+ could not be explained by the presence of a ceiling effect in responding. Furthermore, if the results obtained were due to failure of fear inhibition, high anxious individuals would have shown higher level of responding to all stimuli across all the rule subgroups, which was not observed in the current study. Hence, the current results favor the interpretation of over-generalisation of fear in trait anxious individuals in the presence of ambiguity, rather than a failure to inhibit fear responses. 104
Processing efficiency theory
The processing efficiency theory (Eysenck & Calvo, 1992) suggests that the working memory in trait anxious individuals would be impaired with the induction of anxiety, since they are hypothesized to be constantly worrying about potential threat and using extra cognitive resources to reduce the amount of state anxiety. However, the current findings did not show any anxiety differences in rule formation. High anxious individuals were as likely as low anxious individuals to come up with relational rules, and generalised accordingly to their inferred rules. One possible explanation for this pattern is that participants may have inferred their rules during Test1, where the shock electrodes were disconnected. Therefore, without having to worry about imminent shock, trait anxious individuals’ working memory capacity may be freed and hence leaving more cognitive resources for rule formation. However, it is possible that participants may have come up with a rule during acquisition when the shock electrodes were still connected. Given that the explicit assessment of rule-based generalisation in humans is a relatively new line of research, future studies can aim to investigate when rules are inferred.
Limitations
One limitation was that the trait anxiety effect on fear generalisation was only significant in the expectancy data, but not in skin conductance. One reason for this may be the large inter-individual variability of skin conductance (Lykken & Venables, 1971).
Furthermore, in order to minimize the extinction of learning during test, skin conductance was measured only once for each selected generalisation stimulus during test. This single-trial measurement inevitably induces more variability in the skin conductance measure (Vervoort,
Vervliet, Bennett & Baeyens, 2014). However, even though skin conductance measure is characterized by high variability, the current study still found a high correspondence between the inferred rules and the corresponding SCL gradients. It was also found that the SCL 105 gradients in each rule subgroup were significantly distinctive from each other. Thus it appears that skin conductance may be sensitive enough to pick up inter-rule-subgroup differences, but not a trait anxiety effect within the same subgroup.
Chapter summary
This empirical chapter aimed to investigate the effect of trait anxiety on fear generalisation across a perceptual stimulus continuum. In order to induce ambiguity, a single- cue conditioning procedure was used to create a ‘weak situation’ (Beckers et al., 2013; Lissek et al., 2006). Furthermore, taking advantage of the recent development in the rule-based generalisation literature, participants were categorized into different subgroups according to their reported inferred rules, so that the interaction between trait anxiety and rule formation on fear generalisation could be analyzed. The results were consistent with previous findings in our lab (Ahmed & Lovibond, 2018; Lee et al., 2018; Wong & Lovibond, 2017), in which generalisation patterns were highly consistent with participants’ reported rules. The high correspondence between reported rules and the gradients in both measures also favors the propositional model. Interestingly, trait anxious individuals showed a similar generalisation pattern to their low anxious counterpart when a clear rule was inferred and endorsed, but showed more fear generalisation when no clear rules could be identified. Since not being able to come up with a rule arguably maintains the ambiguity of the generalisation task, the current results suggest that over-generalisation of fear may be a special case of the more general principle that trait anxiety is associated with excessive threat appraisal under conditions of ambiguity.
106
Chapter 6
The effect of trait anxiety on fear generalisation to conceptually
related objects
Wong, A. H. K., & Lovibond, P. F. (submitted). Breakfast or bakery? The role of categorical ambiguity in over-generalization of learned fear in anxiety. 107
The past chapters have emphasized generalisation on a stimulus continuum with stimuli that differ perceptually on a quantitative level. The advantage of using a perceptual stimulus dimension is that similarity between stimuli can be easily manipulated. This allows the analysis of the generalisation pattern in a quantitative way and allows the prediction of generalisation as a function of perceptual similarity. However, the emphasis on perceptual generalisation in the past decades may have distracted researchers from the fact that objects that lack of any physical resemblance to the original threat cue can also trigger fear. For instance, fear memories of a traumatic event in PTSD can be triggered by various reminder stimuli that do not necessarily resemble the traumatic event in a physical way (Parsons &
Ressler, 2013). A child who was abused by a teacher may show fear of other authority figures
(e.g., police) even though they do not physically resemble the original perpetrator. This suggests that cues that are related to the threat cues or traumatic experience on a conceptual level are able to trigger fear. This suggests that, unsurprisingly, humans are capable of generalising between stimuli that are abstractly and conceptually related, presumably via a higher-order cognitive process. In addition, the findings of rule-based generalisation along an arbitrary stimulus dimension in the previous chapter and in our lab (Ahmed & Lovibond,
2018; Lee et al., 2018; Wong & Lovibond, 2017) suggest the importance of cognitive processes in human generalisation. There are also some studies in the literature that have examined how different conceptual relationships between stimuli affect generalisation in humans, and one of the earliest areas of study is semantic generalisation.
Semantic generalisation
Semantic generalisation refers to generalisation that occurs through the linguistic meaning of stimuli. Using a salivary conditioning procedure, Razran (1939) first conditioned participants to salivate when certain words were presented (e.g., style, urn). He then presented participants with synonyms (e.g., fashion, vase) and homophones (e.g., stile, earn) of the CS+, 108 and measured the magnitude of CRs (i.e., amount of saliva produced) during the presentation of these words. He observed that conditioned salivation was transferred to both synonyms and homophones, with more conditioned responses to synonyms than to homophones. Similarly,
Riess (1940) attempted to replicate Razran’s (1939) finding via a fear conditioning procedure.
The same words used in Razran’s (1939) study (e.g., style) were paired with an electric shock, and the same generalisation stimuli (i.e., fashion, stile) were presented during test.
Conditioned skin conductance responses were again found to spread to both synonyms and homophones, with responding to synonyms significantly greater than those to homophones.
The authors of both studies claimed that the homophones served as a ‘visual-auditory’ form of the CS+, and hence were ‘perceptually’ similar to CS+. These results suggest that conditioned responses are more readily to be generalised via conceptual knowledge of semantic meanings rather than perceptual similarity.
Interestingly, Branca (1957) conditioned participants with the word ‘hat’, and presented a novel picture of a hat during test. Participants who were aware of the CS-US contingency showed generalisation of conditioned skin conductance responses to the test picture. This finding further supports the notion that fear can be generalised to objects that belong to different dimensions (words and pictures) but are nonetheless conceptually related.
Furthermore, Lang, Geer and Hnatiow (1963) attempted to obtain a semantic gradient of generalisation. Participants were first conditioned to words with high hostility (e.g., torture, mutilating), and their skin conductance responses to words with medium hostility (e.g., conflict, scolding), low hostility (e.g., displeasure, blunt) and neutral words (e.g., nameless, recommend) were measured. Skin conductance responses generalised to words with varying level of hostility but not to neutral words, again suggesting semantic generalisation took place. However, responding fell off sharply from CS+ (high hostile words) to medium hostile words, while no differentiation in responding was observed between medium hostile and low hostile words, rendering it difficult to establish a gradient of semantic generalisation. During 109 the 1960s and the 1970s, semantic generalisation was extensively studied (e.g., Brotsky &
Keller, 1971; Maltzman, Langdon, Pendery & Wolff, 1977; Peastral, 1964; Staats, Staats &
Crawford, 1962). The general phenomenon of semantic generalisation was often replicated, although the results tended to be variable (see Feather, 1965; Maltzman et al., 1977).
Since the finding that fear can be generalised to stimuli outside of a physical stimulus continuum cannot be easily explained by the associative accounts, Maltzman, Langdon and
Feeney (1970) proposed that some complex thought processes may have be involved, and these processes may mediate the autonomic responses during generalisation. The fact that successful semantic generalisation was often restricted to participants who were aware of the
CS-US contingency (see Maltzman et al., 1977) further suggests the involvement of a higher- order process in human generalisation. However, the theoretical focus in semantic generalisation then shifted into language acquisition (see Goldberg, 2006). It was not until recently that research interests have begun to focus again on the role of higher-order processes in human fear generalisation. Using a semantic generalisation procedure, Boyle, Roche,
Dymond and Hermans (2016) paired a word with shock (e.g., broth; CS+) and another word with no shock (e.g., assist; CS-). During test, novel words semantically related to the CS+
(e.g., soup) and CS- (e.g., help) were presented. It was found that differential responding to
CS+ and CS- had also been transferred to the novel words. In other words, fear selectively generalised to the word that was semantically related to CS+ (e.g., soup) but not to the word semantically related to CS- (e.g., help).
Symbolic generalisation
Another line of research that deviates from semantic generalisation, but nonetheless addresses how a verbal relation between stimuli affects generalisation is called symbolic generalisation. Symbolic generalisation studies often involve the use of arbitrary stimuli, and the categorization of these stimuli into artificial categories via a procedure called a matching- 110 to-sample (MTS) task (Sidman, 1971). In a typical MTS task, a stimulus (A1) is usually presented on the screen with a set of comparison stimuli ([B1, B2, B3] or [C1, C2, C3]).
Participants are required to select one stimulus from the comparison stimulus. Selecting B1 or
C1 in the presence of A1 is reinforced by providing corrective feedback. In other words, participants will learn the artificial relation between A1 and B1, and A1 and C1. Since A1 is directly paired with B1 and C1, the A1-B1 and A1-C1 relations are said to be ‘symmetrical’.
Although B1 and C1 have not been explicitly paired, their mutual symmetrical relations with
A1 render them into the same category, and they are said to have an ‘equivalent’ relation
(Sidman, 1994). Similarly, another artificial category can be created by directly pairing another stimulus A2 with B2 and C2. After the MTS task, Dougher, Augustson, Markham,
Greenway and Wulfert (1994) paired B1 with an electric shock and B2 with no shock. In the subsequent test phase, participants showed increased skin conductance responses to B1 and
C1 but not to B2 and C2. That is, fear was generalised to stimulus equivalently related (C1) to
CS+ (B1), but not to stimulus related to CS- (i.e., C2). Using a larger sample size and a similar MTS paradigm, a series of studies have successfully replicated the finding of selective fear generalisation to novel stimuli that were conceptually related to CS+ (Bennett et al.,
2015a, 2015b; Vervoort et al., 2014). Similar findings were observed in fear avoidance studies, in which participants showed more fear avoidance to stimuli symbolically related to
CS+ but not to those related to CS- (e.g., Augustson & Dougher, 1997; Dymond et al., 2011).
In symbolic generalisation studies, the symmetrical or equivalent relations between stimuli effectively assign them into the same artificial category. By the same token, real-life objects that have already established their categorical membership via prior learning experiences have also been found to promote fear generalisation.
Category-based induction and its role in fear generalisation 111
In the inductive reasoning literature, categorical generalisation is a form of category- based induction, which is usually examined by schematic arguments (Rips, 1975). A list of statements are presented, of which the last one is called the conclusion and the others are called premises. The premises contain information that is relevant to but does not demand that conclusion. The magnitude of inductive generalisation is examined by asking participants to rate the argument strength of the conclusion, given the premises (see Osherson, Smith,
Wilkie, Lopez & Shafir, 1990), as listed below:
Horses have sesamoid bones [premise]
Rats have sesamoid bones [conclusion] (1)
Sparrows have sesamoid bones [premise]
Rats have sesamoid bones [conclusion] (2)
In this case, the strength of the argument is determined by how much the premise tend to lead one to believe the conclusion, or in other words, how one would generalise the knowledge gained from the premise to the conclusion. Typically, the strength of the first argument (1) is found to be stronger than the second argument (2), as horses and rats belong to the same category (i.e., mammal) while rats and sparrows belong to different categories (i.e., mammal and bird respectively; see Heit, 2000) Similarly, consider the following arguments:
Horses secrete uric acid crystal [premise]
Dogs secrete uric acid crystal [conclusion] (3)
Rats secrete uric acid crystal [premise]
Horses secrete uric acid crystal [premise]
Dogs secrete uric acid crystal [conclusion] (4) 112
In this case, although the object in the conclusion belongs to the same category as those in the premise(s), the second argument (4) is perceived to be stronger than the first one (3; Heit,
2000). This phenomenon is called premise monotonicity (Carey, 1985). It refers to the principle that the more premises are available, the stronger the conclusion, since the extra premise(s) provide(s) more positive evidence to support the conclusion.
Although category-based induction has been extensively studied in the development
(Hayes & Thompson, 2007; Lopez, Gelman, Gutheil & Smith, 1992; Sloutsky & Fisher,
2004) and decision making (Hayes & Newell, 2009; Malt, Ross & Murphy, 1995; Murphy &
Ross, 2010) literatures, it was not until recently that the similarity between inductive reasoning and generalisation has been exploited. Researchers started to examine category- based induction in associative learning paradigms. Dunsmoor, Martin and LaBar (2012) presented exemplars from two categories, animals and tools. Critically, each exemplar was only presented once, rendering the presentation of each exemplar unique. Half of the exemplars from one category (e.g., animal) were reinforced by electric shock (i.e., 50% reinforcement rate), while exemplars from the other category (e.g., tool) were never reinforced. Overall, participants developed a significant differentiation in responding to exemplars between the two categories. That is, they showed increased shock expectancy ratings and skin conductance responses only to exemplars that belong to the CS+ category
(e.g., animal) but not to those belonging to the CS- category (e.g., tool). The results strongly suggest that fear generalisation was operating on a categorical level, for two reasons. First, if participants were learning the CS-US contingency on the exemplar level (e.g., cow predicts shock, hammer predicts no shock), no learning would have been occurred given that each exemplar was only presented once. Secondly, shock expectancy ratings to both categories differentiated in early trials, suggesting participants quickly learnt that the categorical membership of the exemplars determined the predictiveness of shock. 113
Empirical evidence like this suggest that generalisation in associative learning is a form of inductive reasoning. In fact, the conclusion in inductive reasoning is uncertain, and the strength of the argument depends on how participants extrapolate from the evidence gathered from the premises. As mentioned in Chapter 5, this is analogous to stimulus generalisation, where participants need to extrapolate the knowledge they gained from the
CSs (premises) to determine how likely the novel GSs predict an outcome (conclusion).
Similarly, Meulders, Vandael and Vlaeyen (2017) presented exemplars of two action categories: opening boxes and closing boxes. Exemplars from one category (e.g., opening boxes) were partially reinforced with shock (CS+), while exemplars of another category
(closing boxes) were not followed by any shock (CS-). Participants showed more generalisation of pain-related fear to novel exemplars that belonged to the CS+ category but not to the CS- category, again demonstrating generalisation is a similar process to inductive reasoning.
If stimulus generalisation in humans derives from the same mechanisms as inductive reasoning, factors that affect inductive reasoning may affect fear generalisation in a similar way. One of these factors is called stimulus typicality. Consider the following arguments:
Horses use serotonin as a neurotransmitter [premise]
Rats use serotonin as a neurotransmitter [conclusion] (5)
Bats use serotonin as a neurotransmitter [premise]
Rats use serotonin as a neurotransmitter [conclusion] (6)
The first argument (5) is perceived to be stronger than the second one (6) though they involve the same category (i.e., mammal). This is because the premise in the first argument uses an object that is typical in its category; that is, the object (horse) is highly representative of its category. In contrast, the premise in the second argument contains an atypical object (bat), 114 that is less representative of its category. Typical stimuli are found to generate more inductive strength than atypical stimuli, a phenomenon called typicality asymmetry (Osherson et al.,
1990). In order to examine whether typicality asymmetry would be observed in human fear generalisation, Dunsmoor and Murphy (2014) assigned participants to two groups: the typical group and the atypical group. The typical group was presented with typical mammal pictures
(e.g., bear, rabbit), that were partially reinforced by a shock, while the atypical group was trained with atypical mammal pictures (e.g., armadillo, otter). Both groups received bird pictures of intermediate typicality (e.g., duck, dove) as CS-. In test, novel pictures of mammals and birds of intermediate typicality were presented and participants’ shock expectancy and skin conductance responses were measured. Both groups showed similar acquisition to exemplars of the CS+ category. Surprisingly, although no group differences were observed in expectancy ratings to the test stimuli, the typical group showed more skin conductance responses to the test stimuli. In other words, the group trained with typical stimuli showed more fear generalisation than those trained with atypical stimuli, similar to the findings of typicality asymmetry in the inductive reasoning literature. This finding further suggests that fear generalisation in humans potentially involves the same processes as those underlying inductive reasoning.
Given that humans are capable of generalising beyond perceptual features, conceptual- or category-based fear generalisation may help explain why patients suffering from anxiety disorders show fear to a wide range of stimuli (see Cahill & Foa, 2007; Foa, Steketee &
Rothbaum, 1989). Indeed, there has been a call for more research on the processes underlying higher-order fear generalisation recently, given the importance of its implications for the understanding of anxiety disorders and development of more effective treatments (Dymond et al., 2015). Accordingly, the following empirical chapter aimed to study categorical fear generalisation, and examined whether trait anxiety has any effect on categorical generalisation. Given that the empirical study in the previous chapter suggests that trait 115 anxiety is associated with over-generalisation of fear only when the level of perceived ambiguity is high (i.e., No rule subgroup), the present study also aimed to manipulate the ambiguity level of category membership.
Manipulation of ambiguity
As discussed in the previous chapter, ambiguity in a laboratory task can arise from different sources. First, being unaware of the CS-US contingency adequately makes all stimuli ambiguous, since participants have no way to determine which stimulus predict a threatening outcome (Chan & Lovibond, 1996). Secondly, similar to the first notion, not being able to infer the causal status of a cue also renders it ambiguous (Boddez et al., 2012; Chen &
Lovibond, 2016). Thirdly, interpretation of a cue that has more than one meaning, usually one affectively negative and one affectively neutral, is ambiguous (e.g., MacLeod, 1990;
MacLeod & Cohen, 1993). The current study aimed to explore a variation of the third strategy, based on the idea of ‘cross-classification’ from the categorical induction literature.
According to this literature, most objects belong to more than one category. For instance, Bill
Gates is a businessman, an American and a man, while a lion is a mammal, a predator and a carnivore. These objects are often termed cross-classified items since they can be classified into multiple categories (Smith & Medin, 1981). When these objects can be simultaneously fitted in categories that have conflicting properties, the categorization process often becomes ambiguous (Hayes, Kurniawan & Newell, 2011; Murphy & Ross, 1999). For example, after discovering a skin blemish on a patients’ back, a dermatologist has to decide (or categorize) whether it is just a harmless sun-spot or a sign of skin cancer. Cross-classified items can usually be categorized into both taxonomic and script categories simultaneously. Taxonomic categorization classifies items based on their shared features or properties. It provides a hierarchical structure, where items at the highest level share the most basic features. Further down the hierarchy, objects share the features from the superordinate level in addition to their 116 own distinguishing features, allowing more specific categorization (Rosch, Mervis, Gray,
Johnson & Boyes-Braem, 1976). On the other hand, script categorization derives from event schemas (Nelson & Nelson, 1990). In other words, items that are categorized based on their roles in a certain event or time, for instance, lunch food or party food.
Therefore, the current study aimed to investigate two research questions. The first research aim was to examine fear generalisation on a categorical level, by pairing exemplars from one category with electric shock, and exemplars from another category with no shock.
The test phase examined whether fear selectively generalises to novel exemplars that belong to the threat category. The second research aim was to study whether trait anxious individuals would show excessive threat appraisal to cues whose categorization was ambiguous. The ambiguous cues were cross-classified items that fitted into both threat and safety categories simultaneously. Food items were used in this study, for two main reasons. First, while most objects can be easily classified into taxonomic categories but not script categories (e.g., animals), food items can be readily classified into both categories (Murphy & Ross, 1999).
Secondly, participants were able to spontaneously cross-classify the same food item (e.g., bagel) into taxonomic (e.g., grain) and script (e.g., breakfast) categories, presumably because we have extensive experience and knowledge of food items as they are accessed many times a day (e.g., Rozin, Dow, Moscovitch & Rajaram, 1998). The food items used in the current study were exemplars from the breakfast and bakery categories for two major reasons. First, according to a national survey (Williams, 2002), certain bakery items like bread are commonly consumed as breakfast in Australia. This effectively enables some bread-breakfast items readily to be cross-classified by an Australian undergraduate sample. Secondly, the breakfast category was chosen because of its relatively low diversity. Other script categories like lunch have a relatively high diversity, that is, many different foods can be classified as lunch. Having a higher diversity runs the risk of exemplars overlapping with other script 117 categories, for example, pasta can be classified as lunch, but also dinner. Having a low categorical diversity encourages participants to classify them only into the desired categories.
Although past studies (e.g., Murphy & Ross, 1999) have already suggested foods that are readily able to be classified as breakfast (e.g., oatmeal), bakery (e.g., pretzels) and both
(e.g., bagel), a questionnaire-based study was run to select the breakfast and bakery items that would be used in the conditioning task. The questionnaires aimed to validate the items based on three criteria: first, to be certain that participants clearly recognised the items being shown; secondly, to ensure that certain items clearly belong to only one category in order to facilitate the learning of categorical membership; finally, to ensure that cross-classified items were perceived to belong to both categories.
Validation Questionnaires
Method
Participants
First year psychology undergraduates were recruited in return for course credit. A total of 51 participants were recruited (35 females, mean age = 18.6, SD = 1.1). Participants were randomly allocated into the Typicality rating group or the Free categorization group, resulting in 26 participants in the former group and 25 in the latter group.
Apparatus and stimulus materials
Participants were tested individually in an experimental room. A 64-cm Dell® LCD computer monitor was used to present the experimental instructions, pictures found via the
Internet and questionnaires. Pictures that either belong to the breakfast category (e.g., bacon, cereal) or bakery category (e.g., baguette, maple Danish), or both categories (e.g., bagel, croissant) were included (see Table 4). Food items not belonging to either category (e.g., 118 steak, spaghetti), and non-food items (e.g., aeroplane) were also included as a validity check
(see Appendix B for the full set of pictures). All 32 pictures shown were 10cm x 8cm in size.
Procedure
In the Typicality rating group, all 32 pictures were shown on the computer screen via
Matlab. Each picture was showed on the center of the screen on a white background, with a question asking ‘How typical is the picture shown of the category BREAKFAST?’, located above the picture. A Likert Scale from 1 to 7 was located at the bottom of the screen, with 1 labelled as ‘Not at all typical’ and 7 labelled as ‘Highly typical’. All pictures were shown for
Table 4. Names of the pictures presented in the Typicality rating and Free categorization groups Intended category Name of pictures
Intended breakfast items Bacon Baked beans Boiled egg Cereal Cornflake Hashbrown Oatmeal Pan fried egg Pancake Sausage Scramble egg Waffles Intended bakery items Apple & walnut log Apple pie Baguette Ciabatta roll Cupcake Custard tart Fingerbun Garlic bread Hamburger bun Hotcross bun Jam bunlet Pretzel Intended cross-classified items Bagel Croissant English muffin Toast Validity check items: Spaghetti Steak Food items (neither breakfast nor bakery) Validity check items: Aeroplane Spongebob® Non-food items 119 a second time with a different question asking ‘How typical is the picture shown of the category BAKERY?’. Participants were required to use the mouse to select the typicality ratings on the scale, and pressed Spacebar once they had finished rating for that trial. There was a total of 64 trials, presented in a pseudo-randomized order so that identical pictures would not appear consecutively. For each trial, the picture, question and scale remained on the screen until participants have made the typicality ratings, with an ITI of 2s.
In the Free categorization group, participants were given a Word document on the screen. The same set of pictures were used, with the addition of a picture of apple and spaghetti for practice. For each picture, participants were asked to type in the first 3 categories that came to mind that the picture might fit in. Participants were also asked to type in the name of the picture shown. If the name of the item was not known (e.g., maple danish), participants were asked to describe the item as much detail as they could. Before participants were shown the test items, they were asked to respond to the practice items: apple and spaghetti. The presentation order of the pictures was counterbalanced, in order to minimize the impact of any practice effect, especially to the pictures that were shown last.
Scoring and analysis
For the Typicality rating group, typicality ratings for each item were calculated. For the intended breakfast items, the 6 items that had the highest typicality ratings for the breakfast category were included for analysis. Similarly, the 6 intended bakery pictures that had the highest typicality ratings for the bakery category were included for analysis. For the intended cross-classified items, the 3 items that had the highest typicality ratings averaged across both categories were included. The typicality ratings for the breakfast and bakery categories were then compared across all selected items, and the averaged typicality ratings to both intended breakfast and bakery items were compared with those to the intended cross- 120 classified items. Finally, the critical interaction examined whether the breakfast and bakery exemplars differed in their breakfast and bakery typicality ratings.
For the Free categorization group, the frequency that the picture was spontaneously recognized as a breakfast or a bakery item (or classified as both categories for the intended cross-classified items) was calculated for each picture.
Results
Exclusion of participants
For the Typicality rating group, 4 items that did not belong to either breakfast or bakery were included for a validity check. Participants having an averaged typicality ratings of 4 or above to these items were excluded. For the Free categorization group, participants had to acknowledge what the picture shown was in order to make their responses valid.
Therefore, if participants could not correctly name the items shown, or provided a wrong description for more than two items, their data were excluded.
Based on the above exclusion criteria, 1 participant in the Free categorization group was excluded.
Typicality rating group
The 6 items that had the highest typicality ratings for the breakfast category were oatmeal, cornflake, pan-fried eggs, pancakes, bacon and hashbrown (see Fig. 12). On the other hand, the 6 items that had the highest typicality ratings for bakery were hamburger bun, garlic bread, maple danish, finger buns, apple and walnut log and cupcake.
Averaged across all items, typicality ratings for breakfast category were significantly higher than that for bakery items, F(1,25) = 33.2, p<0.01, ηp2 = 0.57. The averaged typicality ratings for breakfast exemplars were higher than those for bakery exemplars, however this 121
A
Breakfast typicality ratings 7 Bakery typicality ratings
5
3 Typicality ratings
1
Oatmeal Cornflake Pan-fried Pancakes Bacon Hashbrown B eggs
7
5
3 Typicality ratings
1
Cupcake G arlic Hamburger M aple Finger A pple & C bread bun danish bun W alnut log
7
5
3 Typicality ratings
1
Toast Croissant English muffin
Figure 12. Breakfast and bakery typicality ratings for breakfast items (Top row), bakery items (Middle row) and cross-classified items (Bottom row).
122 difference did not quite reach significance, F(1,25) = 3.5, p=0.07, n.s. Breakfast typicality ratings to the breakfast exemplars were significantly higher than the bakery typicality ratings to the same exemplars, and an opposite pattern was observed in the ratings to the bakery exemplars, confirmed by a significant interaction, F(1,25) = 329.6, p<0.01, ηp2 = 0.93. A follow-up analysis revealed that the breakfast typicality ratings to the breakfast exemplars were significantly higher than the bakery typicality ratings across all 6 items, F(1,25) = 249.0, p<0.01, ηp2 = 0.91, suggesting that these items were highly typical breakfast items (Fig 12A).
Conversely, the bakery typicality ratings averaged across the intended bakery items (Fig.
12B) were significantly higher than the breakfast typicality ratings, F(1,25) = 221.8, p<0.01,
ηp2 = 0.90, suggesting that these items were highly typical bakery items.
The 3 intended cross-classified items that had the highest ratings averaged across breakfast and bakery categories were croissant, toast and English muffin (Fig. 12C). The difference between both typicality ratings to the intended cross-classified items did not quite reach significance, F(1,25) = 3.1, p=0.09, n.s., suggesting they were approximately equally typical in both breakfast and bakery categories. This notion was further supported by the observation that breakfast typicality ratings to the intended cross-classified items were similar to those to the breakfast items, while a similar pattern was observed between bakery typicality ratings to the intended cross-classified items and intended bakery items.
Free categorization group
Only the items included in the Typicality rating group analyses are reported here. For the intended breakfast items, 82% of participants wrote down breakfast as one of the three categories that came into their mind first, suggesting that majority of the participants were able to freely classify them as breakfast items, and none of them classified them as bakery items. Similarly, 78% of participants could freely classify the 6 intended bakery items as typical bakery exemplars, however, 4% of participants also classified them as breakfast items. 123
For the 3 intended cross-classified items, an average of 65% of participants were able to freely classify them as both breakfast and bakery items.
Discussion
The questionnaires sought to validate the typicality of breakfast and bakery items for their corresponding categories. While the intended breakfast items had significantly higher breakfast typicality ratings than the intended bakery items, the intended bakery items had significantly higher bakery typicality ratings. This suggests that the intended breakfast and bakery items are highly typical in their corresponding category. The intended cross-classified items had equally high typicality ratings in both categories, making them suitable candidates for creating categorical ambiguity. Furthermore, the majority of participants in the Free categorization group were able to freely name the categorical memberships of the corresponding items, complementing the findings in the Typicality rating group that both breakfast and bakery items shown were highly typical items in their corresponding category.
Furthermore, a large proportion of participants were able to freely classify the cross-classified items as belonging to both breakfast and bakery categories simultaneously.
Therefore, the breakfast and bakery items’ typicality in their corresponding category was validated. This process ensured that the presentation of a breakfast item would not strongly activate the representation of the bakery category, and vice versa. Furthermore, the cross-classified items were found to be highly typical in both breakfast and bakery categories.
Experiment 2
Using the validated stimuli, the current study sought to examine how fear selectively generalise to novel items that share the same category with the threat cue, by pairing exemplars from one category (e.g., breakfast) with electric shock and exemplars from another category (e.g., bakery) with no shock. A second experimental aim was to investigate whether 124 trait anxiety has any effect on fear generalisation, especially to ambiguous cross-classified cues that share categorical membership with both threat and safety cues.
Method
Participants
Undergraduates were recruited as participants who received course credit or AUD $15 for participation. Participants were pre-screened by the DASS-21 (Lovibond & Lovibond,
1995a). Those with a DASS-anxiety score of 4 or below were recruited to the low anxious
(LA) group, while those with a DASS-anxiety score of 16 or above were assigned to the high anxious (HA) group. Noted that the pre-screening criterion for HA group was lower compared to the previous experiment in Chapter 5 due to difficulty in recruitment. The DASS-21 was re-administered to participants at the time of fear conditioning, and only those who had a
DASS anxiety score consistent with pre-screening were included in the study. Thirty-five participants were recruited in each group, with 10 participants excluded (see Results for detail). The final sample comprised 60 participants (40 females) with a mean age of 19.8 years (SD = 4.5).
Apparatus and materials
Participants were tested individually in an experimental room. The experimental hardware (e.g., skin conductance equipment) and software (e.g., Matlab) were the same as previously.
The 6 breakfast items, 6 bakery items and 3 cross-classified items validated in the questionnaires were presented as stimuli. Two pictures (steak and spaghetti) used as manipulation check items in the previous validation study, that were neither breakfast nor bakery items, were also included. An additional 12 breakfast and 12 bakery pictures were 125 included (see Appendix B) for a categorical task prior to conditioning. All stimuli were 10cm x 8 cm in size and were presented in the center of a white background on the computer screen.
Procedure
After signing the consent form, participants were asked to fill in the DASS-21. Shock electrodes were attached to participants’ index finger, and they were then led through a work- up procedure in which they selected a level of shock that was ‘definitely uncomfortable but not painful’. Isotonic gel was squeezed into the GRASS® silver disc electrodes, to maximize the sensitivity of skin conductance measure, and then attached to participants’ fingers.
Participants were then taken into the experimental room. As shown in Table 5, the study consisted of a categorical task, followed by an acquisition phase and a test phase. Similar to previous studies in our lab (Lee et al., 2018; Wong & Lovibond, 2017), the test phase was divided into two stages. Before the experiment started, headphones were placed on participants. White noise was presented throughout the study for noise cancellation.
Table 5. Design of Current study
Phase Categorical task Acquisition Test1 Test2 12 breakfast (1) 4 CS+ (2) 1 CS+ (1) 1 GEN+ (1) 12 bakery (1) 4 CS- (2) 1 CS- (1) 1 GEN- (1) 2 GEN+ (1) 2 CC (1) 2 GEN- (1) 3 CC (1) 2 Filler (1) Note. Numbers without brackets indicate the number of different exemplars of that trial type; number with brackets indicate the number of trials of that trial type; CS+ indicates reinforced exemplars; CS- indicates non-reinforced exemplars; GEN+ indicates exemplars that were in the same category as CS+; GEN- indicates exemplars that were in the same category as CS-; CC indicates cross-classified items: Filler indicates filler items; shock electrodes were connected in both Acquisition and Test2, while they were disconnected in Test1; all stimuli presented in Test1 and Test2 were not reinforced.
126
Categorical task. This task consisted of the presentation of 12 breakfast items and 12 bakery items. Some of these items were the same as those that would be presented in the subsequent conditioning task, but nonetheless perceptually different. The pictures were presented once in a random order, to the participants. For each trial, the picture appeared at the center of the screen, with the word ‘breakfast’ appearing on the bottom left of the screen and the word ‘bakery’ appearing on the bottom right of the screen. Participants were instructed to categorize the picture shown by either pressing the left arrow key for breakfast or the right arrow key for bakery. No cross-classified items were shown in this task. This categorical task was carried out prior conditioning for 3 reasons. First, categorizing items into either breakfast or bakery categories would increase the salience of these two categories in the subsequent conditioning task. This encouraged participants to consider both categories when making inferences (Murphy & Ross, 1999). In other words, participants would be more likely to have both breakfast and bakery categories in mind when cross-classified items were shown, rather than merely perceiving them as either breakfast or bakery items. Secondly, the increased salience of both breakfast and bakery categories helped facilitate learning of the category-US association. This was intended to minimize the likelihood that participants would learn unrelated categories during acquisition. For instance, participants would be less likely to infer alternative associations like unhealthy foods predict shock while healthy foods predict no shock. Thirdly, this task allowed participants to be aware of the two predictive categories.
This not only facilitated the learning of the CS-US association, but more importantly, it also facilitated the learning of the CS-No US association. It was found that the CS+ category overshadowed the CS- category in a pilot study. That is, participants tended to learn that one category predicted shock (e.g., breakfast), while all items that did not belong to this particular category predicted no shock (e.g., everything except breakfast items predict no shock). This may greatly attenuate the ambiguity of cross-classified items as participants would only perceive them as belonging to the CS+ items. 127
Acquisition (shock electrodes connected). Participants were informed that different pictures would be presented on the computer screen, which may or may not be followed by a shock. They were asked to learn the relationship between the pictures and shock. Participants were then instructed to use the dial to indicate their expectancy of shock whenever a picture appeared, and to turn the dial to the Off position when the picture disappeared from the screen. The acquisition phase was divided into 2 blocks: in each block, four different breakfast exemplars and four different bakery exemplars served as the CSs, and were presented once each, leading to 8 trials per block. All CSs+ were 100% reinforced, while the
CSs- were never reinforced. The categories that served as CS+ and CS- were counterbalanced across participants. The presentation order was pseudo-randomized so that the same trial type never appeared more than twice in a row. The trial structure was made up by a 10-s baseline period, followed by 10-s stimulus presentation. The electric shock appeared at the last 0.5s of
CS+ presentations and co-terminated with the CS. A message then appeared for 2 seconds instructing participants to turn the dial back to the Off position. If no expectancy ratings were made on initial trials, the experimenter went into the room during the ITI to remind participants to respond on the expectancy dial whenever a picture appeared on the screen.
The test phase consisted of two stages, Test1 and Test2. The shock electrodes were disconnected in the former and reconnected in the latter.
Test1 (shock electrodes disconnected). Immediately after acquisition, the experimenter paused the program and went into the experimental room. Participants were informed that due to ethical restrictions, the number of shocks was limited, hence setting up the cover story for disconnecting the shock electrodes. Although no shock could possibly be delivered, participants were explicitly asked to continue making their expectancy ratings, assuming hypothetically that it was still possible for them to receive a shock. As mentioned in the previous chapter, this procedure was used in order to minimize the impact of extinction of 128 learning, and also to reduce the likelihood that participants would modify their response strategy due to confusion instigated by extinction. In this stage, one CS+ and CS- were randomly chosen and shown once each. Two novel generalisation stimuli of the CS+ category
(GEN+) and two novel stimuli of the CS- category (GEN-) were also presented once each.
The three cross-classified items (CC) and the two filler stimuli that were neither breakfast nor bakery items were also presented once each, leading to a total of 11 trials in this stage. The stimuli were presented in a randomized order, and the trial structure followed acquisition with the exception that no electric shock was delivered.
Test2 (shock electrodes reconnected). The experimenter reconnected the shock electrodes and participants were told that it was again possible to receive shock. This stage was included so that skin conductance responses to the test stimuli could be collected. In fact, no electric shocks were presented. One GEN+ and GEN-, randomly chosen from Test1, were presented once. Two CCs, croissant and English muffin were presented. These two stimuli were chosen to be shown in Test2 because their breakfast typicality ratings were highly similar to their bakery typicality ratings compared to toast. Only these 4 selected trials were presented in this stage to minimize extinction. The order of presentation was random.
When the conditioning task was completed, participants were asked to fill in a 2-page questionnaire. On the first page, participants were asked to write down in detail how they predicted whether a picture would be followed by a shock, and also how they predicted whether a picture would not be followed by a shock. The second page was administered only after the first page was completed, and consisted of 5 statements. Each statement described the relationship between the pictures and the shock (breakfast predicts shock; bakery predicts shock; CC items predict shock; food items that were neither breakfast nor bakery predict shock; others). Each statement was followed by a visual analogue scale from 0% to 100% where the left extreme was labelled False and the right extreme was labelled True. (see 129
Appendix B).
Scoring and analysis
Similar to previous studies in our lab (Lee et al., 2018; Wong & Lovibond, 2017), only expectancy ratings collected in Test1 were used for expectancy analyses. For the skin conductance measure, analysis was based on the data collected when the shock electrodes were attached, since this was when physical shock could be delivered (Acquisition and Test2), so anticipatory anxiety was expected to occur. A low-pass digital filter was applied to cut off any skin conductance activity higher than 50Hz, in order to avoid aliasing, The raw skin conductance data were then log transformed to minimize individual differences. Skin conductance scores for each trial were calculated as the difference between the log of mean
SCL during the 10-s stimulus presentation and log mean SCL during the 10-s baseline period for that trial.
Planned contrasts were used to compare groups and to assess acquisition and generalisation to test stimuli. For acquisition, the second block was compared to the first block in both measures to examine the development of differential responding to CSs+ and
CSs-. In test, the expectancy measures were compared between CSs and GENs, and also between threat and safety cues (i.e., CS+ & GEN+ compared to CS- & GEN-). The resulting interaction was also analysed. Ratings to relatively ambiguous cues (i.e., CC and filler) were compared to ratings to relatively unambiguous cues (i.e., CSs & GENs), to examine if threat ambiguity had any effect on responding. The final, critical contrast was the comparison between CC and filler exemplars, to examine whether responding to ambiguous CCs were any different from baseline responding to novel stimuli with neutral threat value. For skin conductance measures, responding to GEN+ was compared against GEN-, and the averaged responding to GENs was then compared to CCs. Group contrasts were used to compare HA 130 and LA participants. Finally, all interactions between the group and repeated measures contrasts were tested to evaluate group differences in responding to test stimuli.
Exclusion of participants
Statistical analyses were restricted to participants who satisfied the acquisition criterion, that is, participants who demonstrated differential conditioning in their shock expectancy ratings. Differential conditioning was defined as a score of above 30 when the average responses of the last 4 trials to CS- were subtracted from the average responses to the last 4 trials to CS+. The acquisition criterion was relatively lenient because previous studies have found that anxious individuals may show a deficit in safety learning (e.g., Andreatta &
Pauli, 2017; Gazendam et al., 2013). A total of 4 participants (3 in HA group and 1 in LA group) were excluded based on this criterion. Interestingly, all 4 participants responded in the post-experimental questionnaire that they had learnt the predictiveness of shock based on other categorical memberships, for instance, savoury foods predict shock while sweet foods predict no shock. Most importantly, all participants who met the acquisition criterion had learnt the correct categorical memberships, hence the ambiguity of the CCs was presumed to be established. Two participants in the LA group were excluded as they did not provide shock expectancy ratings for at least two stimuli in Test1. Furthermore, 2 participants (one in each group) became suspicious about the study aim, as they asked the experimenter how they should categorize items that can be simultaneously classified as both breakfast and bakery items before the categorical task began. One participant did not follow the instructions and expectancy ratings of one participant was not recorded due to technical problem. Altogether, a total of 10 participants were excluded, leaving 29 participants in the HA group and 31 participants in the LA group.
Missing data 131
Missing data occurred on 1.7% and 1.0% of all trials in the HA and LA group respectively. All missing data in the acquisition trials occurred within the initial 4 training trials, presumably when participants were still learning the experimental requirements.
Missing data during acquisition were replaced with the average ratings made during that particular trial type across all participants within the group. Missing data in test only occurred for the 2 excluded participants that did not rate their shock expectancies for more than 2 trials.
Results
Anxiety groups and shock intensities
The mean DASS-anxiety scores were 16.6 and 2.8 for the HA and LA groups respectively. The HA group had a mean shock intensity of 2.2mA while the LA group had a mean shock intensity of 2.1mA. No group differences were found in the tolerance of electric shock, F(1,58) = 0.4, p=0.53, n.s.
Acquisition
Figure 13A shows the mean shock expectancy ratings during acquisition for the HA and LA group. A significant main effect between CS trial types was observed, F(1,58) =
2 1562.9, p<0.01, ηp = 0.96, suggesting that differential shock ratings to both CSs occurred across the 2 acquisition blocks. Average ratings to all CSs in the first block were significantly
2 lower than those in the second block, F(1,58) = 20.5, p<0.01, ηp = 0.26, presumably because the net increase in ratings to CS+ was slightly greater than the decrease in ratings to CS- across the two blocks. Acquisition of discrimination was confirmed by a significant
2 interaction between CS trial types and blocks, F(1,58) = 267.6, p<0.01, ηp = 0.82, indicating that the difference in ratings to CS+ and CS- became more pronounced in late acquisition trials. No interaction effects involving anxiety groups were observed (highest F = 1.1, p=0.3), suggesting that there were no reliable differences in acquisition between the anxiety groups. 132
A HA LA
100
80
60
40
20
Expectancy shock (%) of 0
Block 1 Block 2 Block 1 Block 2
C S+ (s)
C S- (s) B
HA LA 0.02
0.01
0.00
Change in Log-SCL Change in -0.01
Block 1 Block 2 Block 1 Block 2
Figure 13. Mean shock expectancy ratings (Top panel) and skin conductance level (SCL; Bottom panel) across acquisition trials. HA = High Anxious; LA = Low Anxious. 133
Figure 13B shows the mean change in log SCL during acquisition in the HA and LA groups. Across all trials, responding to all CSs+ were significantly higher than all CSs-, as
2 shown by a significant main effect of CS trial types, F(1,58) = 40.0, p<0.01, ηp = 0.4. Across all CS types, there was no difference in responding to early and late acquisition trials, F(1,58)
= 0.56, p=0.46 n.s. However, similar to expectancy ratings, differential skin conductance developed across acquisition trials, resulting in a significant interaction between CS trial types
2 and early vs late acquisition trials, F(1,58) = 5.8, p=0.02, ηp = 0.06. Unlike expectancy, a significant interaction between CS trial types and groups was observed F(1,58) = 5.6, p=0.02,
2 ηp = 0.06. This result is due to HA participants showing a higher level of differential responding to CSs+ and CSs-, especially in late acquisition trials. No other interactions were significant (highest F = 0.16, p=0.69).
Test phase
Figure 14A shows the expectancy ratings to test stimuli in the HA and LA groups in
Test1. Overall, the averaged ratings to threat category cues across groups (i.e., CS+ &
GENs+) were high while those to the safe category cues (i.e., CS- & GENs-) were low. The overall ratings to the CC and filler items were intermediate between those to the threat and safe cues. Both groups showed significantly higher expectancy ratings to threat category cues
2 than to safe category cues (i.e., CS- & GENs-), F(1,58) = 3287.5, p<0.01, ηp = 0.98.
Furthermore, ratings to the CSs were similar to those to the GENs, resulting in a non- significant difference between CSs and GENs, F(1,58) = 0.0002, p=0.99 n.s. The interaction contrast comparing the difference between ratings to CS+ and CS- to the difference between
2 ratings to GEN+ and GEN- was found significant, F(1,58) = 18.1, p<0.01, ηp = 0.24. This suggests discrimination between the threat and safe categories was slightly better for the trained cues than for the generalisation test cues. No interactions with group were found
(highest F = 0.2, p=0.66), suggesting that trait anxiety had no effect on threat appraisal to 134
A
100
80
60
40
20
Expectancy shock (%) of 0 CS+ CS- GEN+ GEN- CC Fillers
HA (n=29)
LA (n=31)
B
0.03
0.02
0.01
0.00
-0.01
Change in Log-SCL Change in GEN+ GEN- CC GEN+ GEN- CCs
Figure 14. Mean shock expectancy ratings (Top panel) and skin conductance level (SCL; Bottom panel) across test trials. HA = High Anxious; LA = Low Anxious. The skin conductance data were collected during Test2, when the shock electrodes were connected.
135 unambiguous threat and safety cues. Across both groups, no reliable difference in expectancy ratings to relatively ambiguous cues (i.e., CC & filler) and to relatively unambiguous cues
(i.e., CSs & GENs) was observed, F(1,58) = 0.03, p=0.86, n.s. However, the HA group showed better discrimination between the unambiguous and ambiguous cues, supported by a
2 significant interaction, F(1,58) = 11.8, p<0.01, ηp = 0.17. This group difference was mostly driven by the higher ratings to CC exemplars among HA individuals. Furthermore, the averaged shock expectancy ratings to CC exemplars across groups were significantly higher
2 than that to filler items, F(1,58) = 11.0, p<0.01, ηp = 0.16, suggesting an overall higher level of fear responding to CC items when compared to novel food items that neither belonged to the threat nor the safe category. Complementing the significant group difference in shock expectancies to the ambiguous and unambiguous cues, the HA group showed higher expectancy ratings to CC exemplars than filler items compared to the LA group; however this interaction did not quite reach significance, F(1,58) = 2.9, p=0.09, n.s.
Figure 14B shows the skin conductance data collected in Test2, which were broadly consistent with the expectancy measures. Averaged across groups, participants showed
2 significantly higher responding to GEN+ than to GEN-, F(1,58) = 16.0, p<0.01, ηp = 0.23.
Furthermore, the averaged responding to GENs were significantly higher than that to CC
2 exemplars, F(1,58) = 15.4, p<0.01, ηp = 0.19. HA participants appeared to respond more to
GENs and CC exemplars than the LA participants, but no interaction effects involving anxiety groups were observed (highest F = 1.3, p=0.26).
Comparison between High and Low trait anxiety groups
The initial analysis indicated no significant group differences in responding to the CSs and GENs. Although a group trend was observed in the shock expectancy ratings to CCs relative to fillers, this group difference did not quite reach significance. Nonetheless, since we hypothesized a trait anxiety difference in responding to the ambiguous CC exemplars, a direct 136 comparison of expectancy ratings to the CC exemplars between groups was carried out. HA participants had higher overall expectancy ratings to the CC exemplars than the LA group,
2 F(1,58) = 9.5, p<0.01, ηp = 0.14. Similarly, the HA group showed significantly higher skin
2 conductance responding to the CC exemplars than the LA group, F(1,58) = 4.7, p=0.03, ηp =
0.08.
Discussion
The current study aimed to investigate whether fear can be generalised to novel objects beyond physical features, that is, if fear can be spread to novel exemplars that are categorically related to the threat cue. Another aim of this study was to examine whether trait anxiety has any effect on higher-order fear generalisation, especially to ambiguous cues that can be seen to fit both the threat and safe categories simultaneously.
The results provided strong evidence that fear can be generalised categorically, since fear responding selectively transferred to novel exemplars that belong to the same category as
CS+, while inhibitory responding generalised to novel exemplars that belong to the CS- category, in both the shock expectancy and skin conductance measures. Across anxiety groups, responding to the ambiguous CC exemplars was higher than to filler items. Given that filler items were also food exemplars but were neither breakfast nor bakery items, they controlled for baseline responding to novel, supposedly threat neutral stimuli. This suggested that the increased shock expectancy to CC exemplars could be attributed to their partial threat value rather than novelty responses.
In terms of trait anxiety effect, there were no group differences in responding to threat cues (i.e., CS+ and GEN+) and safety cues (i.e., CS- and GEN-), for both the shock expectancy and skin conductance measures. This was presumably because of the clear threat value these exemplars possessed. The critical finding was that high anxious individuals showed higher shock expectancy ratings to the ambiguous CC exemplars compared to the low 137 anxious group. Since the CC exemplars could be simultaneously fit in both threat and safe categories, there was a conflict in threat value which could be seen as increasing their level of threat ambiguity. The present findings are therefore consistent with the idea that trait anxious individuals show a bias in threat appraisal to ambiguity, align with findings in the fear conditioning literature (Boddez et al., 2012; Chan & Lovibond, 1996; Chen & Lovibond,
2016) and with Chapter 5 in this thesis. Similarly, the high anxious group showed more conditioned skin conductance responses to the CC exemplars than the low anxious group.
Negative interpretation of ambiguous cross-classified exemplars?
One alternative explanation to the idea that trait anxious individuals show increased threat appraisal to the ambiguous CC exemplars would be that they perceive the CC items as
GEN+ items. However, if this was true, responding to CC items among high anxious individuals would be highly comparable to those to GEN+, which was not observed in the current study. Nonetheless, it is possible that more high anxious participants than low anxious participants perceived the CC items as only belonging to the threat category. Therefore a follow-up study was conducted, adding a force-choice categorical test between acquisition and test, to assess whether the increased threat appraisal to CC exemplars in trait anxious individuals was due to them interpreting these items as threat cues.
Experiment 3
Method
Experiment 3 only differed from 2 in the following aspects.
Participants
Due to difficulty in recruiting high anxious participants, the pre-screening criterion for high anxious individuals was lowered slightly to a DASS anxiety score of 14 or above.
According to Lovibond and Lovibond (1995a), participants having a DASS anxiety score 138 above 14 are at least severely anxious. Thirty undergraduates were recruited in each group, with 6 participants excluded (see Results for detail). The final sample comprised 54 participants (39 females) with a mean age of 19.6 years (SD = 2.3).
Procedure
Immediately after acquisition, participants were asked to complete a forced-choice categorization test. Participants were verbally informed that no shock would be administered in this phase, and that pictures would be presented, with the word ‘breakfast’ located left of the picture and ‘bakery’ located right of the picture. Participants were required to categorize the picture shown into either breakfast or bakery as fast as possible. If participants perceived the item as a breakfast item, they had to press the left arrow key, if they perceived the item as a bakery item, they had to press the right arrow key. After instructions, participants were asked which key they needed to press if they categorized the picture as a breakfast/bakery item. When participants answered correctly, the experimenter left the experimental room and the test started. Three CC exemplars that were identical to the ones to be presented in the subsequent generalisation test were shown once each. After the categorization test, Test1 and
Test2 were presented exactly as in Experiment 2.
Results
Scoring and analysis
For the newly added categorization test, the number of participants perceiving CC exemplars as belonging to either the threat or safe category were analyzed across groups.
Furthermore, expectancy ratings and skin conductance reponses to each CC exemplar were evaluated to see whether the categorization of CC items affected responding to them correspondingly. Additionally, group differences were also evaluated to see if there was any interaction between trait anxiety and categorization on responding to CC exemplars. 139
Exclusion of participants
Four participants in the HA group and one participants in the LA group were excluded based on the acquisition criterion. One participant in the LA group was excluded for not making more than 2 expectancy ratings in Test1. Altogether, 4 HA and 2 LA individuals were excluded, resulting in 26 participants in the HA group and 28 participants in the LA group.
Missing data
Missing data occurred on 0.02% and 0% of all trials in the HA and LA group respectively. All missing data occurred within the initial 2 training trials, while participants were still learning the experimental requirements. Missing data during acquisition were replaced with the average ratings made during that particular trial type across all participants within the group.
Anxiety groups and shock intensities
The mean DASS anxiety scores were 14.7 and 1.36 for the HA and LA group respectively. The mean shock intensities were 2.5mA and 2.7mA for the HA and LA group respectively. There was no significant group difference in the tolerance of electric shock,
F(1,52) = 0.4, p=0.53, n.s.
Acquisition
Figure 15A shows the mean shock expectancy ratings during acquisition for the HA and LA group. The pattern was similar to Experiment 2. Differential shock ratings to CSs were observed across the two acquisition blocks, supported by a significant main effect
2 between CS trial types, F(1,52) = 729.5, p<0.01, ηp = 0.93. Averaged expectancy ratings across CSs in the first block were significantly lower than those in the second block, F(1,52)
2 = 14.9, p<0.01, ηp = 0.22, presumably because the net increase in ratings to CS+ was slightly greater than the decrease in ratings to CS- across the two blocks. Importantly, the interaction 140
A HA LA
100
80
60
40
20
Expectancy shock (%) of 0 Block 1 Block 2 Block 1 Block 2
C S+ (s)
C S- (s) B HA LA 0.02
0.01
0.00
-0.01 Block 1 Block 2 Block 1 Block 2 Change in Log-SCL Change in
Figure 15. Mean shock expectancy ratings (Top panel) and skin conductance level (SCL; Bottom panel) across acquisition trials. HA = High Anxious; LA = Low Anxious. 141
2 between CS type and block was significant, F(1,52) = 270.2, p<0.01, ηp = 0.84, confirming the development of differential responding to CS+ and CS- across blocks. No interactions involving anxiety groups were observed (highest F = 2.2, p=0.14), suggesting that there were no differences in acquisition between anxiety groups.
Figure 15B shows the mean change in log SCL during acquisition in the HA and LA
2 groups. A significant main effect of CS trial types was observed, F(1,52) = 17.3, p<0.01, ηp =
0.24, suggesting that responding to CS+ was higher than to CS- across blocks. Although no differences in responding to the CSs were observed averaged across blocks, F(1,52) = 0.67, p=0.42, n.s., the interaction between CS trial type and block reached significance, F(1,52) =
2 7.7, p<0.01, ηp = 0.11, confirming that responding to CS+ and CS- differentiated across blocks. Similar to the expectancy data, no interaction effects between anxiety groups were found (highest F = 1.7, p=0.20), suggesting no group differences in acquisition.
Test phase
Figure 16A shows the expectancy ratings to test stimuli in the HA and LA groups.
Both groups showed higher threat appraisal to threat cues (i.e., CS+ & GENs+) than to safety cues (i.e., CS- & GENs-), confirmed by a significant difference in shock expectancies to
2 threat and safety cues, F(1,52) = 407.8, p<0.01, ηp = 0.89. Conversely, there was no evidence that expectancy ratings to the CSs were any different from those to the GEN exemplars,
F(1,52) = 0.03, p=0.86, n.s. The interaction between threat type and trial type was significant,
2 F(1,52) = 14.3, p<0.01, ηp = 0.22, suggesting a larger differential responding to the trained cues than to the novel test cues. Surprisingly, a significant group difference was observed in responding between threat cues (i.e., CS+ & GEN+) and safety cues (i.e., CS- & GEN-),
2 F(1,52) = 4.8, p=0.03, ηp = 0.09. Indeed, HA individuals showed lower expectancy ratings to threat cues relative to LA individuals, while an opposite pattern was observed in responding to safety cues (i.e., poorer discriminative learning) in the HA group. Furthermore, although no 142
A
100
80
60
40
20
Expectancy shock (%) of 0 CS+ CS- GEN+ GEN- CC Filler
HA (n=26)
LA (n=28) B
0.03
0.02
0.01 Change in Log-SCL Change in 0.00 GEN+ GEN- CC
Figure 16. Mean shock expectancy ratings (Top panel) and skin conductance level (SCL; Bottom panel) across test trials. HA = High Anxious; LA = Low Anxious. The skin conductance data were collected during Test2, when the shock electrodes were connected. 143 difference was observed in the expectancy ratings between relatively unambiguous cues (i.e.,
CSs & GENs) and relatively ambiguous cues (i.e., CCs & Fillers) across groups, F(1,52) =
1.6, p=0.21, n.s., HA participants had higher shock expectancies to the relatively ambiguous cues than to the unambiguous cues than their LA counterparts, confirmed by a significant
2 group interaction, F(1,52) = 10.9, p<0.01, ηp = 0.17. For the critical comparison, both groups had significantly higher expectancy ratings to the CC exemplars than to the filler items,
2 F(1,52) = 5.9, p=0.02, ηp = 0.10. This contrast also interacted with group, F(1,52) = 7.2,
2 p=0.01, ηp = 0.12, suggesting that the difference in ratings to CC and filler exemplars was not the same across the groups. In fact, the HA group showed higher expectancy ratings to the CC exemplars than to the filler items, and this difference was more pronounced than that in the
LA group. This finding also suggest that the group difference in shock expectancies between ambiguous and unambiguous cues was driven by HA participants showing higher threat appraisal to the CC exemplars. No other interactions were observed between groups (highest
F = 1.2, p=0.28).
Figure 16B shows the skin conductance data collected in Test2. Averaged across groups, participants showed significantly higher responding to GEN+ than to GEN-, F(1,52)
2 = 13.9, p<0.01, ηp = 0.2. No differences were observed in the comparison between responding to GEN and CC exemplars across groups, F(1,52) = 0.2, p=0.66, n.s. However,
2 this contrast significantly interacted with group, F(1,52) = 6.2, p=0.02, ηp = 0.1, suggesting that HA participants responded more to the CC exemplars compared to the averaged GEN exemplars, whereas the LA participants showed the opposite pattern. No other interactions between anxiety groups were observed (highest F = 0.03, p=0.86), suggesting no other differences in skin conductance responding between groups.
144
Comparison between trait anxiety groups
The initial group comparison suggested substantial differences in expectancy ratings to threat and safe cues, and also to CC and filler exemplars. A substantial group difference was also found in the skin conductance responding between GEN and CC exemplars.
The overall group data revealed that the HA group showed lower expectancy ratings to the threat cues (CS+ & GEN+) compared to the LA group, while showing higher expectancy ratings to the safety cues (CS- & GEN-), supported by a significant interaction between group and the contrast comparing threat and safety cues, F(1,52) = 4.4, p=0.04, ηp2 = 0.08. Further examination of the HA group revealed that trait anxious individuals showed a significant larger generalisation decrement from CS- to GEN- than that from CS+ to GEN-, supported by
2 a significant interaction between threat type and trial type, F(1,25) = 8.0, p<0.01, ηp = 0.24.
However, although this interaction seemed to be driven by a stronger generalisation decrement from CS- to GEN-, the direct comparison between these two stimulus types did not reach significance, F(1,25) = 3.5, p=0.07, n.s. Conversely, the LA group showed a stronger generalisation decrement from CS+ to GEN+ compared to CS- to GEN-, supported by a
2 significant interaction between threat type and trial type, F(1,27) = 6.2, p=0.02, ηp = 0.19.
However, similar to the analyses from the HA group, although LA participants showed an apparent stronger generalisation decrement from CS+ to GEN+, the direct comparison in responding between CS+ and GEN+ did not quite reach significance, F(1,27) = 3.1, p=0.09, n.s.
The second set of analyses followed up the critical significant interaction between groups and the difference between ratings to CC and filler exemplars in the expectancy measure. This significant interaction was due to a larger group difference in shock expectancies to CC exemplars than to filler items. This difference was largely due to HA participants having higher expectancy ratings to the CC exemplars than the LA participants. 145
Follow-up analyses confirmed that the HA group showed significantly higher ratings to CC
2 than filler exemplars, F(1,25) = 22.3, p<0.01, ηp = 0.47, while no evidence indicated any differences in ratings to both CC and filler exemplars in the LA group, F(1,27) = 0.03, p=0.86, n.s.
The third set of follow-up analyses examined the critical group difference between skin conductance responding to GEN and CC exemplars. This significant interaction was due to a larger group difference in skin conductance responding to CC exemplars than to the GEN exemplars. In fact, this difference was driven by LA participants responding less to the CC
2 exemplars than to the GEN exemplars, F(1,27) = 5.0, p=0.03, ηp = 0.18, while there was no evidence that HA participants responded differently to both GENs and CC exemplars, F(1,25)
= 1.8, p=0.19 n.s. This suggests that the HA group showed relatively more fear responding to the CC exemplars than the LA group, aligning with the expectancy measure.
Categorical test
An omnibus Cochrans-Q test was carried out to examine any differences in the categorization of CC exemplars. Surprisingly, more LA participants categorized the CC exemplars as belonging to the threat category averaged across all 3 items, χ2 [2] = 10.6, p<0.01. However, all further Chi-square tests for each CC exemplar were found non- significant (lowest p=0.2), suggesting no trait anxiety differences in categorizing CC exemplars into the threatening or safe category.
Nonetheless, follow-up analyses were carried out to examine how the categorization of
CC exemplars and trait anxiety may affect responding to these stimuli in both expectancy and skin conductance measures. However, an overall analysis was impossible to be carried out since the same participant could have rated one CC exemplar (e.g., croissant) as threatening while rating another CC exemplar (e.g., toast) as safe. Therefore, we only analyzed participants who either rated all CC exemplars into the threat or the safe category. This 146
A
100
80
60
40
20
Expectancy of shock (%) shock of Expectancy 0 Pure threat Pure safe
HA
B LA
0.05
0.04
0.03
0.02
0.01
0.00
-0.01 Change in Log-SCL Change in Pure threat Pure safe
Figure 17. Mean shock expectancy ratings (Top panel) and skin conductance level (SCL; Bottom panel) to the cross-classified exemplars according to how participants categorized them. 147 analysis provides two advantages. First, it allows an overall analysis to be carried out first.
Secondly, trait anxiety effects may be more pronounced.
Five and eleven HA and LA participants categorized all CC exemplars into the threat category respectively, while five and two HA and LA participants categorized all CC items into the safe category respectively. Figure 17A shows the overall expectancy ratings to the CC exemplars according to how participants categorized them. HA participants had higher expectancy ratings to CC exemplars regardless of how they categorized them, supported by a
2 significant main effect of anxiety, F(1,19) = 9.0, p<0.01, ηp = 0.32. Averaged across groups, participants also showed higher expectancy ratings to CC exemplars when they classified them into the threat category, supported by a significant main effect of categorization, F(1,19)
2 = 5.3, p=0.03, ηp = 0.22. However, these effects did not interact, F(1,19) = 0.4, p<0.54, n.s.
Figure 17B shows the skin conductance data. Similar to the expectancy measure, HA participants showed more fear responding to all CC exemplars regardless of how they categorized the exemplars, confirmed by a significant main effect of anxiety, F(1,19) = 6.3,
2 p=0.02, ηp = 0.23. However, no other effects reached significance (highest F = 0.6, p=0.45).
Discussion
The current study followed the aims of Experiment 2, and further examined whether the increased threat appraisal bias to the ambiguous CC exemplars among high anxious individuals was due to a bias in the categorical classification of these items. The current results generally replicated the findings of Experiment 2. First, fear and inhibitory responses were selectively transferred to novel items beyond physical features, evident in both expectancy and skin conductance measures. This suggests that generalisation of fear was operating on a higher-order, categorical level. Secondly, no trait anxiety effect was found in responding to novel exemplars that had clear threat value (i.e., GEN+ & GEN-), but an anxiety effect was observed in responding to the ambiguous CC exemplars that had 148 conflicting threat value. Thirdly, high and low anxious individuals showed similar shock expectancies to the novel filler exemplars that were neither breakfast nor bakery items.
The novel feature of this study was the addition of a categorical test between acquisition and test, to examine whether the increased threat appraisal to the ambiguous CC exemplars was due to more high anxious individuals categorizing the CC exemplars into the threat category. Interestingly, trait anxiety did not have any effect on how individuals categorized the ambiguous CC exemplars, refuting the idea that increased threat appraisal was due to negative categorisation of ambiguous exemplars among trait anxious individuals. To further examine why increased threat appraisal was found to ambiguous exemplars in the high anxious group, analyses were carried out for the CC items to investigate how the categorization of these exemplars and trait anxiety modulated fear responding. Trait anxious individuals were found to have an elevation in fear responding to ambiguity regardless of how they categorized the ambiguous exemplars, evident in both expectancy and skin conductance measures. Although it is tempting to suggest that the increase in threat appraisal was mainly driven by increased responding to items even when they were categorized as safe, this claim was not statistically supported. The effect of categorization was also evident, where participants showed higher expectancy ratings to the CC exemplars when they categorized them into the threat category, and vice versa when the exemplars were categorized into the safe category, however, this effect was not observed in the skin conductance data.
There are three possible reasons for the main effect of trait anxiety. First, trait anxious individuals often utilize a “better safe than sorry” strategy (Eysenck, MacLeod & Mathews,
1987; Lommen et al., 2010). Indeed, some participants reported that though they perceived the CC exemplars as safe items, they made higher expectancy ratings to mentally prepare themselves for shock in case the CC exemplars would be followed by a shock in Test2. This 149 may help explain the elevated shock expectancies to CC exemplars among high anxious individuals even though they perceived them as belonging to the safe category.
Secondly, preliminary evidence has shown that trait anxious individuals are less confident about their judgments when under stress (Fathi-Ashtiani, Ejei, Khodapanahi &
Tarkhorani, 2007; Goette, Bendahan, Thoresen, Hollis & Sandi, 2015). Therefore, trait anxious individuals may not have been fully confident about their judgment that CC exemplars were safe items and utilized a “better safe than sorry” strategy similar to the one mentioned above.
Thirdly, trait anxiety differences were observed to CC exemplars but not to novel
GEN exemplars, presumably due to the greater ambiguity of cross-classified items. This suggests that trait anxious individuals show probability biases to ambiguous threat, similar to the threat estimation biases found in the cognitive literature (e.g., Butler & Mathews, 1987;
Mitte, 2007).
General Discussion
Across two studies using a differential fear conditioning paradigm, it was found that fear selectively generalised to novel items that belong to the CS+ category, but not to those belonging to the CS- category. In other words, the categorical membership of exemplars allowed participants to evaluate their threat value accordingly. Furthermore, both studies showed no trait anxiety differences in responding to threat cues (i.e., CS+ & GEN+) or to safety cues (i.e., CS- & GEN-). In contrast, the critical finding was the increased threat appraisal to the ambiguous CC exemplars among trait anxious individuals. These findings will be discussed below, followed by theoretical considerations.
150
Categorical fear generalisation
The current results replicated findings that fear can be generalised to novel cues that are perceptually dissimilar but categorically similar to the threat cues (e.g., Dunsmoor et al.,
2012, 2014; Meulders et al., 2017). Two aspects of the studies provided strong evidence for higher-order fear generalisation. First, differential responding to CS+ and CS- exemplars was established rapidly within the first acquisition block. Given that each exemplar was presented only once per block, this pattern suggests that participants had learnt that the shock predictiveness was based on the categorical membership of the exemplar. If participants were instead learning the CS-US association for every exemplar, responding should have been highly irregular in the first acquisition block. This finding is consistent with Dunsmoor and colleagues’ studies (2012, 2014), which also found differential responding early in acquisition, favoring the interpretation that participants learnt categorically. Secondly, little generalisation decrement was observed from CS+ to GEN+ exemplars, and similarly from
CS- to GEN- exemplars. This finding is conceptually parallel to a phenomenon commonly observed in the inductive reasoning literature, termed premise monotonicity (Carey, 1985).
As previously discussed in the present chapter, premise monotonicity refers to the principle that the more premises are available, the stronger the conclusion, since the extra premise(s) provide(s) more positive evidence to support the conclusion. The presentation of multiple training exemplars is analogous to presenting multiple premises in inductive reasoning, where the degree of generalisation decrement inversely reflects the strength of the conclusion. That is, the lack of generalisation decrement found in the current studies parallels premise monotonicity due to the usage of multiple training exemplars.
The effect of trait anxiety
Both high and low anxious groups had similar shock expectancies and skin conductance to the CS exemplars, suggesting that trait anxious individuals did not show either 151 enhanced conditionability to CS+ (e.g., Santibáñez-H, & Schroeder, 1988; Zinbarg & Revelle,
1989) nor a failure to inhibit fear responses to safety cues (Gazendam et al., 2013; Kindt &
Soeter, 2014). However, the current study cannot conclude that trait anxious individuals showed no enhanced conditionability to the threat cue since the 100% reinforcement rate rendered responses at ceiling. Conversely, the finding that trait anxious individuals did not display increased fear responding to safety cues cannot be explained by a ceiling effect and is inconsistent with past studies (e.g., Andreatta & Pauli, 2017; Gazendam, Kamphuis & Kindt,
2013; Haaker et al., 2015). However, past studies have shown that CS-US contingency learning plays an important role that determines safety learning among trait anxious individuals. Chan and Lovibond (1996) found that trait anxious individuals showed indiscriminative fear responses to both threat and safety cues, but only when they were unaware of the CS-US contingency. Baas, van Ooijen, Goudriaan and Kenemans (2008) found a similar effect of trait anxiety using a conditional discrimination paradigm.
Participants were presented with the CSs in one of the two contexts. While CS- was never reinforced regardless of the context it was presented in, CS+ was only followed by a shock when presented in one context (i.e., shock context) but not the other (i.e., safe context).
Participants were assessed on whether they were aware that the shock contingency of CS+ was context dependent, and were categorized into the aware and unaware group accordingly.
The authors found that the unaware group had higher trait anxiety scores, and showed increased fear responding to CS- when presented in the shock context in both self-report and skin conductance measures. Andreatta and Pauli (2017) found a negative correlation between differential responding to CSs and trait anxiety, which indicated an increase in fear responding to the safety cue with an increase in trait anxiety. The authors also found a decrease in discrimination between CSs in the unaware group in the overall data. Although the authors did not explicitly assess how aware and unaware trait anxious individuals differed in fear responding, their findings provided some evidence that awareness of CS-US 152 contingency affects fear responding among trait anxious individuals. The finding that trait anxious individuals did not show a deficit in safety learning in the current study could be attributed to them being aware of the CS-US contingency, since the post-experimental questionnaire ensured that all participants had learnt the correct category-shock relationship.
Across both studies, no trait anxiety effect was found in responding to the novel generalisation exemplars (GEN+ & GEN-) in either the expectancy or skin conductance measures. Since participants were presented with multiple exemplars in training, it was likely that the GEN exemplars were perceived as another CS exemplar, therefore reducing their threat ambiguity. Alternatively, the clear categorical memberhip possessed by the GEN exemplars attenuated the level of threat ambiguity induced by their novelty. In these ways, our finding is potentially consistent with the hypothesis that a trait anxiety effect will only be observed when the cues are ambiguous, and is also consistent with past studies that found no trait anxiety effect to unambiguous cues whether threatening or safe (Boddez et al., 2012;
Chan & Lovibond, 1996; Chen & Lovibond, 2016).
The critical finding was that trait anxious individuals showed higher shock expectancy ratings to the ambiguous CC exemplars compared to the low anxious group. Since the CC exemplars could be simultaneously fit in both threat and safe categories, there was a conflict in threat value and hence their predictiveness of shock became ambiguous. This finding is consistent with the idea that trait anxious individuals show a bias in threat appraisal to ambiguity, consistent with the broader cognitive literature (e.g., Eysenck, 1987; MacLeod &
Cohen, 1993). The skin conductance data only provide partial support for this interpretation since the group anxiety difference was only significant in the second study. The mixed findings in the skin conductance data may have been due to the high variability in this measure (Lykken & Venables, 1971).
153
Close negative evidence
Interestingly, both high and low anxious groups showed similar shock expectancy ratings to the filler items. The fillers were food exemplars that were neither breakfast nor bakery items, and they served to control for orienting responding to novel items. One may argue that if trait anxious individuals show over-generalisation of fear when faced with ambiguity, they should also have shown a heightened response to filler items, since the fillers were ambiguous in the sense that their threat value was unknown at all. One possible reason this was not observed in the current findings was the presence of negative evidence. In the inductive reasoning literature, negative evidence refers to premises that do not have the relevant property (here, to not cause an outcome). In contrast, positive evidence refers to premises that do have the property (cause an outcome). It has been found that the presence of negative evidence reduces the breadth of generalisation (Kalish & Lawson, 2007; Voorspoels,
Navarro, Perfors, Ransom & Storms, 2015). Consider the following statements:
Horses have sesamoid bones [premise]
Sharks have sesamoid bones [conclusion] (1)
Horses have sesamoid bones [premise]
Eagles do NOT have sesamoid bones [premise]
Sharks have sesamoid bones [conclusion] (2)
The conclusion in the first argument (1) is considered to be stronger than the second one (2). Since there is only one piece of positive evidence acting as the premise in the first argument, and both the premise and conclusion belong to the superordinate category ‘animal’, one may generalise broadly and arrive to a conclusion that ‘all animals have sesamoid bones’, resulting in a stronger argument. The second argument consists of a premise that contains negative evidence and hence dismisses the conclusion that ‘all animals have sesamoid bones’. 154
Therefore, one may less likely generalise to the superordinate category, and rendering the conclusion relatively weak. Furthermore, the type of negative evidence also affects the breadth of generalisation. Consider the following statements:
Horses use serotonin as a neurotransmitter [premise]
Eagles do NOT use serotonin as a neurotransmitter [premise]
Sharks use serotonin as a neurotransmitter [conclusion] (3)
Horses use serotonin as a neurotransmitter [premise]
Palm trees do NOT use serotonin as a neurotransmitter [premise]
Sharks use serotonin as a neurotransmitter [conclusion] (4)
Although both arguments contain a piece of negative evidence, the first argument (3) is considered to be weaker. This is because it contains “close” negative evidence, where the exemplar in the negative premise belongs to a subcategory (bird) relatively closer to the one in the conclusion (fish). In contrast, the second argument (4) contains “distant” negative evidence, where the exemplar in the negative premise comes from a distant category (plants).
Past studies have found that the presence of distant negative evidence increases generalisation from the positive evidence to the conclusion (Heussen, Voorspoels, Verheyen, Storms &
Hampton, 2011; Voorspoels et al., 2015). In contrast, the presence of close negative evidence may lead to less generalisation.
Positive and negative evidence are analogous to CS+ and CS- in conditioning paradigms respectively, as the former predicts an outcome while the latter predicts the omission of an outcome. The CS- exemplars used in the current study resembled close negative evidence, since they belong to the same superordinate category as the filler items
(food). Therefore, they served to lower the probability of participants making a strong conclusion from the positive evidence. In other words, it would be less likely that participants 155 would come to the conclusion that ‘all foods predict shock’, hence decreasing the generalisation of fear to other novel food items. The presence of close negative evidence may have also attenuated any trait anxiety effect on fear generalisation to the filler items via two processes. First, it may have led to the conclusion that ‘not all food causes shock’, therefore reducing the perceived ambiguity induced by novel food items that had unknown threat value.
Secondly, the conclusion that ‘not all food cause shock’ may have helped to confine fear to exemplars fully or partially belonging to the threat category (e.g., CS+, GEN+ & CC exemplars).
Associative and cognitive accounts
The current results appear to be outside the scope of traditional associative accounts.
First, despite the fact that each exemplar was shown only once in each acquisition block, differential responding was observed to develop in the first block. This finding suggests that learning had been facilitated by other factors rather than associative strength that was accumulated on a trial-by-trial basis to individual stimulus - for instance, inductive reasoning.
Similarly, fear generalised to novel stimuli that were perceptually dissimilar to the CS+ exemplars, a result that cannot be readily explained by the associative accounts since they predict generalisation as a function of physical similarity (e.g., Blough, 1975; McLaren &
Mackintosh, 2002). One may argue that most bakery exemplars shown were perceptually similar in terms of sharing similar brownish colours, hence promoting generalisation in a perceptual way. However, strong generalisation (i.e., little generalisation decrement) was also observed in the breakfast items despite them being dissimilar in their perceptual features.
In contrast, the cognitive account is more consistent with the current findings. As discussed previously, generalisation in humans can be seen as a form of inductive reasoning.
That is, knowledge of an exemplar’s property (e.g., predicts shock or not) is more likely to transfer to novel exemplars if they belong to the same category. This was evident in the early 156 development of differential responding to the CSs and the strong generalisation observed from CS exemplars to novel exemplars from the same category. Furthermore, the increased responding to CC exemplars across groups was due to them sharing category membership with the threat cues, rather than them being merely more physically similar to these cues.
Interpretation of CC exemplars
The second study provided no evidence that trait anxious individuals were more likely to categorize the ambiguous CC exemplars into the threat category. In other words, no negative interpretation of ambiguity was found among trait anxious individuals at the level of categorisation, which is inconsistent with some models of cognitive bias (e.g., Eysenck, 1987;
MacLeod & Cohens, 1993). Instead, trait anxious individuals showed a general increase in fear responding to the ambiguous CC items regardless of how they categorized them.
However, as previously discussed, the fact that a trait anxiety effect was not observed in responding to other exemplars that had a clearly defined threat value (e.g., GEN+ & GEN-) suggested that trait anxious individuals were showing threat appraisal biases to CC items due to their ambiguous threat value and possibly their usage of a “better safe than sorry” strategy.
Limitation
One limitation was that no assessments were taken to identify whether there was a mismatch between participants’ own categorization of the CC exemplars and their perception of how the experimenter categorized the CC exemplars. For instance, a participant may have categorized CC exemplars into the safe category, but thought that the experimenter classified them into the threat category. This mismatch in the perceived categorization may have increased participants’ fear responding to the CC exemplars despite them having categorized these exemplars as safe items, and vice versa if participants categorized them as threat items.
It is possible that the high anxious participants had a mismatch between their own categorization and their perception of how the experimenter classified the items, combined 157 with trait anxious individuals less confident about their judgements under stress (Fathi-
Ashtiani et al., 2007; Goette et al., 2015) that resulted in the apparent increased threat appraisal to CC exemplars. Future studies could include a formal assessment to separate the effect of trait anxiety and mismatch in categorization.
Chapter summary
This chapter presented two experiments to examine the impact of trait anxiety on higher-order fear generalisation in humans, namely categorical generalisation. Multiple exemplars from two categories were shown, with exemplars from one category paired with an aversive shock, and exemplars from another category never paired. Fear generalisation was assessed by the presentation of novel breakfast and bakery exemplars in test, and also ambiguous cross-classified exemplars that fit in both categories simultaneously. The results were consistent with past studies (Dunsmoor et al., 2012, 2014; Meulders et al., 2017), showing that fear responses selectively transfer to novel exemplars that belong to the same category as CS+, while inhibitory responses transferred to novel exemplars that belong to the
CS- category. Trait anxiety was found to have no effect on responding to the novel GEN exemplars, but increased fear responding to the ambiguous CC exemplars. This trait anxiety bias in fear responding to CC exemplars was not due to trait anxious individuals interpreting the CC exemplars as threat items, but instead was driven by an increase in fear response regardless of how participants categorized the items. In other words, the ambiguous nature of the CC exemplars increased fear generalisation among trait anxious individuals, presumably due to the anxious individuals utilizing a ‘better safe than sorry’ strategy when faced with threat ambiguity.
158
Chapter 7:
The effect of trait anxiety on the generalisation of extinction
159
The previous chapters have focused on the generalisation of fear acquisition, and how trait anxiety affects this process. Despite the importance of advancing the understanding of the etiology of anxiety disorders, it is equally important to study how to reduce the acquired fear. Therefore, the present chapter will focus on the reduction of conditioned fear to a threat cue, which is termed fear extinction (Pavlov, 1927; Mowrer, 1939) and examine if extinction learning can be generalised to other objects effectively.
Extinction and exposure-based therapies
Pavlov (1927) discovered that once conditioned salivary responses had been established to a specific cue such as a tone (CS+), subsequent unreinforced presentations of
CS+ led to a gradual decrease in CR. This decrease in responding to a cue that previously predicted an outcome is called extinction. Since Pavlov’s (1927) demonstration of extinction, numerous animal studies have been carried out to examine this phenomenon. Similar to
Pavlov’s (1927) finding, the animal literature commonly found a gradual decrease in CR when the CS+ was no longer reinforced (e.g., Bouton & King, 1983; Gormezano,
Schneiderman, Deaux & Fuentes, 1962; Schmaltz & Theios, 1972). Similar findings were observed in instrumental/operant conditioning paradigms, where a decrease in responding to
S+ was observed when responding in its presence was no longer reinforced (e.g., Rescorla,
1992; Rescorla & Skucy, 1969; Wagner, 1961). Conditioned responses to CS+ or instrumental responding to S+ were said to be ‘extinguished’ when they were similar to responding to non-reinforced cues (CS- or S-). In the context of fear conditioning, the decrease in the magnitude of conditioned fear to CS+ is called fear extinction. Similar to the findings in appetitive conditioning, animals showed a reduction in conditioned fear to a CS+ that was no longer followed by an aversive outcome, evident in terms of a decrease in freezing (e.g., Bouton & Bolles, 1979; Ledgerwood, Richardson & Cranney, 2003) and conditioned suppression (e.g., Bouton, 1986; Reberg, 1972). By the same token, instrumental 160 responding like fear avoidance was found to decrease when S+ was no longer reinforced
(Baum, 1966, 1969).
The reduction of conditioned fear in extinction paradigms provides a plausible explanation for the effectiveness of exposure-based therapies. The process of fear extinction learning is analogous to fear reduction in exposure-based treatments, which involve systematically exposing anxious patients to a fear-provoking stimulus (e.g., exposing arachnophobics to spiders; Öst et al., 1997a, 1997b). In fact, the laboratory model of fear extinction provides great face validity, construct validity and predictive validity for exposure- based treatments (Scheveneels et al., 2016). These evidence-based exposure interventions have been found to effectively attenuate anxiety symptoms (Butler, Chapman, Forman &
Beck, 2006; Hofmann & Smits, 2008; Norton & Prince, 2007). However, although exposure- based treatments are undoubtedly effective in treating anxiety disorders, it is also common to see relapse after successful treatment (Craske and Mystkowski, 2006; Ginsburg et al., 2014).
Return of fear
Return of fear refers to the re-emergence of conditioned fear following successful extinction. It is a robust phenomenon and is analogous to relapse after exposure-based therapies. There are three known pathways for the return of fear, namely spontaneous recovery, renewal and reinstatement. Pavlov (1927) discovered that when the conditioned salivary responses had been successfully extinguished, the mere passage of time allowed the recovery of conditioned response to the CS+. Spontaneous recovery of conditioned fear has been observed in numerous animal studies (e.g., Brown et al., 1951; Burdick & James, 1970;
Shurtleff & Ayres, 1980) and human studies (e.g., Huff, Hernandez, Blanding & LaBar, 2009;
Norrholm et al., 2008). This phenomenon helps explain why some patients show a relapse of anxiety symptoms some time after successful treatment. 161
Renewal refers to return of fear after a change in context between extinction and test
(Bouton & Bolles, 1979; Bouton & King, 1983). Context refers to the surroundings where the
CSs are presented. Empirical studies have shown that a mismatch in extinction and test contexts leads to return of fear in both animal (e.g., Bouton & Peck, 1989; Bouton & Ricker,
1994; Holmes & Westbrook, 2014) and human studies (e.g., Alvarez, Johnson & Grillon,
2007; Effting & Kindt, 2007; Vansteenwegen et al., 2005). Similarly, when patients receive treatment, they are surrounded by the therapist and other cues in the clinical setting. When patients later encounter the fear cue in other settings, their anxiety symptoms may relapse due to this change in context (Mystkowski et al., 2002; Rodriguez, Craske, Mineka & Hladek,
1999).
Reinstatement refers to the return of conditioned fear to a CS+ if the US is presented alone after fear extinction (Rescorla & Heth, 1975; Westbrook, Iordanova, McNally,
Richardson & Harris, 2002). Similar to other return of fear phenomena, reinstatement is a robust finding evident in both animal (e.g., Bouton, 1984; Rescorla & Heth, 1975; Westbrook et al., 2002) and human studies (e.g., Dirikx, Hermans, Vansteenwegen, Baeyens & Eelen,
2007; Hermans et al., 2005; Norrholm et al., 2006). It has been hypothesized that exposure to another traumatic event (analogous to the US) may trigger a return of fear after successful treatment in anxious patients.
Theories of extinction
The phenomenon of return of fear suggest that extinction is not a process of unlearning the original CS-US association. Indeed, Pavlov (1927) was the first to suggest that extinction involves inhibitory learning, forming a preliminary idea that extinction involves the formation of a new CS-no US association. Pavlov (1927) also assumed that this inhibitory association is relatively fragile compared to the original CS-US association, hence explaining why conditioned fear would return after extinction. Since then, associative theorists have proposed 162 the inhibitory learning model, emphasizing the idea that extinction is not unlearning, but instead involves inhibitory learning of a new CS-no US association (Bouton, 1993; Miller &
Matzel, 1988). Bouton (1993, 2007) proposed that the extinguished CS+ have two meanings: the original CS-US association and the newly formed CS-no US association. These two associations compete with each other when the CS is presented; if the original CS-US association dominates over the inhibitory association, return of fear occurs. Since the CS is now capable of activating either the excitatory or the inhibitory association, the CS+ becomes ambiguous.
Besides empirical demonstrations of the return of fear in behavioural and psychophysiological measures, neuroimaging studies also provide supportive evidence for inhibitory learning during fear extinction. While it has been known that there is an increase in amygdala activity during fear acquisition (Davis, 1992; Maren & Fanselow, 1996), an increase in prefrontal cortex (PFC) activity and a decrease in amygdala activity has been found during extinction (Phelps, Delgado, Nearing & LeDoux, 2004; Quirk & Mueller, 2008).
The PFC has been suggested to be involved in the inhibitory circuit responsible for downregulating amygdala activity during extinction (Morgan, Romanski & LeDoux, 1993;
Quirk, Garcia & Gonzalez-Lima, 2006; Sotres-Bayon, Cain & LeDoux, 2006), therefore inhibiting the expression of conditioned fear to CS+ (Gottfried & Dolan, 2004; Greco &
Liberzon, 2016).
Prediction error, which is thought to be critical in CS-US acquisition, also plays an important role in inhibitory learning (Miller & Matzel, 1988). As suggested by Rescorla and
Wagner’s (1972) contingency model, when a CS that previously predicted an outcome is now unreinforced, a large negative prediction error is produced, resulting in significant inhibitory learning. In other words, the strength of an inhibitory (CS-no US) association increases across extinction trial, competing with the original excitatory link. 163
Similar to fear learning in humans, cognition appears to play an important role in fear extinction. In cognitive terms, extinction refers to the establishment of a belief that the CS will no longer followed by the outcome it previously predicted (Hofmann, 2008; Lovibond,
2004). It has been found that extinction of conditioned fear closely follows participants’ causal belief that the CS will no longer predict the US (Biferno & Dawson, 1977; Dawson &
Schell, 1985). That is, the stronger the belief that the US will no longer follow the CS, the stronger the reduction in CRs. However, if the belief that the CS leads to the US is not disconfirmed, protection of extinction occurs. Lovibond, Davis and Flaherty (2000) trained participants with several cues. Two cues, C and D, predicted an aversive outcome during acquisition. In the subsequent extinction phase, while D was presented alone, C was presented in compound with a stimulus, E, that had been previously established as a conditioned inhibitor, or a safety signal. That is, participants were trained that when a cue that predicted shock was presented with cue E, no shock would occur. It was found that participants showed successful extinction to D, but responding to C was comparable to another unextinguished
CS+. In other words, fear responding to C was not extinguished. Lovibond et al. (2000) argued that was because participants attributed the omission of shock following CE to the presence of E, hence their belief that C predicts shock remained intact. The phenomenon of protection of extinction can also be explained by certain associative accounts, such as the
Rescorla-Wagner (1972) model. The model assumes that the negative associative strength of the conditioned inhibitor (cue E) balances out the positive associative strength of the excitatory cue (cue C), leading to little to no mismatch between the predicted outcome
(nothing) and the actual outcome (nothing). This leads to little to no prediction error, resulting in the lack of extinction learning to the excitatory cue.
Awareness of the change in CS-US contingency also plays an important role in extinction. For instance, Biferno and Dawson (1977) found that participants who were unaware that the CS was no longer followed by an US during extinction showed significantly 164 less decrease in conditioned skin conductance responses to the CS. Empirical studies have also shown that verbal instruction that the CS will no longer followed by an US is sufficient to produce extinction learning. After fear acquisition, Colgan (1970) informed participants that the CS tone will no longer followed by a shock, and participants showed a sharp reduction in anticipatory skin conductance responses to the tone on the first extinction trial.
Similarly, Grings, Schell and Carey (1973) showed that after establishing differential responding to the CSs, participants who were instructed about a reversal in CS-US contingency demonstrated immediate discrimination reversal in the skin conductance measure. In other words, participants immediately showed an increase in CRs to CS- while showing a decrease in CRs to CS+ after the reversal instructions.
In summary, the cognitive account suggests that the formation of a causal belief that the CS will no longer followed by an US is a prerequisite for extinction learning to CS to take place.
Clinical anxiety in fear extinction
The fear extinction model has been used to study the association between clinical anxiety and extinction learning, to further our understanding of the process underlying extinction in anxiety and improve the effectiveness of exposure-based therapies. Empirical studies have been carried out to examine how clinically anxious patients differ in the process of fear extinction from the healthy population. One early study by Pitman and Orr (1986) compared fear extinction between healthy participants and a cohort with mixed anxiety disorders, such as GAD, PD and SAD patients. While the healthy controls showed non- differential conditioned skin conductance responses to the CSs after extinction, the anxious patients maintained differential responding to the CSs throughout extinction, suggesting that they did not show as much extinction learning to the CS+. More recent studies examined the process of fear extinction in specific anxiety disorders. Some evidence that has been 165 mentioned previously in Chapter 3 suggested that PTSD patients showing enhanced conditionability to CS+ (Orr et al., 2000) also showed differential conditioned skin conductance responses to the CS+ and CS- after extinction. This suggested that PTSD patients are resistant to fear extinction. This effect was not simply due to patients showing more conditioned fear to CS+ hence requiring more extinction trials to extinguish their responding to CS+, as patients also showed a significant elevation in responding to CS- too. Other studies that did not find enhanced conditionability to CS+ in PTSD patients also showed resistant to fear extinction in conditioned heart rate responses (Peri et al., 2000) and in US expectancy and valence ratings (Blechert, Michael, Vriends, Margraf & Wilhelm, 2007). It was also suggested that the severity of PTSD symptoms is positively correlated with impaired fear extinction (Norrholm et al., 2011).
Similar findings in fear extinction were observed in patients suffering from other anxiety disorders. For instance, PD patients showed more conditioned skin conductance responses to the CS+ during extinction and evaluated the CS+ as more unpleasant (Michael,
Blechert, Vriends, Margraf & Wilhelm, 2007, but see Del-Ben et al., 2001). SAD patients continued to show differential conditioned skin conductance responses to the CSs throughout fear extinction (Hermann et al., 2002).
Trait anxiety in fear extinction
Similar to the findings in fear extinction in clinical studies, trait anxious individuals show a deficit in the extinction of fear learning. After fear extinction, trait anxious individuals showed an overall higher responding to the extinguished CS+ in both eyeblink startle responses and self-report distress ratings (Gazendam et al., 2013). In addition, trait anxious individuals showed a significantly slower decrease in startle responses to CS+ during extinction, further supporting the idea that individuals at risk of developing anxiety disorders exhibit resistance in extinction learning. The delivery of a surprising reinstatement shock after 166 extinction led to an increase in startle responses and self-report distress ratings to both CSs in the trait anxious group and control group. However, while the control group quickly showed re-extinction to CS+ in both startle responses and distress ratings, trait anxious individuals continued to show differential responding to both CSs at the end of the re-extinction phase.
Dibbets, van den Broek and Evers (2014) found that trait anxious individuals showed significantly higher expectancy ratings to CS+ after extinction compared to a non-anxious control group. Similarly, individuals high in IU (intolerance of uncertainty), a vulnerability factor for developing anxiety disorders (e.g., Boelen & Reijntjes, 2008; Dugas et al., 1998;
Fetzner et al., 2013) continued to show differential skin conductance responding to CS+ and
CS- after extinction, suggesting individuals at risk of developing anxiety disorder show resistance to fear extinction (Morriss, Christakou & van Reekum, 2016).
Complementing the behavioural studies, neuroimaging studies have also found a similar pattern in fear extinction among trait anxious individuals. Barrett and Armony (2009) found a positive correlation between trait anxiety and amygdala activity during fear extinction. Furthermore, empirical evidence has shown that trait anxious individuals exhibit lower activity in the PFC (Indovina et al., 2011) and the dorsal anterior cingulate cortex
(dACC; Sehlmeyer et al., 2011) during extinction. As previously mentioned, the PFC and the associated dACC regions are involved in the ‘extinction circuit’ that inhibits the amygdala activity during extinction (Morgan et al., 1993; Quirk et al., 2006; Sotres-Bayon et al., 2006).
Therefore, neuroimaging studies suggest that trait anxious individuals have a deficiency in the inhibition of amygdala activity that leads to resistance to fear extinction.
Generalisation of extinction
Traditionally, fear extinction studies involve fear acquisition to the CS+, and then extinction of the exact same stimulus (i.e., non-reinforced presentations of CS+). In exposure- based therapies, however, it is very unlikely that the original CS+ can be presented. For 167 instance, it is highly unlikely to present the original perpetrator to a rape victim in exposure- based therapies. Technically speaking, therefore, GSs that are perceptually or conceptually related to the original threat cue are presented during treatment. In light of this issue, there have been a few laboratory studies investigating how effectively presenting a GS during extinction reduces conditioned fear to the original CS+, in other words, how well can extinction learning to a GS be generalised to the CS+ (generalisation of extinction effect).
Using geometrical shapes as stimuli, Vervliet, Vansteenwegen, Baeyens, Hermans and Eelen
(2005) trained participants to respond differentially to a CS+ (e.g., a triangle) and to a CS-
(e.g., a parallelogram). In the subsequent extinction phase, one group of participants were presented with a stimulus perceptually similar to CS+ (i.e., GS+; a perceptually different triangle) and a stimulus perceptually similar to CS- (i.e., GS-; a perceptually different parallelogram), while another group of participants were presented with the original CSs. In the following test phase, the CSs were presented and participants’ responding to them was assessed. The group that were presented with GSs during extinction showed more conditioned skin conductance responses and retrospective expectancy ratings to the CS+ compared to the control group that received standard fear extinction. In contrast, no group differences were observed in responding to CS- regardless of whether GS- or CS- was presented in extinction.
The results suggested that while presenting GS- during extinction has no effect on responding to CS- in test, extinction with GS is less effective than standard fear extinction in terms of inhibiting conditioned fear to the CS+. Similarly, Vervliet and Geens (2014) presented two
GSs that shared a unique perceptual feature with the CS+ during extinction. For instance, participants were trained with a yellow triangle (CS+). In the subsequent extinction phase, a
GS that shared the shape of CS+ (i.e., blue triangle) and a GS that shared the colour of CS+
(i.e., yellow square) were presented. In the critical test phase, participants continued to show heightened expectancy ratings and skin conductance responses to the CS+. The results 168 support the notion that presenting a GS+ in extinction cannot effectively reduce fear responding to the original CS+.
The effect of generalisation of extinction has also been examined in symbolic generalisation (Vervoort et al., 2014). Using a MTS task, participants learnt to classify some arbitrary stimuli into two categories ([A1, B1, C1 & D1] and [A2, B2, C2 & D2]). A stimulus from one category was then paired with a shock (B1+) while a stimulus from another category was not reinforced (B2-). In the subsequent generalisation task, participants showed fear generalisation to stimuli that belong to the same category as B1+ (A1, C1 & D1) but not to stimuli that shared categorical membership with the safe cue (A2, C2 & D2). One group of participants were then presented with the CSs (B1 & B2) in extinction, while another group of participants were presented with the GSs (C1 & C2) in extinction. In test, participants who received GSs in extinction showed higher US expectancies to the CS+ than those who received CSs in extinction. Therefore, similar to the generalisation of extinction studies that used perceptually similar stimuli, preliminary evidence suggests that extinction learning to a
GS that is categorically related to the CS+ cannot effectively reduce conditioned fear to the original CS+.
The associative accounts explain the failure of generalisation of extinction in terms of shared elements between the CS+ and the GS. The GS shares some common elements with the CS+, and these common elements gain inhibitory strength when the GS is presented in extinction. However, the remaining elements in the CS+ that are not shared with the GS retain their excitatory strength. Therefore, inhibitory learning to the GS fails to extinguish responding to CS+ completely because of these unextinguished elements uniquely possessed by the CS+. In cognitive terms, participants may not be willing to extrapolate the extinction knowledge to GS to the original CS, hence displaying heightened fear responses to CS+ after the extinction of GS. 169
Theoretically, increasing the perceptual similarity between CS+ and GS can arguably promote the generalisation of extinction learning from the GS to the CS+. In associative terms, the increase in perceptual feature between stimuli will result in more common elements in CS+ gaining inhibitory strength during extinction, therefore leaving fewer elements in the
CS+ that retain their excitatory strength (Blough, 1975; McLaren & Mackintosh, 2002). In cognitive terms, increasing the perceptual similarity between stimuli may give individuals more confidence to extrapolate extinction learning from GS to CS+. Therefore, the failure to find a generalisation of extinction effect in Vervliet and colleagues’ (2005, 2014) and
Vervoort et al.’s (2014) studies may be due to the CS and GS not being perceptually similar enough.
The current study followed and modified the paradigm used by Vervliet et al. (2005) to further examine generalisation of extinction. The aspect that we followed was the usage of
ABA and AAA groups. Although this terminology is usually used in context renewal studies where different letters stand for different contexts (Bouton & Bolles, 1979; Bouton & King,
1983), the context remained constant in the current study. Instead, each letter represented a stimulus, where A stood for CS+ and B stood for the GS. In other words, both groups were presented with the CS+ during training and in test, but the ABA group received a GS during extinction while the AAA control group received standard extinction with the CS+.
Additionally, the current study sought to examine the effect of trait anxiety on the generalisation of extinction learning. This led to a 2 (ABA and AAA) by 2 (trait anxiety) between-subject design.
There were two aspects that were modified from Vervliet et al.’s (2005) study. First, instead of geometrical shapes that differed in size, circles differing in hue on a blue-green dimension were used in order to increase the perceptual similarity between the CS and the GS
(see Livesey & McLaren, 2009). Secondly, a single-cue conditioning procedure was used 170 instead of differential conditioning for two reasons. First, if a perceptually similar CS- was included, a reduction in fear responses to CS+ in test could be attributed to the inhibitory strength of the shared elements between CS+ and CS-, or to the extinguished excitatory elements shared between the GS and CS+, or a combination of both. It is difficult to unravel these effects contributing to the reduction of CRs to the CS+. Therefore, a safety cue (CS-) was not included in the current study in order to solely focus on the effect of presenting a GS in extinction on the conditioned fear to CS+. Secondly, the presence of a CS- may increase the opportunity for rule formation, potentially decreasing the ambiguity of the experimental configuration (see Wong & Lovibond, 2017). Hence, participants were only trained with a single CS+ to maintain the ambiguity of a ‘weak’ situation to maximise the opportunity of observing an effect of trait anxiety.
Experiment 4
Method
Participants
Undergraduate students were recruited as participants who received course credit or
AUD $10 for participation. Only non-colour-blind individuals were eligible for participating the current study. Participants were also pre-screened with the DASS-21 (Lovibond &
Lovibond, 1995a) via the SONA recruiting system. Participants with a DASS-anxiety score of
4 or below were recruited to the low anxious (LA) control group, while those with a DASS- anxiety score of 14 or above were assigned to the high anxious (HA) group. Participants were re-administered the DASS-21 at the time of testing, and only those who had a DASS anxiety score consistent with pre-screening were included in the study. Thirty participants were recruited in each group (across anxiety and conditions), resulting in a total of 120 participants.
Twenty participants were excluded (see Results for more detail). The final sample comprised
100 participants (74 females) with a mean age of 19.5 (SD = 2.6). 171
Apparatus and materials
The hardware (e.g., shock stimulator) and software (e.g., programming language) were the same as previously. Figure 18 shows the circle stimuli (A & B) used in the current study, with a radius of 5.75cm. Both stimuli lay on a blue-green dimension, varying in their hue value (HSV value). Stimulus A was an aqua colour circle with a HSV value of 0.446, and stimulus B was a green circle with a HSV value of 0.479. The saturation and brightness for both stimuli were held constant at 1 and 0.75 respectively. All stimuli were presented in the center of a grey background (RGB value = [200 200 200]) on the computer screen.
A B
Figure 18. Stimulus A served as the CS+. Stimulus B served as the GS.
Procedure
After signing the consent form, participants were asked to fill in the DASS-21. Shock electrodes were attached to participants’ fingers, and they were then led through a work-up procedure in which they selected a level of shock that was ‘definitely uncomfortable but not painful’. Isotonic gel was squeezed into the GRASS® silver disc electrodes, to maximize the sensitivity of skin conductance measure, and then attached to participants’ fingers.
Participants were then taken into the experimental room. As shown in Table 6, the study consisted of an acquisition phase, followed by an extinction phase and a test phase. Unlike studies in the previous chapters, the shock electrodes were connected throughout the entire study. Before the experiment started, headphones were placed on participants. White noise was presented continuously throughout the study for noise cancellation. 172
Table 6. Design of current study Phase/Condition Acquisition Extinction Test
ABA A+ (6) B- (8) A- (3)
AAA A- (2) A- (8)
Note. + indicates shock presentation; - indicates shock omission; numbers in brackets indicate the number of trials of that type in each phase.
Acquisition. Participants were informed that some circles would be presented on the screen, which may or may not be followed by a shock. They were asked to learn the relationship between the pictures and shock. Participants were then instructed to use the dial to indicate their expectancy of shock whenever a circle appeared, and to turn the dial to the
Off position when the picture disappeared from the screen.
Both ABA and AAA conditions received 8 trials of stimulus A presentations, in which
6 trials were reinforced (i.e., 75% reinforcement). Stimulus A was not fully reinforced for two reasons. First, extinction learning in human occurs rapidly. Partial reinforcement of A+ slows down extinction (Mackintosh, 1974) and allows room to examine the effect of trait anxiety on fear extinction. Secondly, partial reinforcement is thought to increase the ambiguity of the experimental configuration, rendering it a ‘weak’ situation (Lissek et al., 2006), again increasing the opportunity to observe any trait anxiety effects. The presentation order was pseudo-randomized, so that non-reinforced A trials did not occur twice in a row, and the first and last trials were always reinforced. The trial structure was made up by a 10-s baseline period and a 10-s stimulus presentation. Electric shock, if presented, was delivered in the last
0.5s of A+ trials and co-terminated with the CS. The ITI varied between 10 and 21s, starting from stimulus offset to the onset of the baseline period for the next trial.
Extinction. In this stage, participants in the ABA condition received 8 trials of B- presentations, while those in the AAA condition received 8 trials of A- presentations. All 173 stimulus presentations in this stage were non-reinforced. The trial structure was identical to
Acquisition, with the exception that no shock was delivered at all.
Test. Participants in both the ABA and AAA conditions received 3 trials of A- presentations. This phase was to assess how extinction of the training CS (AAA) and extinction with a novel GS (ABA) would affect fear responding to A. The trial structure was identical to Extinction.
When the conditioning task was completed, participants were asked to fill in a 1-page questionnaire (see Appendix C). Participants were asked how many different colours were presented in the experiment, and were asked to name or describe the colours they saw. Given that the stimuli used were highly similar, this questionnaire served as an important check for whether participants in the ABA condition were able to distinguish between stimuli A and B, or if participants in the AAA condition had a misperception that more than one colour was presented. For instance, if a participant in the ABA condition failed to distinguish between stimuli A and B, any effect on responding to A in test would be due to standard extinction (an
AAA effect) rather than a genuine effect of generalisation of extinction (an ABA effect).
Scoring and analysis
A low-pass digital filter was applied to cut off any skin conductance activity higher than 50Hz, in order to avoid aliasing. The raw skin conductance data were then log transformed to minimize individual differences. Skin conductance scores for each trial were calculated as the difference between the log of mean SCL during the 10-s stimulus presentation and log mean SCL during the 10-s baseline period for that trial.
Planned contrasts were used to compare the two group factors – trait anxiety and conditions across acquisition, extinction and test. Acquisition, extinction and test data were analyzed with a linear trend repeated measures contrast. All interactions between each factor 174 and repeated measures contrasts were tested to evaluate group differences in linear trend in each phase.
In addition, responding on the last acquisition trial and the first extinction trial was directly compared, to assess any changes in the level of responding with a change in stimulus
(i.e., generalisation decrement; ABA condition). The critical contrast was the comparison of responding between the last extinction trial and the first test trial, to examine whether a change in perceptual feature would affect the responding to the test stimulus.
Exclusion of participants
Statistical analyses were applied to participants who correctly answered the number of colours presented depending on their condition. Across the ABA condition, 8 participants were excluded because they either reported there was only one colour or more than two colours presented throughout the study. Five participants in the AAA condition reported seeing more than one colour in the study, and hence were excluded. Participants who did not show an averaged expectancy rating to stimulus A above 50 in the last 4 acquisition trials were also excluded. Across anxiety groups, 6 participants were excluded based on this criterion. Additionally, 1 participant in the AAA condition was excluded due to not giving expectancy ratings for more than 2 trials in Extinction and Test. Altogether, 20 participants were excluded, leaving 100 participants in the final sample.
Missing data
Missing data occurred on 1.6% and 1.3% of all trials in the HA and LA group respectively. Most missing data occurred within the initial 4 acquisition trials, when participants were still learning the experimental requirements. Missing data were replaced with the average ratings during that particular trial type across all participants within the condition. 175
Results
Anxiety groups and shock intensities
The mean DASS-anxiety scores were 14.8 and 3.8 for the HA and LA groups respectively. The mean shock intensities for both groups were 2.5, indicating no anxiety difference in the tolerance of electric shock, F(1,98) = 0.001, p=0.98, n.s.
Acquisition
Figure 19A shows the mean shock expectancy ratings during acquisition. All participants showed a rapid increase in expectancy ratings to A+ (CS+) which leveled off for the remaining acquisition trials after reaching asymptote, supported by a significant main
2 effect of linear trend across trials, F(1,96) = 53.4, p<0.01, ηp = 0.36. No main effect involving anxiety (HA vs LA) or condition (ABA vs AAA) was significant, nor the interaction involving anxiety and condition (highest F = 0.4, p=0.53). Participants in the ABA condition showed a more significant decrease in expectancy ratings to CS+ after reaching asymptote than those in the AAA condition. This resulted in a relatively steeper linear trend in the AAA condition, confirmed by a significant interaction between condition and linear trend,
2 F(1,96) = 6.1, p=0.02, ηp = 0.06. This effect further interacted with trait anxiety, where the drop-off in expectancy ratings in the ABA conditions was more pronounced in the HA group, while LA participants showed similar acquisition trends in both conditions, F(1,96) = 11.9,
2 p<0.01, ηp = 0.11. Trait anxiety per se, had no effect on fear learning across the Acquisition phase, F(1,96) = 0.0001, p=0.99, n.s.
Figure 19B shows the mean change in log SCL during acquisition. Across anxiety groups and conditions, participants showed a steady decrease in skin conductance responding to CS+, confirmed by a significant main effect of linear trend across trials, F(1,96) = 18.0,
2 p<0.01, ηp = 0.17. Similar to the acquisition data in Experiment 1 in Chapter 5, the skin 176
A
100
80
60
40
20
Expectancy shock (%) of 0 A cquisition Extinction Test
HA_ABA (n=26)
HA_AAA (n=26)
LA_ABA (n=24)
LA_AAA (n=24)
B
0.02
0.01
0.00 Change in Log-SCL Change in
-0.01 A cquisition Extinction Test
Figure 19. Mean shock expectancy ratings (Top panel) and skin conductance level (SCL; Bottom panel) across acquisition, extinction and test trials. HA = High Anxious; LA = Low Anxious; ABA = ABA condition; AAA = AAA condition.
177 conductance data did not directly align with the expectancy data, as the level of responding to
CS+ decreased across trials. This could be accounted for by the habituation of skin conductance to the CS and US across trials. No effects of anxiety and condition, nor the interactions involving these factors, were found to significantly affect skin conductance during acquisition (highest F = 1.9, p=0.17).
Extinction
All participants showed a decrease in their expectancy ratings to the extinction cue
2 (Fig. 19A), supported by a significant main effect of linear trend, F(1,96) = 184.2, p<0.01, ηp
= 0.66. HA participants had higher averaged expectancy ratings to the extinction cue than LA participants averaged across conditions and trials, resulting in a significant main effect of
2 anxiety, F(1,96) = 10.1, p<0.01, ηp = 0.10. Averaged across anxiety groups and trials, participants in the AAA condition had higher overall expectancy ratings to the extinction cue than participants in the ABA condition, confirmed by a significant main effect of condition,
2 F(1,96) = 27.9, p<0.01, ηp = 0.23. The interaction involving anxiety and condition did not reach significance, F(1,96) = 1.5, p=0.22, n.s. Averaged across conditions, HA participants showed an overall less decrease in shock expectancies than their LA counterparts, confirmed
2 by a significant interaction involving anxiety and linear trend, F(1,96) = 5.3, p=0.02, ηp =
0.05. This suggested that HA participants showed an overall slower extinction than their LA counterparts. Condition had no reliable effect on the rate of extinction, F(1,96) = 1.0, p=0.32, n.s. However, the effect of anxiety on linear trend was somewhat more pronounced in the
AAA condition than in the ABA condition. That is, the effect of trait anxiety on fear extinction was mostly driven by the AAA condition, whereas the linear trends between the
HA and LA group in the ABA condition were fairly parallel. However, this 3-way interaction did not quite reach significance, F(1,96) = 3.5, p=0.06, n.s. 178
Unlike the expectancy measure, the skin conductance data displayed an irregular pattern (Fig. 19B), showing no reliable extinction effect across anxiety groups and conditions,
F(1,96) = 1.4, p=0.24, n.s No main effects regarding anxiety or conditions, nor any interactions involving linear trend were significant (highest F = 2.6, p=0.11).
Comparison of last acquisition trial to first extinction trial
A direct comparison between responding on the last acquisition and the first extinction trial was carried out, to examine whether a perceptual change in stimulus would affect the level of responding (i.e., generalisation decrement). Across anxiety groups and conditions, participants showed a significant decrease in expectancy ratings from the last trial of
2 acquisition to the first trial of extinction, F(1,96) = 13.6, p<0.01, ηp = 0.12. Participants in the
AAA condition had overall higher expectancy ratings across these 2 trials compared to those
2 in the ABA condition, F(1,96) = 16.1, p<0.01, ηp = 0.14. In fact, this effect was mainly driven by participants in the ABA condition having a significant decrease in expectancy ratings to cue B (GS) in the first extinction trial, while participants in the AAA condition
2 showed a continuity in responding between these two trials, F(1,96) = 5.0, p=0.03, ηp = 0.05.
The decrease in expectancy ratings between acquisition and extinction in the ABA condition suggested a generalisation decrement from stimulus A (CS+) to B (GS). No other effects reached significance (highest F = 1.9, p=0.17).
For skin conductance, participants showed no significant difference in responding between the last trial of acquisition and the first trial of extinction averaged across anxiety groups and conditions, F(1,96) = 0.22, p=0.64, n.s. No main effects of anxiety or condition reached significance, nor the interaction involving both anxiety and condition (highest F =
0.4, p=0.53). Interestingly, HA participants showed an increase in skin conductance responding from acquisition to extinction in the ABA condition, while LA participants showed no observable differences in responding between these 2 trials in the same condition. 179
While HA participants in the AAA condition showed a decrease in responding from acquisition to extinction, LA participants showed an opposite pattern in the same condition.
This led to a significant 3-way interaction involving anxiety groups, conditions and the
2 comparison between acquisition and extinction, F(1,96) = 4.2, p=0.04, ηp = 0.06. However, neither trait anxiety or condition per se had a significant impact on the responding to the last trial of acquisition and the first trial of extinction (highest F = 0.6, p=0.44).
Test
Averaged across anxiety groups and conditions, participants showed a decrease in expectancy ratings to the CS+ across the 3 test trials (Fig. 19A), confirmed by a significant
2 linear trend, F(1,96) = 52.4, p<0.01, ηp = 0.35. HA participants showed higher overall expectancy ratings to the test stimulus averaged across conditions and trials, supported by a
2 significant main effect of anxiety, F(1,96) = 13.5, p<0.01, ηp = 0.12. Averaged across anxiety groups and test trials, participants in the ABA condition had higher expectancy ratings to the test stimulus than those in the AAA condition, confirmed by a significant main effect of
2 condition, F(1,96) = 6.8, p=0.01, ηp = 0.07. Participants in the ABA condition showed a significantly faster decrease in expectancy ratings across the test trials than their counterparts in the AAA condition, resulting in a significant interaction between condition and linear trend,
2 F(1,96) = 11.4, p<0.01, ηp = 0.11. This suggests that participants who were presented with a
GS in extinction showed a faster decline in responding to the CS+ in test. No other interactions that involved trait anxiety reached significance (highest F = 0.7, p=0.40).
Similar to the expectancy measure, participants showed an overall decrease in skin conductance responding to the test stimulus across test trials averaged over anxiety groups and conditions (Fig. 19B), confirmed by a significant linear trend over the test trials, F(1,96)
2 = 5.3, p=0.02, ηp = 0.06. Averaged across anxiety groups, participants in the ABA condition had higher fear responding to the test stimulus, but this effect did not quite reach significance, 180
F(1,96) = 3.8, p=0.054, n.s. No main effect of anxiety on responding in test was observed,
F(1,96) = 0.25, n.s. Similar to the expectancy measure, participants in the ABA condition showed a more rapid decrease in responding across the test trials than those in the AAA condition, resulting in a significant interaction involving conditions and linear trend, F(1,96)
2 = 13.7, p<0.01, ηp = 0.12. No other interactions reached significance (highest F = 0.4, p=0.53).
Comparison of the last extinction trial to the first test trial
This critical contrast directly compared responding between the last extinction trial and the first test trial, to examine whether a change in perceptual feature would affect the responding to the CS+ in test. Averaged across anxiety groups and conditions, participants showed higher expectancy ratings on the first test trial compared to the last extinction trial,
2 F(1,96) = 30.6, p<0.01, ηp = 0.24. Overall, HA participants showed significantly higher expectancy ratings averaged across the two trials compared to their LA counterparts,
2 confirmed by a significant main effect of anxiety, F(1,96) = 13.7 , p<0.01, ηp = 0.13.
However, no observable effect of condition was found on the shock expectancies averaged across the two trials, F(1,96) = 0.4, p=0.53, n.s. Participants in the ABA condition showed a larger increase in shock expectancies from extinction to test, while those in the AAA condition showed similar expectancy ratings to the stimuli in the transition from extinction to test. This resulted in a significant interaction involving condition and the comparison between
2 extinction and test trials, F(1,96) = 35.5, p<0.01, ηp = 0.27. This suggests that participants who received standard extinction in the AAA condition continued to show low shock expectancies to the CS+, while those who received a GS in extinction showed heightened expectancy ratings when the CS+ was presented again in test. No other interactions reached significance (highest F = 0.07, p=0.79). 181
For skin conductance, participants showed an overall increase in responses from
2 extinction to test, F(1,96) = 12.3, p<0.01, ηp = 0.13. Averaged across anxiety groups, participants in the ABA condition showed larger responses averaged across the 2 trials
2 compared to those in the AAA condition, F(1,96) = 8.1, p<0.01, ηp = 0.10. This difference in responding across trials was mainly driven by the ABA condition – in fact, participants in the
ABA condition showed a large increase in responding from the last trial of extinction to the
2 first trial of test, confirmed by a significant interaction, F(1,96) = 8.2, p<0.01, ηp = 0.07. No other effects reached significance (highest F = 1.0, p=0.32).
Exploratory analyses
Although the above analyses showed that participants who were presented with a GS in extinction displayed more fear responding to the CS+ in test than those who received standard extinction with the CS+, it was unknown whether presenting a GS in extinction had any effect at all on the responding to CS+ in test. Therefore, we carried out an exploratory analysis comparing responding on the last acquisition trial to the first test trial, to examine whether presenting a GS in extinction had any effect on subsequent responding to CS+ in the
ABA condition.
For the expectancy measures, no main effect of trait anxiety on responding was evident across the two trials, F(1,48) = 1.1, p=0.3, n.s. Across anxiety groups, participants showed significantly lower expectancy ratings on the first test trial compared to the last
2 acquisition trial, F(1,48) = 8.3, p<0.01, ηp = 0.15. This effect was more pronounced among the LA participants compared to the HA participants, confirmed by a significant interaction between anxiety group and the comparison between the last acquisition trial and the first test
2 trial, F(1,48) = 4.4, p=0.04, ηp = 0.08.
Unlike the expectancy measures, participants showed significantly higher skin conductance responses on the first test trial compared to the last acquisition trial averaged 182
2 across conditions and anxiety groups, F(1,48) = 21.5, p<0.01, ηp = 0.15. No other effects reached significance (highest F = 0.4, p=0.53).
Discussion
Using a single-cue conditioning paradigm, the current study aimed to examine whether the usage of highly perceptually similar stimuli will increase the effectiveness of generalisation of extinction learning. In other words, whether presenting a GS that strongly resembles the CS+ in extinction would reduce fear responding to CS+ in the test phase.
Another aim of this study was to examine whether trait anxiety has any effect on fear extinction, and the generalisation of extinction.
Effect of conditions (ABA and AAA)
Averaged across anxiety groups, participants in the ABA condition showed a decrease in expectancy ratings from the last acquisition trial to the first extinction trial. In fact, participants in the ABA condition were presented with different stimuli between these two trials, A (CS+) and B (GS). The decrease in expectancies between these two trials suggests a generalisation decrement from CS+ to GS, and the lower averaged shock expectancies across extinction in the ABA condition can be attributed to persistence of the generalisation decrement from CS+ to GS seen at the beginning of extinction.
The critical finding was the increase in responding to the trained stimulus at test, in the
ABA, but not in the AAA condition. This pattern was observed in both the expectancy and skin conductance measures. In other words, participants who received the CS+ in extinction continued to show low level of fear responding to the CS+ in test, while those who received a
GS in extinction showed strong responding to the CS+ in test. This finding is consistent with previous studies that found a change in stimulus during extinction causes an increase in responding to the trained value in test (Vervliet et al., 2005, 2014), despite our attempt to use 183 a high level of perceptual similarity between the threat cue and extinction cue in the present study. However, participants in the ABA condition did show a slight decrease in shock expectancies to the CS+, confirmed by the exploratory analyses comparing responses on the last acquisition trial to the first test trial. This suggests that although extinction learning to a stimulus highly similar to the threat cue can slightly decrease fear to the original threat cue itself, it is not as effective as extinction learning to the threat cue itself.
Although participants in the ABA condition showed more fear responding to the CS+ on the first test trial, responding to the CS+ extinguished rapidly compared to the AAA control condition, evident in both expectancy and skin conductance measures. This is a novel finding since previous studies in generalisation of extinction analyzed the data averaged over test trials instead of on a trial-by-trial basis (Vervliet et al., 2005, 2014), and hence could not observe any differences in the rate of extinction in responding during test. There are two possible reasons for this finding. First, because participants in the ABA condition had higher shock expectancies to the CS+ on the first test trials compared to the AAA condition, there was more room for responding to CS+ to drop. Secondly, after experiencing a non-reinforced
CS+ on the first test trial, participants in the ABA condition may had a stronger belief that the
CS+ no longer predicted shock given that they had already experienced a similar exemplar that did not predict shock.
Effect of trait anxiety
Across both conditions, trait anxious individuals showed slower fear extinction. This is consistent with the idea that anxious individuals show resistance in fear extinction (Dibbets et al., 2014; Gazendam et al., 2013). Although high anxious individuals tended to show slower fear extinction in the control AAA condition than those in the ABA condition, suggesting that high anxious individuals showed slower fear extinction to the original CS+ 184 than to a novel GS, this pattern was not statistically supported. Furthermore, these effects were only evident in the expectancy measure but not the skin conductance measure.
Trait anxiety did not interact with condition when comparing responding between the last acquisition trial and the first extinction trial in both the expectancy and skin conductance measures. Specifically, high anxious individuals in the ABA condition had a similar difference in responding from CS+ to GS compared to low anxious individuals in the same condition. This suggests that trait anxiety has no effect on the generalisation decrement from the CS+ to GS. This was an unexpected finding since the GS was expected to have an ambiguous threat value, given that it was novel and highly similar to the CS+. It was possible that high anxious individuals had already inferred some verbal rules during acquisition, such as ‘the bluer/greener the circle, the more/less likely I would get shocked’, that may have decreased the ambiguity of the situation. Although one may argue that it was unlikely that participants were able to identify a rule since they had only experienced one stimulus during training (see Wong & Lovibond, 2017), Lee et al. (2018) found that approximately 49% of participants came up with a linear rule across two studies that used the same colour dimension with a single-cue conditioning procedure.
Presenting a GS in extinction slightly decreased participants’ shock expectancies to the CS+ on the first test trial, but this effect was more pronounced in the low anxious group than the high anxious group. This was presumably due to trait anxious individuals possessing a stronger memory of CS+ (see Hagenaars, 2012; Soeter & Kindt, 2013), which in turn triggered a stronger fear response to CS+ (see Wegerer, Blechert, Kerschbaum & Wihelm,
2013).
Abnormalities in acquisition
Surprisingly, a significant 3-way interaction was observed in the expectancy ratings during acquisition. All participants showed a drop-off in expectancy ratings to the CS+ after 185 shock expectancies had reached asymptote, which may have contributed to the significant interaction. The drop-off in shock expectancies was likely caused by the partial reinforcement of CS+, which induced uncertainty of shock. High anxious individuals in the ABA condition showed a significantly larger decrease in expectancy ratings than those in the AAA condition, while conditions did not affect the magnitude of the decrease in expectancy ratings to CS+ among low anxious individuals. However, it was impossible for condition to affect acquisition since the acquisition phases in both conditions were identical. A possible explanation for this is that the partial reinforcement of CS+ increased the uncertainty of its shock predictiveness, leading to a large variability in expectancy ratings to it, increasing the chance of the occurrence of a Type I error.
In summary, the current study replicated the finding that a reduction in fear responding to the CS+ is less effective if extinction is carried out with a GS than with the original CS+.
Consistent with the trait anxiety literature, high anxious individuals showed resistance to fear extinction across both conditions.
On a positive note, participants who were presented with a GS during extinction showed a rapid reduction in fear responding to the CS+ during test. This suggests that although extinction learning to GS does not effectively generalise to the CS+ initially, it is beneficial to the subsequent exposure to non-reinforced CS+ presentations.
However, one important research question remains: despite exposure to a GS does not effectively extinguishing fear responding to the CS+, can this extinction knowledge be effectively generalised to another novel GS? To our knowledge, only two studies have addressed this question. Vervliet, Vansteenwegen and Eelen (2004) assigned participants to one of the two groups: ABA and AAB. Similar to the current study, the ABA group received a novel GS in extinction and the CS+ in test, while the AAB group received the CS+ in extinction and a novel GS in test. They found that participants in the ABA group showed 186 heightened fear responding to the CS+ in test, but those in the AAB group showed relatively low responding to a novel GS in test. Although these results suggest that extinguishing the
CS+ directly is more effective in reducing fear responding to a similar and novel GS, the study did not directly address whether presenting a GS in extinction would reduce fear to another novel GS (i.e., an ABC paradigm). A more recent study conducted by Zbozinek and
Craske (2018) compared how extinction with a previously reinforced CS+, a GS, or a variety of different GSs affected responding to the CS+ and other GSs. One hypothesis the authors tested was whether extinction learning to the previously reinforced CS+ (Extinction _CS group) and extinction with multiple GSs (Extinction_Multiple group) would be more effective in lowering conditioned fear to a novel GS than presenting a single GS during extinction
(Extinction_Singular group). The authors did not find any evidence to support this hypothesis; however, since the test stimuli were presented in a peusdo-randomized order with the novel
GS always presented first, the authors argued that the orienting responses to this novel GS may have attenuated any group differences.
Therefore, the following study was designed to specifically examine how effective extinction with a GS is in reducing conditioned fear to another novel GS by using an ABC paradigm, compared to an AAC control group. In other words, the following study sought to examine whether exposure to a GS (i.e., extinction with stimulus B) can be generalised to a novel cue (stimulus C) that is similar to both the CS+ (stimulus A) and the extinguished GS
(stimulus B). Furthermore, the effect of trait anxiety on the generalisation of extinction was also investigated.
Experiment 5
Method
Experiment 5 only differed from Experiment 4 in the following aspects. 187
Participants
Thirty-three undergraduate students were recruited in each group (anxiety groups and conditions), resulting in a total of 132 participants. Thirty-one participants were excluded (see
Results for more detail). The final sample comprised 101 participants (74 females) with a mean age of 20.3 (SD = 2.8).
Apparatus and materials
The additional stimulus C lay on the same blue-green dimension as stimuli A and B
(Fig. 20). It was a blue circle with a HSV value of 0.512. The saturation and brightness for stimulus C were held the same as A and B at 1 and 0.75 respectively. The radius of stimulus C was 5.75cm.
A B C Figure 20. Stimulus A served as CS+ in both AAC and ABC conditions. Stimulus A was presented in extinction in the AAC condition, while stimulus B was presented in extinction in the ABC condition. Stimulus C served as the test stimulus in both conditions.
Procedure
The acquisition and extinction phases in the current study were exactly the same as
Experiment 4. During test, both conditions received 3 trials of a novel GS (Stimulus C) without any reinforcement (Table 7).
188
Table 7. Design of current study Phase/Condition Acquisition Extinction Test
ABC A+ (6) B- (8) C- (3)
AAC A- (2) A- (8)
Note. + indicates shock presentation; - indicates shock omission; numbers in brackets indicate the number of trials of that type in each phase. Exclusion of participants
In the ABC condition, 15 participants were excluded because they failed to report seeing exactly 3 different colours in the post-experimental questionnaire. Eight participants in the AAC condition failed to correctly report there were 2 colours presented in the study, and hence were excluded. Five participants (1 in the HA group and 4 in the LA group) were excluded because they did not meet the acquisition criteria. Additionally, 3 participants were excluded because their skin conductance was not measured properly due to technical problems. Altogether, 31 participants were excluded, leaving 101 participants in the final sample.
Missing data
Missing data occurred on 1.2% and 1.3% of all trials in the HA and LA group respectively. Missing data were replaced with the average ratings during that particular trial type across all participants within the condition.
Results
Anxiety groups and shock intensities
The mean DASS-anxiety scores were 15.8 and 3.9 for the HA and LA groups respectively. The mean shock intensities for the HA and LA group were 2.2mA and 2.3 mA respectively, and there was no evidence that there was any anxiety difference in the tolerance of electric shock, F(1,97) = 0.3, p=0.59, n.s. 189
Acquisition
Figure 21A shows the mean shock expectancy ratings during acquisition. Across anxiety groups and conditions, participants showed a sharp increase in expectancy ratings to
CS+, and expectancies dropped off after reaching asymptote at around 80%. The drop-off in expectancy ratings was presumably caused by an uncertainty of shock predictiveness, induced by the partial reinforcement of CS+. This pattern was supported by a significant main effect
2 of linear trend, F(1,97) = 61.2, p<0.01, ηp = 0.39. Unlike Experiment 4, no main effects of anxiety or condition, or interactions involving either factors and linear trend reached significance (highest F = 1.2, p=0.28).
Figure 21B shows the mean change in log SCL during acquisition. Unlike the expectancy measure, participants showed a reduction in skin conductance responding to CS+ across anxiety groups and conditions, confirmed by a significant main effect of linear trend,
2 F(1,97) = 10.0, p<0.01, ηp = 0.10. Across conditions, the averaged responding to CS+ was higher in the HA group than in the LA group across trials, but this anxiety difference on fear acquisition did not reach significance, F(1,97) = 2.5, p=0.12, n.s. The anxiety difference in skin conductance was seemingly larger in the AAC condition than in the ABC condition, but this difference did not reach significance, F(1,97) = 2.4, p=0.13, n.s. No other main effects or interactions reached significance (highest F = 1.2, p=0.28).
Extinction
Across anxiety groups and conditions, participants showed a decrease in expectancy ratings to the extinction cue across trials (Fig. 21A), supported by a significant main effect of
2 linear trend, F(1,97) = 86.2, p<0.01, ηp = 0.47. On average, HA participants had higher expectancy ratings to the extinction cue than their LA counterparts, resulting in a significant
2 main effect of anxiety, F(1,97) = 5.3, p=0.02, ηp = 0.05. Participants in the AAC condition also showed significantly overall higher expectancy ratings to the extinction cue than those in 190
A
100
80
60
40
20
Expectancy shock (%) of 0 A cquisition Extinction Test
HA_ABC (n=25)
HA_AAC (n=28)
LA_ABC (n=25)
LA_AAC (n=23) B
0.03
0.02
0.01
0.00
-0.01
-0.02 Change in Log-SCL Change in
-0.03 A cquisition Extinction Test
Figure 21. Mean shock expectancy ratings (Top panel) and skin conductance level (SCL; Bottom panel) across acquisition, extinction and test trials. HA = High Anxious; LA = Low Anxious; ABC = ABC condition; AAC = AAC condition.
191
2 the ABC condition, F(1,97) = 62.6, p<0.01, ηp = 0.39. This difference was presumably due to the persistence of generalisation decrement from A (CS+) to B (GS) in the ABC condition seen at the beginning of extinction. Averaged across anxiety groups, participants in the AAC condition showed a relatively faster decline in expectancies across trials, resulting in a
2 significant interaction that involved condition and linear trend, F(1,97) = 5.0, p=0.03, ηp =
0.05. In other words, participants showed a faster rate of extinction when presented with the
CS+ (A) compared to a novel GS (B). This difference in extinction rate between conditions was more pronounced in the LA group than the HA group – in fact, HA participants showed a significantly slower decrease in shock expectancies than the LA group in the AAC conditions, whereas the linear trends between the HA and LA group in the ABC condition were fairly parallel. This pattern was confirmed by a 3-way interaction involving anxiety, condition and
2 linear trend, F(1,97) = 5.2, p=0.03, ηp = 0.05. No other interactions reached significance
(highest F = 1.5, p=0.22).
In contrast to the expectancy measure, participants showed no reliable extinction effect in skin conductance responding (Fig. 21B) across anxiety groups and conditions, F(1,97) =
0.7, p=0.41, n.s. In fact, the skin conductance measure showed a highly irregular pattern.
Across conditions, HA participants had an overall higher responding across trials than the LA participants, but this difference did not reach significance, F(1,97) = 2.8, p=0.10, n.s. No other main effects and interactions reached significance (highest F = 2.1, p=0.15).
Comparison of the last acquisition trial to the first extinction trial
Similar to Experiment 4, this contrast directly compared responding on the last acquisition trial and the first extinction trial, to examine whether a perceptual change in stimulus would affect the level of responding (i.e., generalisation decrement). Averaged across anxiety groups and conditions, participants showed a significant drop in expectancy
2 ratings from acquisition to extinction, F(1,97) = 44.7, p<0.01, ηp = 0.32. Participants in the 192
ABC condition also had lower expectancy ratings averaged across anxiety groups and the two
2 trials, F(1,97) = 33.3, p<0.01, ηp = 0.26. In fact, these two effects were mainly driven by a significant decrease in expectancy ratings from acquisition to extinction from participants in the ABC condition, supported by a significant interaction involving conditions and the
2 comparison between acquisition and extinction, F(1,97) = 35.2, p<0.01, ηp = 0.27. This suggests that participants in the ABC condition showed a generalisation decrement due to a change of stimulus between acquisition and extinction. This generalisation decrement effect appeared to be larger among the LA participants than the HA participants. In other words, HA participants in the ABC condition showed stronger generalisation from the CS+ to GS than their LA counterpart in the same condition; however, this difference did not quite reach significance, F(1,97) = 3.0, p=0.09, n.s. No other effects were significant (highest F = 2.1, p=0.15).
For skin conductance, there was no reliable difference in responding between the last acquisition trial and the first extinction trial averaged across anxiety groups and conditions,
F(1,97) = 0.09, p=0.77, n.s. Averaged across anxiety groups and trials, participants in the
ABC condition showed no reliable difference in their responding compared to those in the
AAC condition, F(1,97) = 0.06, p=0.81, n.s. HA participants showed higher skin conductance responding to the LA participants averaged across conditions and trials, but this difference did not reach significance, F(1,97) = 2.1, p=0.15, n.s. No other interactions reached significance
(highest F = 0.1, p=0.75), suggesting that neither trait anxiety or condition had any effect on the responding between acquisition and extinction.
Test
HA participants had higher expectancy ratings to the novel test stimulus (C) than LA participants averaged across condition and trials (Fig. 21A), confirmed by a main effect of
2 anxiety, F(1,97) = 9.5, p<0.01, ηp = 0.09. Averaged across anxiety groups and trials, 193 participants in the ABC condition showed higher expectancy ratings to the test stimulus than
2 those in the AAC condition, resulting in a main effect of condition, F(1,97) = 4.2, p=0.04, ηp
= 0.04. Participants showed a steady decline in expectancy ratings to the novel test stimulus across trials averaged across anxiety groups and conditions, supported by a main effect of
2 linear trend, F(1,97) = 73.3, p<0.01, ηp = 0.43, due to the test stimulus being presented without any reinforcement. Participants in the ABC condition showed a more rapid decrease in expectancy ratings across the test trials, leading to a significant interaction that involved
2 condition and linear trend, F(1,97) = 10.0, p<0.01, ηp = 0.09. This suggests that participants who experienced a GS in extinction (ABC condition) showed faster extinction learning to a novel GS than those who experienced standard extinction with the original threat cue (AAC condition). No other interactions reached significance (highest F = 1.2, p=0.28).
For skin conductance (Fig. 21B), participants showed a general decline in responding across the test trials averaged across anxiety groups and conditions, confirmed by a significant
2 main effect of linear trend, F(1,97) = 11.9, p<0.01, ηp = 0.13. Similar to the expectancy measure, participants in the ABC condition showed a more rapid decrease in responding than those in the AAC condition, although this difference did not reach significance, F(1,97) =
2.9, p=0.09. n.s. No other effects reached significance (highest F = 0.43, p=0.51).
Comparison of last extinction trial to first test trial
This critical contrast directly compared responding between the last extinction trial and the first test trial, to examine whether a change in perceptual feature would affect the responding to a novel GS in test. Averaged across trials and conditions, HA participants showed higher expectancy ratings than their LA counterparts, confirmed by a significant main
2 effect of anxiety, F(1,97) = 8.7, p<0.01, ηp = 0.08. Participants also showed higher expectancy ratings to the first test trial than the last extinction trial averaged across conditions
2 and anxiety groups, F(1,97) = 47.2, p<0.01, ηp = 0.33. This difference in responding 194 between extinction and test was mainly driven by participants in the ABC condition having a significant increase in shock expectancies from extinction to test, while those in the AAC condition showed similar expectancy ratings between extinction and test, F(1,97) = 29.9,
2 p<0.01, ηp = 0.24. This critical contrast suggests that participants who were presented with a
GS in extinction (ABC condition) showed a high level of fear responding to a novel GS in the first test trial, while those who received extinction with the original threat cue (AAC condition) responded to the novel GS similarly to an extinguished cue. No other effects reached significance (highest F = 0.3, p=0.59).
Averaged across anxiety groups and conditions, participants showed an overall increase in skin conductance responding from extinction to test, however, this difference between trials did not quite reach significance, F(1,97) = 3.6, p=0.06, n.s. No main effects of anxiety or condition reached significance, nor did any interactions involving linear trend
(highest F = 2.2, p=0.14).
Exploratory analyses
Similar to Experiment 4, an exploratory analysis was carried out to examine if presenting a GS in extinction has any effect at all on responding to another novel GS, by comparing responding between the last acquisition trial and the first test trial between both conditions.
For the expectancy measure, participants showed lower expectancy ratings on the first test trial compared to the last acquisition trial averaged across anxiety and condition, F(1,97)
2 = 46.5, p<0.01, ηp = 0.32. HA participants showed higher expectancy ratings than LA participants averaged across condition and trials; however this main effect of trait anxiety did not quite reach significance, F(1,97) = 3.5, p=0.06, n.s. Averaged across anxiety and trials, participants in the ABC condition showed significantly higher expectancy ratings than the
2 AAC condition, confirmed by a main effect of condition, F(1,97) = 4.4, p=0.04, ηp = 0.04. 195
This effect was mainly driven by participants in the ABC condition exhibiting significantly higher expectancy ratings in the first test trial than the last acquisition trial compared to the
2 AAC condition, F(1,97) = 13.1, p<0.01, ηp = 0.12. No other effects reached significance
(highest F = 1.8, p=0.18). Given the significant interaction between condition and the contrast comparing the last acquisition trial and the first test trial, follow-up analyses were carried out for each condition. In the AAC condition, participants showed a significant decrease in expectancy ratings from the last acquisition trial to the first test trial, F(1,49) = 51.3, p<0.01,
2 ηp = 0.51. Although participants in the ABC condition showed a significantly smaller decrease in expectancy ratings between the two trials, this decrease in expectancy ratings was
2 significant, F(1,48) = 5.5, p=0.02, ηp = 0.10.
For the skin conductance measure, responding on the first test trial was significantly higher than the last acquisition trial averaged across conditions and anxiety groups, F(1,97) =
2 20.7, p<0.01, ηp = 0.15. No other effects reached significance (highest F = 1.8, p=0.18).
Discussion
The current study sought to examine whether presenting a GS in extinction is as effective as presenting the original CS+ in extinction to reduce conditioned fear to a novel
GS. The major finding of the current study was that participants were able to generalise extinction learning to the CS+ to a novel GS (AAC effect). However, extinction learning to a
GS did not generalise to another novel GS as well as extinction learning to the CS+ (ABC effect), evident by the heightened responses to the novel GS in test.
During extinction, high anxious individuals again showed higher expectancy ratings than their low anxious counterparts across conditions and trials, due to two reasons. First, high anxious individuals showed slower extinction learning than low anxious individuals when they were presented with the original threat cue (AAC condition). This finding can be attributed to the difference in ambiguity level between the two stimuli. A CS+ in extinction is 196 thought to be ambiguous because it has two meanings: the CS-shock association and the newly formed CS-no shock association (Bouton, 1993, 2007). The GS can also be considered to have an ambiguous threat value because of its novelty and its perceptual similarity with the threat cue. However, it had never been reinforced before and the accumulation of the non- reinforced trials of GS suggested its safety, hence rendering its threat value relatively less ambiguous than the CS+ in extinction. Therefore, the slow extinction learning to the relatively ambiguous CS+ among high anxious individuals is consistent with the idea that trait anxious individuals show a bias in threat appraisal to ambiguity. Secondly, the finding that high anxious individuals had higher averaged expectancy ratings during extinction across condition could also be attributed to anxious individuals having stronger fear generalisation from the
CS+ to the GS. However, this claim was not statistically supported.
In test, high anxious individuals continued to show higher shock expectancies averaged across trials. This was primarily due to their slow extinction learning, especially in the AAC condition. Interestingly, despite participants who received a GS in extinction showing heightened shock expectancies on the first test trial, their expectancies declined more rapidly than those who received the CS in extinction. This finding was similar to the results in
Experiment 4, where participants also showed a faster extinction rate in test when they were presented with a GS in extinction. This rapid decline in expectancy ratings observed in the
ABC condition could possibly be due to participants having experienced an extra exemplar
(stimulus B) in extinction, hence they were more confident to extrapolate extinction to the novel GS, especially when they had learnt that the novel GS did not predict shock on the first test trial.
General Discussion
Across two studies using a single-cue fear conditioning paradigm, it was found that direct extinction with the CS+ was the most effective way to reduce conditioned fear to the 197
CS+ itself, or to a novel GS. In other words, a weak generalisation of extinction effect was found, since participants who were extinguished with a GS showed heightened responses to the CS+ or to a novel GS on the initial test trial. Interestingly, both studies found that participants who were exposed to a GS during extinction showed a significantly faster decline in responding to the test stimuli compared to those who received the CS+ during extinction.
Trait anxiety was found to slow down the rate of extinction learning. The extinction rate among trait anxious individuals also differed between conditions. Generally speaking, those who received the CS+ in extinction showed slower extinction learning than to those who received a GS in extinction. These findings will be discussed below, followed by theoretical considerations.
Effects of conditions (ABA and ABC)
The current results replicated findings that extinction with the CS+ is the most effective way to reduce conditioned fear to either the CS+ itself or another novel GS (Vervliet et al., 2004, 2005; Vervoort et al., 2014). In contrast, although extinction learning to a GS did slightly decrease the shock expectancies to the CS+ or to another GS in the first test trial, it was not as effective as direct extinction learning to the CS+.
Interestingly, both studies showed that when participants received a GS in extinction, they showed a sharp decrease in expectancy ratings (ABA & ABC conditions) and in skin conductance responses (ABA condition) in test. This was possibly due to participants in these conditions having simply experienced an extra exemplar in extinction, hence giving them more confidence to extrapolate that the test stimulus was safe, especially after they had experienced the first non-reinforced test trial. Practically, this means presenting a GS in exposure-based therapies may not effectively reduce fear to the original fear cue or a similar threat cue initially, but may enhance the subsequent inhibitory learning to these cues. 198
A plausible explanation was that responding to the test stimulus in both AAA and
AAC conditions had reached the floor, hence rendering little room for a sharp decline in responding. Alternatively, the rapid decline in responding after receiving a novel GS in extinction was conceptually parallel to a recently found phenomenon – novelty-facilitated extinction. In a study conducted by Dunsmoor, Campese, Ceceli, LeDoux and Phelps (2015), participants were first trained to discriminate between two facial CSs. In the following extinction phase, one group of participants received standard fear extinction with a non- reinforced CS+ (EXT group) while another group of participants received a CS-novel tone pairing in extinction (NFE group). Twenty-four hours after extinction, the facial CSs were presented and participants’ spontaneous recovery of conditioned fear to the stimuli was measured. The NFE group showed less conditioned fear to the CS+ in the spontaneous recovery test compared to the EXT control group. At face value, this seems to contradict the current results; however, the test data analyzed were averaged across early test trials. That is, it is possible that the lower spontaneous recovery observed in the NFE group may have attributed to a rapid decline in conditioned fear to the CS+, similar to the current results.
Similarly, Lucas, Luck and Lipp (2018) found that the NFE group showed less conditioned fear in a reinstatement test when compared to the EXT control group. These studies suggested that the involvement of a novel stimulus in extinction leads to a reduction in the return of fear; combined with the current findings, this decrease in the return of fear is possibly due to a rapid decrease in fear responding in test after the first non-reinforced CS+ trial.
Effect of trait anxiety
Across both studies, generalisation decrement was observed between the last acquisition trial and the first extinction trial in the ABA and ABC conditions. The high anxious individuals showed stronger generalisation from the acquisition trial to the extinction trial in both conditions, however, this finding was not statistically confirmed. High anxious 199 individuals showed stronger generalisation primarily because the novel GS was ambiguous, however, the trait anxiety difference did not quite reach significance presumably because we did not assess if rule formation took place during the study. That is, some participants may have come up with relational rules during acquisition (e.g., the bluer the circle is, the more likely it predicts shock) that potentially decreased the ambiguity of the situation. Therefore, any trait anxiety effect on generalisation may have been attenuated because we were not able to separate the ones who had inferred a rule from those who were unable to identify any clear rules.
A trait anxiety effect on extinction was generally observed when participants received standard fear extinction (AAA and AAC conditions), which is consistent with the idea that trait anxious individuals show resistance in fear extinction (Dibbets et al., 2014; Gazendam et al., 2013). However, high anxious individuals who received a novel GS in extinction (ABA and ABC conditions) showed a similar extinction learning rate to their low anxious counterparts. As previously discussed, this was potentially due to the difference in the level of ambiguity of the extinction stimulus. In the ABA and ABC conditions, the novel GS was relatively less ambiguous than the CS+ in the AAA and AAC conditions. Although the initial presentation of the GS may be highly ambiguous, it had never been previously paired with a shock. In addition, the subsequent non-reinforced trials of the GS presumably rendered it relatively less ambiguous. Conversely, the CS+ was relatively more ambiguous since it signaled both threat and safety after extinction learning. This finding emphasizes the importance of ambiguity in fear learning among trait anxious individuals, consistent with
Experiments 1, 2 and 3 in the current thesis. While some studies have shown that trait anxious individuals display a deficit in inhibitory learning to unambiguous safety cues (e.g.,
Gazendam et al., 2013; Torrents-Rodas et al., 2013), there is some evidence that trait anxious individuals show impaired safety learning only when they perceive the situation as ambiguous
(e.g., Baas et al., 2008; Chan & Lovibond, 1996; Chen & Lovibond, 2016). The current 200 finding arguably supports the latter notion since high anxious individuals did not show impaired inhibitory learning to a relatively less ambiguous GS; however, they showed impaired inhibitory learning to the relatively more ambiguous CS+ in extinction.
Interestingly, studies that found the novelty-facilitated extinction effect (Dunsmoor et al., 2015; Lucas et al. 2018) provide some evidence to suggest that extinguishing the threat cue renders it ambiguous. Both studies found a positive correlation between self-reported IU and the magnitude of return of fear, but this effect was only observed in participants who received standard fear extinction with CS+. As individuals with high IU are thought to be vulnerable to ambiguous situations and interpret ambiguity in a negative way (e.g., Chen &
Lovibond, 2016; Dugas et al., 2005; Koerner & Dugas, 2008), this suggests that the extinguished threat cue is more ambiguous compared to when a novel cue is involved in extinction, complementing the current finding that trait anxious individuals showed resistance in fear extinction only to the CS+ but not to a GS.
No trait anxiety difference was observed in the rapid decline in responding to the test stimuli (CS+ or another novel GS) among the ABA and ABC conditions. This could be accounted for by individuals having received a GS in extinction, effectively providing them with an extra exemplar that signaled safety. This increased their confidence to infer the test stimuli to be safe, especially after the first non-reinforced test trial. In other words, this experience with an extra safe exemplar may have reduced the ambiguity of the stimuli in test, hence high anxious individuals showed a rapid decline in fear responding in test similar to the low anxious group.
Associative and cognitive accounts
The critical finding that participants showed heightened fear responses to the test stimulus in the ABA and ABC conditions can be accommodated by associative accounts. In the ABA condition, only the elements shared between the CS+ and GS gained inhibitory 201 strength, while elements in the CS+ not shared with the GS retained their excitatory properties. Therefore, these unextinguished excitatory elements in the CS+ were still able to trigger conditioned fear. Similarly, elements that were uniquely shared between the CS+ and the GS in test (C) remained their excitatory properties in the ABC condition, and hence were able to trigger a higher magnitude of conditioned fear compared to the AAC condition.
The associative accounts predict that increasing the perceptual similarity between the
CS+ and GS will increase the number of shared elements, hence more shared elements will gain inhibitory strength. This will theoretically leave fewer excitatory elements that are possessed by the CS+ but not the GS, hence improving the generalisation of extinction effect.
Intriguingly, the current studies found a weak effect of generalisation of extinction despite using CS+ and GSs that were highly similar. The associative accounts are still able to explain this surprising finding via conditioned inhibition. When the GS was first presented, the novel elements uniquely possessed by the GS had no associative strength. Across extinction trials, these novel elements gradually gained inhibitory strength and acted as conditioned inhibitors, protecting the excitatory elements shared between the CS+ and the GS from extinction. In the case of the ABA condition, the CS+ in test triggered conditioned fear because of the protected excitatory, shared elements, in addition to the unextinguished excitatory elements uniquely possessed by the CS+. In the ABC condition, the novel GS (C) was able to trigger a relatively strong CR because of the protected excitatory elements shared among the CS+ and the two
GSs, in addition to the elements uniquely shared between the CS+ and the novel GS.
However, the associative accounts provide no clear explanation for the rapid decline in responding to the test stimuli seen in the ABA and ABC conditions.
According to the cognitive account, the increase in fear responding to the test stimuli in the ABA and ABC conditions was due to participants not being willing to extrapolate safety learning from the GS to the test stimuli. This was presumably because the CS+ and GS 202 were merely connected via perceptual similarity. Participants may not be willing to extrapolate the safety properties of GS to the test stimuli based on this perceptual connection, especially on the first test trial. In contrast, participants who received the CS+ in extinction
(AAA & AAC conditions) were more willing to extrapolate safety learning of the CS+ to the same stimulus (AAA) or to a novel stimulus (AAC), since the only cue that previously signaled threat no longer predicted shock. The cognitive account can also explain the rapid decline in fear responding to the test stimuli in the ABA and ABC conditions. After experiencing the first non-reinforced test trial, the belief that the test stimuli no longer signaled threat was strengthened. This belief was stronger in the ABA and ABC conditions since participants had received a GS in extinction, which acted as an extra piece of evidence that strengthen the belief that the test stimuli were safe. Therefore, participants showed a rapid decrease in fear responding in test because they had stronger beliefs that the test stimuli predicted safety.
Limitations
One limitation of the current studies is the imbalance between gender among the high anxious groups. The high anxious groups comprised of 73% and 78% of female participants in Experiment 4 and 5 respectively. Numerous empirical studies have found that hormonal status fluctuates across the menstrual cycle, which directly affects extinction learning to a threat cue. Specifically, a low level of estradiol and progesterone impairs fear extinction in females (e.g., Glover et al., 2012; Graham & Daher, 2016; Lebron-Milad et al., 2012; Merz &
Wolf, 2016). Given the majority of high anxious participants were females in the current studies and their menstrual cycle was not assessed, it was possible the finding that trait anxious individuals showed impaired fear extinction may have been confounded by the hormonal effect. 203
Another limitation is that we did not assess if participants derived any inferred rules throughout the study. When a rule was identified, the experimental configuration may have become less ambiguous and trait anxious participants may not have exhibited impaired fear extinction. In contrast, if they failed to identify any clear rules, the experimental configuration become more ambiguous and they may show slower inhibitory learning. Therefore, trait anxious individuals may differ in their rate of fear extinction depending on whether they came up with a verbalized rule or not.
Chapter summary
The current chapter presented two experiments to examine whether using two highly similar stimuli will improve the effect of generalisation of extinction to the original threat cue or to a novel cue. In other words, the experiments examined whether presenting with a GS highly similar to the CS+ in extinction will effectively reduce conditioned fear to the CS+ or another novel cue. The effect of trait anxiety on the generalisation of extinction was also investigated. The results showed that extinction to the CS+ was more effective to reduce conditioned fear to the CS+ itself or to a novel GS compared to the extinction of a GS. This finding is consistent with past studies (Vervliet et al., 2004, 2005; Vervoort et al., 2014).
Interestingly, participants who received a GS in extinction showed a sharp decrease in responding to the test stimuli across test trials. This was primarily because participants had received a different exemplar in extinction; after the first non-reinforced test trial, they were more willing to extrapolate that the test stimuli were safe, hence resulting in a rapid decline in responding. Trait anxiety was found to slow down extinction learning to the CS+, but not to extinction learning to a GS. This was interpreted as being due to the extinguished CS+ being more ambiguous than the GS, since the extinguished CS+ was associated with a shock or no shock, while the GS were never followed by a shock.
204
Chapter 8:
General Discussion
205
Excessive generalisation of fear has been recently thought to be a pathogenic marker of anxiety disorders (Kaczkurkin et al., 2017; Lissek et al., 2010, 2014). Given that trait anxiety is thought to be a vulnerability factor of anxiety disorder, the current thesis aimed to investigate the effect of trait anxiety on fear generalisation using a ‘weak’ conditioning task, and the effect of trait anxiety on fear learning in general. Another aim of the current thesis was to examine whether extinction learning would effectively generalise to the original threat cue or another novel cue when relevant stimuli were highly similar in perceptual terms. The effect of trait anxiety on generalisation of extinction was also investigated. The final aim was to examine how the current findings could be accounted for by the associative and the cognitive accounts. The following will present a summary of the key findings, followed by how the present work contributes to the literatures on i) the effect of trait anxiety on fear learning in general, in particular generalisation, and ii) theoretical aspects of generalisation in humans.
Key empirical findings
The first empirical chapter (Chapter 5) examined fear generalisation along an arbitrary, spatial dimension, using a single-cue fear conditioning procedure. Both high and low anxious groups showed a peaked gradient in their overall generalisation pattern, with the highest responding to CS+. While the low anxious group showed a pronounced decline in responding to stimuli that were perceptually less similar to the CS+, the high anxious group showed a relatively flat generalisation gradient. These peaked gradients at the group level aligned with the predictions of associative accounts that response strength is a function of physical similarity of novel stimuli to the CS+.
However, participants reported using a variety of rules during the conditioning task.
After categorizing participants into subgroups according to their reported inferred rules, distinctive generalisation patterns were observed. Peaked gradients were observed in the 206
Similarity subgroup, consisting of participants who inferred a rule that the likelihood of shock was predicted by the perceptual similarity of stimuli to the CS+. Linear gradients were found in the Linear subgroups, consisting of participants who inferred a linear relational rule that if the more the black dot was located to the right of the CS+, the more likely the stimulus predicted shock. The remaining participants reported not being able to identify any clear rules
(No rule subgroup). They showed a relatively flat generalisation gradient with responding peaked at CS+. No reliable trait anxiety differences were found in the subgroups that identified a clear rule (i.e., Similarity and Linear subgroups); however, trait anxious individuals in the No rule subgroup showed stronger generalisation of fear than their non- anxious counterparts. The results suggest that without a relational rule to guide fear generalisation (No rule subgroup), the perceived threat ambiguity of novel generalisation stimuli increased, leading to over-generalisation of fear among high anxious individuals.
Similar to past studies, the formation of rules about the basis for generalisation and the fact that such rules predict patterns of fear generalisation suggest that cognitive processes are heavily involved in fear generalisation in humans. (Ahmed & Lovibond, 2018; Lee et al.,
2018; Shanks & Darby, 1998; Wong & Lovibond, 2017). This motivated the research aim of studying the generalisation of fear to novel objects beyond perceptual features (Chapter 6). In
Experiment 2 and 3 we adopted a categorical induction paradigm and tested it within a fear conditioning framework. Exemplars from one category (e.g., breakfast foods) were paired with electric shock while exemplars from another category (e.g., bakery foods) were not reinforced. In test, novel breakfast and bakery items were presented to measure how acquired fear to the training exemplars generalised to these novel items. Based on the finding that trait anxious individuals only showed over-generalisation of fear under conditions of threat ambiguity in Experiment 1, the threat level of certain exemplars was manipulated to be highly ambiguous (i.e., items that were consistent with both threat and safe categories). The results showed that conditioned fear selectively transferred to novel exemplars that belonged to the 207 threat category, but not to novel exemplars that belonged to the safe category. No trait anxiety effects were observed in the generalisation of fear to these exemplars that had clear threat value. However, trait anxious individuals showed an increase in fear responding to items that had ambiguous threat value.
The research focus of Experiments 4 and 5 expanded from the generalisation of acquired fear to the generalisation of extinction learning. The latter research focus is as important as the former, since further understanding of how extinction learning generalises from one object to another provides important insight for exposure-based therapies. This chapter aimed to test whether presenting a stimulus that was perceptually similar to CS+ would be able to effectively reduce conditioned fear to the CS+ (ABA effect), or to another novel GS (ABC effect). The results consistently showed that presenting a GS in extinction was not effective in reducing conditioned fear to either the CS+ or a novel GS in test, especially on the first test trial. This finding was consistent with past studies that found heightened conditioned fear to the CS+ when a GS had been presented in extinction (Vervliet et al., 2004, 2005; Vervoort et al., 2014). However, the present work found that participants who received a GS in extinction showed a rapid decline in fear responding in test, especially after the first non-reinforced test trial. In terms of trait anxiety, trait anxious individuals showed slower extinction learning, but this was only observed in groups that received the
CS+ in extinction. High and low anxious individuals also showed no significant differences in the magnitude of generalisation of extinction learning.
A common theme revealed across all the present studies is that individuals high in trait anxiety show increased threat appraisal to stimuli that have a high level of threat ambiguity.
For instance, high anxious individuals showed increased conditioned fear to novel generalisation stimuli that had ambiguous threat value (Experiments 1, 2 & 3). The resistance in fear extinction to CS+ observed among high anxious individuals is also thought to be 208 attributed to the ambiguous threat value possessed by the CS+ (Experiments 4 & 5), since the
CS+ arguably signals both threat and safety at the same time (Bouton, 1993, 2007). Another common theme of the present work is that all findings can be readily explained by the cognitive account. These findings will be discussed in detailed below.
What do the findings mean?
Trait anxiety
One of the primary aims of the current thesis was to examine the effect of trait anxiety on fear learning, especially on fear generalisation. The effect of trait anxiety on different aspects of fear learning will be discussed below.
Physiological responses: The neurobiological models proposed by Eysenck (1957,
1967) and Gray (1981, 1982) suggested that anxiety-prone individuals have an oversensitive neurobiological system. This is thought to lead to an excessive level of physiological responses when innocuous cues are presented, resulting in a heightened level of anxiety. The current studies did not aim to specifically target the underlying pathways of hyper- physiological activity, such as a failure of habituation in physiological responses (e.g., Hart,
1974), or a higher baseline of physiological responses among trait anxious individuals
(Hodges & Speielberger, 1966; Kelley et al., 1970). However, no symtematic differences in skin conductance responses to the CSs and novel GSs were found between high and low anxious individuals in the current studies (except responding to the ambiguous cross- classified exemplars across Experiments 2 & 3). Although this may suggest that trait anxious individuals do not have an oversensitive physiological system that leads to hyperactivity in physiological responses, consistent with the majority of findings in the literature (e.g.,
Fahrenberg et al., 1983; McReynolds et al., 1966; Neary & Zuckerman, 1976; O'Gorman,
1977), this could be alternatively attributed to the high variability of skin conductance responses. Nonetheless, most of the significant findings in trait anxiety were found in the 209 expectancy measure, suggesting trait anxious individuals showed higher threat appraisal
(Lazarus, 1966, 1981, 1991) rather than heightened physiological fear responses to the same degree of threat appraisal. Therefore, the following will focus on the biases in threat expectancy found among trait anxious individuals in the present work.
Conditionability: A theory that has long been suggested is that trait anxious individuals show stronger acquisition to a fear cue, hence potentially explaining why they are prone to develop an anxiety disorder after a traumatic experience (Eysenck, 1967; Levey &
Martin, 1991). In a conditioning paradigm, this can be examined by testing whether trait anxious individuals show stronger and more rapid acquisition of responding to the CS+. In the current thesis, trait anxious individuals did not show a more rapid conditioning to the CS+, nor did they show stronger fear responses to the CS+ than those low in trait anxiety. This is especially evident across Experiments 1, 4 and 5, where the CS+ was partially reinforced.
This indicates that the failure to observe any enhanced conditionability to a CS+ among trait anxious individuals was not due to the presence of a ceiling effect in responding. The current findings are consistent with the literature, where majority of the studies have not observed enhanced conditionability to the CS+ in trait anxious individuals (e.g., Chan & Lovibond,
1996; Haddad et al., 2012; Otto et al., 2007; Sehlmeyer et al., 2011). Although some neuroimaging studies have found a positive correlation between differential responding to the
CSs and trait anxiety scores (e.g., Indovina et al., 2011), which suggest enhanced conditionability to CS+ among trait anxious individuals, there are alternative interpretations.
Since the activity in the amygdala circuit was found to be highly correlated with fear expression (Knight, Smith, Cheng, Stein & Helmstetter, 2004; Phelps et al., 2001), it is straightforward to interpret the observation of heightened amygdala activity in the presence of the CS+ as increased conditionability to the CS+. However, empirical studies have also found that an increased activation in the amygdala circuit mediates the detection of threat-related stimuli (LeDoux, 2000; Öhman, Flykt & Esteves, 2001), especially if the stimuli are phobic- 210 related (Dilger et al., 2003; Larson et al., 2006). That is, the heightened amygdala activity to the CS+ observed in trait anxious individuals may be associated with rapid detection of threat- related stimuli, rather than an increase in the magnitude of conditioned fear. This interpretation is more in line with the broader cognitive literature where consistent findings that trait anxious individuals showing attentional bias to threat-related stimuli have been shown (e.g., Broadbent & Broadbent, 1988; MacLeod et al., 1986; MacLeod & Mathews,
1988).
Impaired safety learning: The inability to inhibit fear responses to safe objects has also been proposed to be a pathogenic marker for anxiety disorders (Davis et al., 2000). A considerable body of evidence has shown that clinically anxious patients display elevated conditioned fear to a safety cue or context (e.g., Grillon & Morgan III, 1999; Hermann et al.,
2002; Lissek et al., 2009, 2010). However, there has been no concrete evidence that trait anxious individuals show a specific deficit in safety learning. While some studies found an increase in conditioned fear to safety cues or contexts among trait anxious individuals
(Gazendam et al., 2013; Grillon & Ameli, 2001; Haaker et al., 2015), other studies suggested that this observation is modulated by other factors (Baas et al., 2008; Boddez et al., 2012;
Chan & Lovibond, 1996; Chen & Lovibond, 2016; Haddad et al., 2012). One factor that is thought to affect fear responses among trait anxious individuals is the perceived ambiguity of the situation. Chan and Lovibond (1996) found that trait anxious individuals failed to inhibit their fear responses in the presence of a safe conditioned inhibitor, but this was only observed in trait anxious participants who were not aware of the CS-US contingency. In other words, being unaware of the CS-US contingency rendered all CSs equally unpredictable of the US, making the experimental configuration ambiguous. Similarly, Baas et al. (2008) found that only trait anxious individuals who were unaware that the shock contingency of the CS+ was context dependent showed an increase in fear responding to CS- when presented in a shock context. Furthermore, Chen and Lovibond (2016) found no differences in fear responding to 211 the CS- between individuals high and low in IU, when they were explicitly instructed that the
CS- would never be followed by an aversive outcome. This null difference in responding to the safety cue was thought to be attributed to the unambiguous threat value of the cue.
Another factor that may cause an apparent deficit in safety learning is over- generalisation. Haddad et al. (2012) presented two safety cues in a conditioning task, one that was perceptually similar to the CS+ (i.e., similar CS-) and one that was not (i.e., dissimilar
CS-). Trait anxious individuals showed more EMG eyeblink startle to the similar CS-, but not to the dissimilar CS-. This suggests that the apparent impaired safety learning was due to stronger perceptual fear generalisation among trait anxious individuals. Interestingly, the aforementioned positive evidence of deficient safety learning in trait anxious individuals can be potentially modulated or caused by one of these two factors. Some of these studies did not assess participants’ awareness of the CS-US or Context-US contingency (Grillon & Ameli,
2001; Haaker et al., 2015); therefore it is possible that the increase in conditioned fear to safety cues observed in these studies was largely driven by trait anxious participants who were not aware of the relevant contingencies. Other studies used perceptually similar CSs; for instance, Gazendam et al. (2013) used similar facial stimuli. In these cases, the apparent elevation in fear responding to safety can be alternatively interpreted as stronger perceptual generalisation of fear in trait anxious individuals.
Across Experiments 2 and 3, trait anxious individuals did not show any significant differences in safety learning when compared to their low anxious counterparts. If anything, trait anxious individuals showed a significantly larger differential skin conductance responding to the CSs in Experiment 3, and this effect was due to high anxious individuals having lower responding to the CS- exemplars. Additionally, high anxious participants showed seemingly poorer discrimination between threat cues (CS+ & GEN+) and safety cues 212
(CS- & GEN-) than the low anxious control group. However, this effect was driven by trait anxious individuals somewhat showing lower shock expectancies to the threat cues.
The current results also support the view that ambiguity is a modulating factor of the apparent impaired safety learning in trait anxious individuals. A post-experimental questionnaire was included in both Experiments 2 and 3, to ensure that all participants included in the data analysis were aware of the correct CS category-US association. Since all participants had learnt about the shock predictiveness of the cues, the threat value of the safety cues (CS-) became unambiguous and hence no impaired safety learning was found in trait anxious individuals. Interestingly, trait anxious individuals did not show stronger fear generalisation to the novel GEN- exemplars. Although one may argue their novelty may render their threat value relatively ambiguous, the clear categorical membership of the GEN- exemplars may have attenuated the threat ambiguity. Therefore, the current findings emphasize the role of ambiguity in modulating safety learning, rather than supporting the notion that the apparent impaired safety learning among trait anxious individuals is due to over-generalisation of fear.
Fear extinction: Across Experiments 4 and 5, trait anxious individuals showed significantly slower reduction in conditioned fear during extinction, consistent with past findings (e.g., Dibbets et al., 2014; Gazendam et al., 2013; Morriss et al., 2016). However, this effect seemed to be confined to standard fear extinction with the CS+, but not to a GS in extinction. This finding again fits in with the account that ambiguity modulates inhibitory learning in trait anxious individuals. According to Bouton (1993, 2007), an extinguished CS+ becomes ambiguous, since it can activate either the excitatory CS-US association or the inhibitory CS-no US association. In contrast, although the GS was also ambiguous in terms of its novelty and perceptual similarity to the CS+, it had never been associated with an aversive
US. The subsequent non-reinforced trials of GS in extinction further established its safety 213 properties, attenuating its threat ambiguity. Therefore, one possible reason why trait anxious individuals showed impaired inhibitory learning to the CS+ but not the GS, was that the threat value of the GS was less ambiguous.
Generalisation of fear and extinction: One of the primary aims of this thesis was to examine if trait anxiety has any effect on fear generalisation. The current results strongly suggest that trait anxious individuals show over-generalisation of fear, to novel stimuli on the same perceptual stimulus dimension with the CS+ (Experiment 1), or to novel stimuli that were categorically related to the CSs (Experiments 2 & 3). Importantly, this trait anxiety effect was only observed when participants perceived the situation as ambiguous or when the cues had ambiguous threat value. The collective evidence in the current thesis shed light on past studies that did not find any trait anxiety effect on fear generalisation (Arnaudova et al.,
2017; Torrents-Rodas et al., 2013). As previously discussed, one possible reason why these studies found null effects of trait anxiety on fear generalisation was because of the usage of a differential training procedure and an intensity stimulus dimension with the CSs located at the extreme ends. The combination of these experimental parameters may have rendered the generalisation task unambiguous - it was rather straightforward for participants to infer the threat value of each test stimulus based on their respective similarity to the CS+ or CS-. In
Experiment 1, no trait anxiety effect was found in fear generalisation for those who successfully identified a relational rule, aligning with the hypothesis that a trait anxiety effect would not manifest in unambiguous situations. However, participants who failed to identify a relational rule would have no basis to infer the threat value of the novel stimuli, rendering the shock predictiveness of the novel stimuli ambiguous, allowing the expression of trait anxiety effects.
In contrast, the generalisation of extinction studies (Experiments 4 & 5) did not find enough evidence to suggest stronger fear generalisation from the CS+ (A) to a GS (B) among 214 trait anxious individuals. One possible reason for this was that some trait anxious individuals may have already come up with a relational rule (e.g., the less green the circle is, the less likely I will be shocked) that attenuated the ambiguity of the conditioning task. However, since whether participants successfully identified a rule or not was not assessed, it was not possible to separate them into rule subgroups like Experiment 1.
Another aim of the current thesis was to examine the effect of trait anxiety on the generalisation of extinction, which was specifically investigated in Experiments 4 and 5. In both experiments, both high and low anxious individuals generalised extinction learning of
GS to the CS+ or to a novel GS in a similar way. That is, both anxiety groups showed little generalisation of extinction learning. As previously discussed, a deficit in safety learning in trait anxious individuals is most likely modulated by the ambiguity of the experimental configuration. It is possible that after experiencing a different stimulus in extinction, participants may have identified a relational rule, rendering the experimental configuration less ambiguous. In this case, trait anxiety effect may not impact the transfer of inhibitory learning.
Collectively, these findings suggest that fear generalisation is largely an adaptive process for both high and low anxious individuals. When the stimuli presented were arbitrary, some of the participants were able to identify and extrapolate a relation rule that guided generalisation, attenuating the threat ambiguity of the novel stimuli. When the novel exemplars were objects commonly seen in real life, participants were able to selectively generalise their fear according to their categorical membership. However, fear generalisation became maladaptive in trait anxious individuals only when they perceived the situation as ambiguous, that is, when they had no clear basis to judge the threat value of the novel stimuli.
The finding that trait anxious individuals showed over-generalisation of fear for ambiguous threat cues provides some preliminary evidence that excessive fear generalisation 215 is a predispositional factor for the development of anxiety disorders, rather than a consequence of anxiety disorders. This interpretation is supported by a study by Lenaert and colleagues (2014). In this laboratory study, participants who showed more fear generalisation to novel stimuli that resembled a safe cue subsequently showed a higher level of anxiety in a six-month follow-up. In addition, some preliminary clinical evidence showed that individuals who displayed impaired inhibition of fear responses to an extinguished CS+ showed severe
PTSD symptoms after trauma exposure (Guthrie & Bryant, 2006; Pole et al., 2009). These studies suggested that the inability to suppress fear responses can be used as a behavioural marker for predicting clinical anxiety symptoms. Similarly, the present work provides preliminary evidence that over-generalisation of fear under conditions of threat ambiguity can be used as a behavioural marker for individuals at risk of developing anxiety disorders.
Ambiguity: The current findings emphasize the role of ambiguity in modulating fear generalisation, safety learning and fear extinction in trait anxious individuals. This aligns with the general principle that trait anxiety is associated with excessive threat appraisal under conditions of ambiguity (e.g., Baas et al., 2008; Boddez et al., 2012; Chan & Lovibond; Chen
& Lovibond, 2016). Furthermore, the current results complement findings in the broader cognitive literature that trait anxiety is associated with vulnerability to ambiguity. For instance, trait anxious individuals in the No rule subgroup showed expectancy biases to all the novel stimuli in Experiment 1. This aligned with the findings that trait anxious individuals showed interpretation biases (e.g., Eysenck et al., 1987; MacLeod & Cohen, 1993), which anxious individuals interpret ambiguity in a threatening way. The current findings also aligned with the findings that anxious individuals overestimate the probability of a negative event when relevant contingencies are ambiguous (i.e., interpretation bias; Butler & Mathews,
1987; Mitte, 2007; Stöber, 1997). Across Experiments 2 and 3, trait anxious individuals showed increased threat appraisal to the ambiguous cross-classified exemplars than their low anxious counterparts. Although it is intuitive to interpret this finding as a result of negative 216 interpretation to ambiguous cues (e.g., Eysenck et al., 1987; MacLeod, 1990; MacLeod &
Cohen, 1993), the results in Experiment 3 suggest otherwise. In fact, high and low anxious individuals showed no significant differences in how they interpreted the categorical membership of ambiguous exemplars in a forced-choice categorization task. The increase in threat appraisal to the ambiguous cross-classified exemplars observed in high anxious individuals is thought to be due to trait anxious participants utilizing a ‘better safe than sorry’ strategy when faced with ambiguous stimuli (Eysenck et al., 1987; Lommen et al., 2010). In other words, even if high anxious individuals classified the ambiguous exemplars as safe items, they made precautionary higher shock expectancies to mentally prepared themselves in case the exemplars would be followed by a shock.
In Experiment 1, some participants were able to identify a relational rule to guide fear generalisation. Several factors may contribute to how likely participants are to engage in rule formation in a conditioning task. First, the information available for rule extrapolation. The more information available, the more likely individuals are able to infer a relational rule. This is evident in previous conditioning studies (e.g., Lee et al., 2018; Wong & Lovibond, 2017), where it was found that more participants who received a differential conditioning procedure successfully identified a rule than those who received a single-cue conditioning procedure, since the former gives an extra piece of information (i.e., CS-). Therefore, the single-cue conditioning procedure may have limited the number of individuals who could successfully identify a rule (see Premack, 1995). This finding also suggests that the use of a single-cue conditioning procedure did successfully induce ambiguity to the experimental configuration.
Secondly, individual differences other than trait anxiety may affect the likelihood that participants will identify a rule. For example, participants more open to experience may be more likely to entertain a rule during test (see Kaufman, 2013), while participants more motivated to learn could be more engaged in goal-oriented learning, and hence be more likely to infer a relational rule (see De Houwer, 2009). 217
Strong and weak situations
The current thesis relied heavily on the usage of a ‘weak’ situation to reveal any trait anxiety differences in the generalisation of fear acquisition and extinction. It also provided some new interpretations for the induction of a ‘weak’ situation. Lissek et al. (2006) argued that partial reinforcement of a CS+ may induce a ‘weak’ situation since its shock predictiveness is uncertain. However, the current results suggest otherwise. Across
Experiments 1, 4 and 5, the CS+ was partially reinforced at a 75% rate. Both high and low anxious individuals showed similar fear responding to the CS+, that is, trait anxious individuals did not show more fear responding to the threat cue (i.e., enhanced conditionability to CS+). In fact, all individuals seemed to have learnt the reinforcement rate of the CS+, since their averaged shock expectancies were approximately at 75%. This suggests that partial reinforcement per se may not be strong enough to reveal any trait anxiety differences on fear learning.
Conversely, although the argument that a differential conditioning procedure induces a
‘strong’ situation (Beckers et al., 2013; Lissek et al., 2006), the findings in Experiment 2 and
3 suggest that the level of ambiguity for different stimuli could be individually manipulated in a differential conditioning procedure. For instance, the GEN exemplars were used as ‘strong’ cues where they had clear threat value due to their clear categorical membership despite of their novelty. Conversely, the CC exemplars were used as ‘weak’ cues because of their ambiguous categorical membership and therefore ambiguous threat value. Importantly, these
‘strong’ and ‘weak’ cues could not be established without a differential conditioning procedure.
In conclusion, the single-cue conditioning procedure successfully induced a ‘weak’ situation, where over half of the participants were not able to identify any clear rules in
Experiment 1. Although partial reinforcement of CS+ was also suggested to be able to induce 218 a ‘weak’ situation (Lissek et al., 2006), no evidence was found to support this notion since trait anxious individuals did not show any biases in responding to the CS+. The current work also suggests that a differential conditioning procedure can be tailored to induce ambiguous
‘weak’ stimuli.
Cognitive and associative accounts of human generalisation
Collectively, the findings from Experiment 1, 2 and 3 were more consistent with the cognitive (e.g., Mitchell et al., 2009) than the associative accounts (e.g., McLaren &
Mackintosh, 2002; Rescorla & Wagner, 1972). Using a continuous perceptual dimension, a peaked gradient with the peak responding at CS+ was observed in the overall data in
Experiment 1. This pattern aligned with the predictions of associative accounts: responding peaked at CS+ was consistent with the idea that CS+ had the highest level of associative strength (Rescorla & Wagner, 1972). The gradual decline in responding towards both ends of the dimension indicated progressively lower associative strength as stimuli share fewer common elements with CS+ (Blough 1975; McLaren & Mackintosh, 2002). However, participants reported using a variety of rules during the generalisation task. After classifying participants into different subgroups according to their reported rules, distinctive non- similarity-based gradients (e.g., linear) in fear generalisation were observed, which cannot be readily explained by the associative accounts.
The operation of rules in human generalisation is consistent with past findings (e.g.,
Livesey & McLaren, 2009; Shanks & Darby, 1998). Specifically, the reported rules by participants in Experiment 1 closely aligned with their corresponding generalisation patterns, consistent with previous findings in our lab that examined rule-based generalisation in humans (Ahmed, 2014; Ahmed & Lovibond, 2018; Lee et al., 2018; Wong & Lovibond,
2017). According to the cognitive account, individual participants’ level of threat beliefs for each novel GS were guided by their relational rules. That is, participants determined the threat 219 value of each novel GS based on the identified rules, and responded accordingly. In fact, rules were formed primarily because participants were extrapolating the shock predictiveness of the novel GSs based on what they had learnt from the CS+. The inferred rules subsequently formed the basis of threat belief to every GS, and the threat belief to each GS directly affected how participants responded. This is evident in both expectancy and skin conductance measures since responding to the same GS was directly determined by different threat beliefs.
For instance, participants in the Linear subgroup inferred that the stimulus at the right end of the dimension (i.e., stimulus I) had the highest threat value and therefore showed the highest level of fear responding to it. In contrast, participants in the Similarity subgroup showed a low level of conditioned fear to the same stimulus since they inferred that stimuli at both ends of the dimension (i.e., stimuli A & I) had the lowest threat value.
Similarly, the finding that fear generalised to novel cues beyond perceptual features
(Experiments 2 & 3) cannot be readily explained by the associative accounts. If fear generalisation was guided only by the perceptual features of the exemplars, the generalisation pattern would be highly irregular since the training and testing exemplars were perceptually connected with each other in a range of ways (e.g., colour, shape). Instead, participants showed clear differential responding to the novel exemplars based on their categorical membership. That is, participants showed heightened fear responses to novel exemplars that belonged to a category associated with shock while showing a low level of fear to exemplars that belonged to a category that signaled safety. Threat belief played an important role in determining the fear responses to the novel exemplars; it was formed based on exemplars’ categorical membership once the categorical membership-shock association was learnt.
Conversely, most of the findings in the generalisation of extinction studies
(Experiments 4 & 5) were equally amenable to both associative and cognitive accounts. All participants showed a gradual decrease in responding to either the CS+ or GS in extinction. 220
The associative accounts explain the declining responses because of the accumulating inhibitory strength across the non-reinforced trials. The cognitive account explains this pattern in terms of the increasing strength of the belief that the stimulus would no longer be followed by a shock. According to the associative accounts, inhibitory learning to a GS cannot be fully generalised to the CS+ because only the elements shared between the CS+ and the GS gained inhibitory strength (Blough, 1975; McLaren & Mackintosh, 2002). Therefore, the excitatory elements that were uniquely possessed by the CS+ were still able to trigger conditioned fear to a certain extent.
Despite the effort to increase the perceptual similarity between stimuli, therefore theoretically increasing the number of shared elements that could be extinguished via extinction, the current studies only found a weak effect of generalisation of extinction. An alternative explanation is the protection of extinction effect exerted by the novel elements uniquely possessed by the GS. These novel elements had no associative strength initially, but subsequently gained negative associative strength across extinction trials. This process would transform these novel elements into conditioned inhibitors. Inhibitory learning to the GS would be attributed to the inhibitory associative strength of these conditioned inhibitors, but not to the elements shared between the GS and the CS+, nor to the elements shared between the CS+, the extinguished GS and another novel GS. Therefore, this protected the positive associative strength possessed by these shared elements and hence they were able to trigger conditioned fear when the CS+ or the novel GS was presented. On the other hand, the cognitive account can also explain this pattern under the notion of protection of extinction.
Participants may have attributed the absence of an aversive outcome to the novel features of the GS. For example, when a participant received an aqua-coloured circle as CS+ and then received non-reinforced blue circles (GS) in extinction, he/she may attributed the absence of shock to the ‘blueness’ of the circle. Therefore, when this feature was absent (or attenuated) in the stimulus (either the original CS+ or another novel GS) presented in test, participants may 221 expect an imminent shock. An alternative explanation by the cognitive account is the unwillingness of extrapolation of safety learning. After learning that a stimulus similar to the threat cue was not followed by a shock, participants may not be confident to generalise the safe properties from a GS to the CS+ or to another novel GS just because of their mere perceptual similarity.
However, the associative accounts cannot accommodate the rapid decline in responding to the test stimulus in participants who received a GS in extinction. The cognitive account, in contrast, suggests that participants who received a different exemplar in extinction, after experiencing the first non-reinforced test trial, may be more confident to connect the extinction and test stimuli together (in terms of their expected threat value), hence leading to a rapid decline in fear responding.
Dual-system or single-system learning models?
Overall, most findings in the current thesis showed the involvement of cognitive processes in human fear generalisation, while some findings were equally amenable with the associative accounts. As discussed in Chapter 1, a conventional approach to reconcile both cognitive and associative accounts in human associative learning is the proposal of a dual- system model (e.g., Clark & Squire, 1998; McLaren et al., 2014). The dual-system model suggests that humans possess an associative system and a cognitive system that independently contribute to learning. The reflexive associative system involves learning in an automatic and unconscious manner, while the higher-order cognitive system is responsible for the development of explicit propositional beliefs about the relationship between events. A couple of assumptions have been typically adopted in many dual-system models. First, it is often assumed that these systems work independently from each other (Clark & Squire, 1998; Wills
& Mackintosh, 1998). Secondly, some dual-system models assume that when cognitive resources are sufficient, the cognitive system will override the associative system and guide 222 learning (Livesey & McLaren, 2009; McLaren et al., 2014). Thirdly, associative system will guide learning when cognitive resources are not available (McLaren, Green & Mackintosh,
1994; McLaren et al., 2014).
Alternatively, the propositional model denies the existence of the associative system in humans, and proposes that associative learning in humans is only accounted for by the cognitive system (e.g., De Houwer, 2009; Mitchell et al., 2009). The propositional model suggests that the formation of a CR is derived from the propositional belief of the CS-US contingency, that is, an individual will show CRs to the CS only if he/she is aware that the CS predicts the US. Indeed, as previously discussed in Chapter 1, empirical studies have challenged the existence of an automatic learning system in humans (e.g., Dawson & Biferno,
1973; Ross & Nelson, 1973; Weidemann & Antees, 2012; Weidemann, Satkunarajah &
Lovibond, 2016).
Although reconciling fear generalisation in humans within either of the learning models was not the primary aim of the current thesis, the present findings highlighted the importance of cognitive processes in human fear generalisation. More importantly, all of the current results can be accommodated by the cognitive account. For instance, across
Experiments 1, 2 and 3, the propositional model can readily explain that participants’ threat beliefs to the novel GSs determined the diverse generalisation patterns observed, and how threat beliefs determined how fear generalised beyond perceptual features of stimuli. In contrast, the dual-system models provide a less clear explanation for the current results. First, the current findings did not fully support the notion that there were two learning systems operating independently from each other. While dual-system learning models (e.g., Clark &
Squire, 1998; Razran, 1955) typically postulate a dissociation between cognitively-controlled expectancy ratings and associatively-controlled skin conductance in non-similarity-based gradients (Linear and No rule), the gradients in both measures were aligned with participants’ 223 reported rules. Furthermore, these strict learning models view skin conductance responses as
CRs generated by a low-level, automatic system, and suggest that the generalisation of skin conductance responses depends on perceptual similarity of the stimuli. However, the finding that skin conductance responses generalised beyond perceptual features in Experiments 2 and
3 violated this postulation.
Secondly, some dual-system accounts assume that the cognitive system overrides the associative system when cognitive resources are available (e.g., Livesey & McLaren, 2009).
Although this assumption resolves the conflict mentioned above, it does pose some questions.
For example, how can dual-system accounts predict when the cognitive system will override the associative system? Some studies suggested that when experimental configuration limits rule learning, the cognitive system will remain dormant and the associative system will guide learning. For instance, studies that used a stimulus dimension that was difficult to verbalize as a rule (Wills & Mackintosh, 1998) or that used a reduction in training trials (Jones &
McLaren, 1999) found generalisation patterns that were more consistent with a peak shift effect, a learning phenomenon predicted by the associative accounts as discussed in Chapter
2. Similarly, Livesey and McLaren (2009) used a blue-green stimulus dimension that was difficult to discriminate, and found that the generalisation pattern in early test trials tended to be consistent with a peak shift effect, but the pattern became more linear in late test trials. The authors suggested that because the stimuli were difficult to discriminate, participants were unable to verbalize any relational rules in early test trials. At this stage, generalisation was thought to be guided by the associative system, hence the generalisation pattern aligned with an associatively-driven peak shift effect. In later trials, participants had more experience with the stimulus dimension and were able to identify relational rules such as a linear rule, hence the cognitive system took over and participants displayed rule-based generalisation. However, the findings in Experiment 1 did not fully align with this notion. Participants in the No rule subgroup failed to identify any clear rules; according to the dual-system accounts, 224 generalisation should nonetheless have been guided by the associative system which would produce a sharp, peaked gradient. However, the current findings showed a peaked gradient in the No rule subgroup that was significantly flatter than the similarity-based gradient, suggesting that generalisation was not being driven by an associative system. This finding aligned with some evidence in the literature that opposes the idea that cognitive load reveals the operation of the associative system (Dawson, 1970; Dawson & Biferno, 1973). Instead, these studies found that the addition of cognitive load during conditioning decreased both electrodermal CRs and explicit knowledge of CS-US contingency (see also Mitchell et al.,
2009).
Although peaked gradients have been seen as a hallmark of associatively-guided generalisation (Blough, 1975; McLaren & Mackintosh, 2002), recent evidence has shown that the same gradient can also be observed in participants who infer a similarity-based rule
(Ahmed & Lovibond, 2018; Lee et al., 2018; Wong & Lovibond, 2017). In Experiment 1, the generalisation pattern in the Similarity subgroup closely resembled an associatively-driven peaked gradient, while other rule subgroups showed generalisation patterns that could not be accounted for by the associative accounts. In fact, the peaked gradient observed in Experiment
1 arised from an explicit similarity rule.
The above argument suggests that a peaked gradient is not an exclusive support for associatively-driven generalisation. This phenomenon is equally amenable to explanation by the cognitive account. While some of the findings across Experiments 4 and 5 are equally amenable by both associative and cognitive accounts, the majority of the current findings are more readily accommodated by the cognitive account. The present work emphasizes the importance of cognitive processes in human learning and generalisation. In fact, it suggests that generalisation in humans is a form of inductive reasoning, a conscious cognitive process that involves the extrapolation of known evidence to novel situations. In Experiment 1, some 225 participants were able to identify a rule and generalised their fear accordingly. However, similar to inductive reasoning, the validity of the inferred rules is not guaranteed. Similarly, participants in Experiments 2 and 3 inferred that the categorical membership of the exemplars was the predictor of an aversive outcome or not, but, the validity of this logical induction is also not guaranteed. Given that no evidence has been found for a separate associative system in the current work, it highlights the parsimonious explanation provided by the propositional model.
Clinical implications
Associative accounts propose that the magnitude of conditioned fear is directly determined by the strength of the CS-US link (e.g., Rescorla-Wagner model, 1972).
Conditioned fear is generalised to other stimuli that are perceptually similar to the CS
(Blough, 1975; McLaren & Mackintosh, 2002), and when the CS is no longer reinforced (i.e., fear extinction), an inhibitory CS-no US association will be produced that competes with the original CS-US link (Bouton, 1993; Miller & Matzel, 1988). According to the inhibitory model, the inhibitory strength of the CS-no US association would also spread to stimuli of perceptual similarity, reducing the degree of fear generalisation (e.g., Hull, 1943). This theoretical concept forms the basis of exposure therapy, aimed at extinguishing maladaptive clinical fear.
Some of the current findings are compatible with some aspects of the above framework, such as the decrease in conditioned responses in extinction (e.g., Bouton, 1993;
Rescorla & Wagner, 1972); however, the present work is more readily explained by the cognitive account and highlights the importance of the role of cognition in fear generalisation in humans. Specifically, the findings emphasized how relational rules and threat beliefs determine how fear is generalised to novel objects that have unknown threat value. This finding aligns with the core component of cognitive-behavioural therapies (CBT), which 226 target the correction of maladaptive threat beliefs to innocuous objects in order to reduce emotional distress and maladaptive behaviours among clinically anxious patients (Beck, 1970,
1979). Given that the current findings can all be accounted for in cognitive terms, and a separate associative system is not required to account for the findings, the present work suggests that threat belief and conditioned fear may arise from a common system (Lovibond,
2011; Mitchell et al., 2009). Thus, the current research calls for an increased emphasis on the
‘cognitive component’ in exposure-based therapies, and a focus on the deconstruction of patients’ maladaptive beliefs. In fact, CBT has been found to be highly efficient in reducing anxiety symptoms in anxiety disorders (see Hofmann, Asnaani, Vonk, Sawyer & Fang, 2012).
The finding that trait anxious individuals showed adaptive generalisation of fear when clear rules were available or when the threat value of cues were clear suggest that anxious patients will benefit from therapeutic strategies that help them to disambiguate a potentially threatening situation, or to generate rules to assess the situation more adaptively. Chen and
Lovibond (2016) suggested to explicitly train anxious patients to quantify threat probability in ambiguous situations, to help patients to more adaptively evaluate the probability of novel threatening events in the future. Similarly, other researchers have suggested that encouraging anxious patients to identify a rule may help guide their generalisation of extinction memories
(Treanor, Stapes-Bradley & Craske, 2015; Vervliet, Baker & Craske, 2012). It has also been suggested that the return of fear via contextual renewal is due to the ambiguous threat of the extinguished CS+ (Craske, Liao, Brown & Vervliet, 2012). The extinction context is thought to act as an occasion setter for the inhibitory CS-no US association, disambiguating the current meaning of the extinguished CS+. Therefore, when the extinguished CS+ is presented in a context different from the extinction context, the ambiguity of the CS+ returns, leading to a return of fear. This suggests that carrying out exposure-based therapies in different contexts
(e.g., inside or outside of clinical settings) may facilitate the generalisation of extinction contexts, therefore disambiguating the meaning of the extinguished threat cue. This notion is 227 supported by empirical evidence showing significantly less return of fear via contextual renewal after exposure to multiple contexts during extinction (Balooch, Neumann & Boschen,
2012; Bandarian-Balooch, Neumann & Boschen., 2015; Neumann, 2006; Vansteenwegen et al., 2007; but see Neumann, Lipp & Cory, 2007).
The findings in Experiments 4 and 5 also suggest that fear extinction with a GS is not as effective as extinction with the original CS+ to reduce fear to the CS+ itself or to a novel cue. This suggests that extinction with a GS is a potential new pathway for the ‘return of fear’ in real life, especially evident in the ABC paradigm. Since it is highly unlikely that a therapist will be able to expose the original threat cue in treatment (e.g., a PTSD patient who experienced a traumatic car accident), patients are presumably instead exposed to cues that are perceptually or conceptually similar to the original threat cue. The heightened fear response to a novel cue in test is conceptually equivalent to clinically anxious patients showing fear to novel objects that resemble the original threat cue after successful treatment.
However, despite this initial increase in fear responses, participants showed a rapid decline in conditioned fear after learning that the cue is non-threatening (i.e., after the first non- reinforced test trial). This suggests that when avoidance is restricted when patients encounter the novel cue (e.g., exposure response prevention; preventing patients from protection of extinction), they will have the opportunity to learn that the cue is non-threatening more effectively. This suggests that exposing multiple stimuli in extinction may facilitate safety learning to novel cues. In fact, empirical evidence has shown that arachnophobics showed stronger short- and long-term fear reduction after being exposed to different spider stimuli
(Shiban, Schelhorn, Pauli & Mühlberger, 2015). A more recent study also showed that participants who were exposed to multiple stimuli in fear extinction showed a significant reduction in conditioned skin conductance responses to the CS+ and other novel cues in test
(Waters, Kershaw & Lipp, 2018). 228
Limitations and future directions
One limitation of the current work was the lack of strong skin conductance data across
Experiments 1, 4 and 5. All trait anxiety differences in these studies were only evident in the expectancy measure, but not in the skin conductance measure. This suggests that skin conductance may not be a sensitive measure to pick up any trait anxiety differences, particularly when the stimuli in use were arbitrary. This is one potential reason why we did not find enhanced physiological responses in trait anxious individuals. Furthermore, the skin conductance data did not fully align with the expectancy measure in the acquisition and extinction phases in Experiment 4 and 5. This pattern may be attributable to the high variability of skin conductance responses. However, both measures aligned with each other in the test phases in Experiments 4 and 5, while all the generalisation gradients in skin conductance aligned closely with the expectancy measures and inferred rules in Experiment 1.
Therefore, even though the skin conductance was highly variable, there was no indication of a substantial dissociation between the two measures.
Alternatively, one may argue that the significant trait anxiety effect on the expectancy measure was simply an artefact, since the measure itself may merely reflect the anticipation of an outcome but not how fearful an individual is. In other words, trait anxious individuals may simply anticipate the outcome more, but not necessarily feel an increased level of fear, hence explaining the null trait anxiety effect in skin conductance. However, as discussed in Chapter
3, the expectancy measure is similar to threat appraisal in the sense that it not only reflects the perceived probability of a negative event to occur, but also the perceived cost of the event
(Paterson & Neufeld, 1987). Given the intensity of the shock US was individually tailored at a level perceived to be aversive, the expectancy measure would likely reflect both outcome anticipation and conditioned fear to the stimulus (see Boddez et al., 2013). Empirical studies have also shown a decrease in threat expectancy in patients with specific phobias after 229 successful treatment (e.g., Adler, Craske, Kirshenbaum & Barlow, 1989; Williams, Kinney &
Falbo, 1989), suggesting that outcome expectancy and fear are closely associated with each other.
The discrepancy in trait anxiety effect between the expectancy and skin conductance measures could also be potentially explained by the three-system model of anxiety (Frijda,
1986; Hugdahl, 1981; Lang, Bradley & Cuthbert, 1998). As discussed in Chapter 1, it has been suggested that fear can be expressed via three response systems, namely cognitive, physiological and behavioural. These three systems are thought to work independently to each other, therefore making it possible to observe different levels of responding across measures to the same stimulus or situation. Fear expressed via these three systems weakly correlate with each other (Mauss, Levenson, McCarter, Wilhelm & Gross, 2005), which has been seen as supportive evidence for the three-system model. Although the three-system model is able to account for by the discrepancy between different fear expressions, it also raises some questions. For example, it does not give a clear explanation why certain fear responses (e.g., physiological responses) diverge from others (e.g., behavioural responses). It also does not give a clear and testable prediction for when the fear responses will diverge from each other, and when they will express themselves in parallel with each other. For instance, the three- system model may not be able to predict nor give a clear explanation for why the effect of trait anxiety was found to be significant in both the cognitive and physiological data in
Experiments 2 and 3, but differed between the two measures in Experiment 1. In fact, the weak correlation across the three measures may merely reflect unique sources of measurement error in the three response systems (see Beckers et al., 2013). This idea also provides a viable explanation for the variation in the significance of the trait anxiety effect between measures found in the current work. 230
One may also question whether the lack of significant skin conductance results might be attributed to the way in which skin conductance was analysed. Specifically, we analysed the change in tonic skin conductance level (ΔSCL) rather than the conventional analysis of discrete skin conductance responses (SCR). However, these two measures are highly correlated, and it has been argued that ΔSCLs are a more face valid measure of sustained fear/anxiety responses during a period of anticipation of an aversive event (Lovibond, 1992).
Furthermore, a significant amount of research from our lab has confirmed that the ΔSCL is a sensitive measure of conditioned fear (e.g., Lovibond, Davis & O’Flaherty, 2000; Lovibond,
Mitchell, Minard, Brady & Menzies, 2009; Mitchell & Lovibond, 2002; Ng & Lovibond,
2017; Sokol & Lovibond, 2012; Wong & Lovibond, 2017). The findings from our lab are also highly comparable to other fear conditioning studies that have measured SCR (e.g., Dunsmoor et al., 2012, 2014; Vervliet et al., 2004, 2005, 2010; Zbozinek & Craske, 2018). Therefore, the lack of significance of a trait anxiety effect in skin conductance in Experiments 1 ,4 and 5 is unlikely to be due to the analysis of ΔSCLs.
One may argue that another limitation of the present work was that trait anxious participants’ diagnostic status was not formally assessed, hence it is difficult to come to a strong conclusion that over-generalisation of fear is a predispositional factor of anxiety disorder or a consequence of anxiety disorder. However, most anxious patients have high trait anxiety (Beck et al., 1988; Clark, Watson & Mineka, 1994); the current project arguably shows trait anxiety effects in participants before they develop an anxiety disorder. This suggests that over-generalisation of fear under conditions of ambiguity is a vulnerability factor of anxiety disorder.
The current thesis examined fear generalisation using a Pavlovian fear conditioning paradigm. However, within a Pavlovian framework, only passive fear learning can be measured. For instance, participants had to merely rate their shock expectancies while having 231 their skin conductance measured throughout the experiment. In contrast, an operant/instrumental conditioning procedure involves active learning, since one’s actions determine the occurrence of an outcome or not. For instance, after learning that a cue (e.g.,
A+) predicted a shock, and performing a particular behaviour (e.g., pressing a particular key) led to the omission of a shock, participants quickly learnt to press the key when A+ was presented in order to avoid the shock (e.g., Lovibond et al., 2009; Vervliet & Indekeu, 2015).
Excessive fear generalisation is thought to lead to maladaptive fear avoidance, since this results in the avoidance of a wide variety of innocuous objects or situations (Dymond et al.,
2015; Pittig et al., 2018). This excessive fear avoidance undoubtedly impairs daily function, and leads to ‘protection of extinction’ (Lovibond et al., 2009; see also Rescorla, 2003). That is, the constant avoidance of fear-related cues prevents the patients from learning that the cues pose no actual threat. Maladaptive fear avoidance is considered to be one of the pathological features of anxiety disorders (Craske, 1999; Dymond & Roche, 2009). Since the current work has provided preliminary evidence that trait anxious individuals show over-generalisation of fear in the presence of threat ambiguity within a Pavlovian framework, it will be beneficial for future studies to examine if trait anxious individuals generalise fear avoidance within an operant/instrumental paradigm in a similar way.
Recent fear conditioning studies have also utilized the Virtual Reality (VR) technique
(e.g., Baas, Nugent, Lissek, Pine & Grillon, 2004; Genheimer, Andreatta, Asan & Pauli, 2017;
Glotzbach-Schoon, Andreatta, Mühlberger & Pauli, 2013; Grillon, Baas, Cornwell &
Johnson, 2006; Kroes, Dunsmoor, Mackey, McClay & Phelps, 2017). The advantage of using
VR is the manipulation of different stimuli in a naturalistic environment. That is, the stimuli and contextual background can be manipulated to resemble how anxious patients acquire their pathological fear in a naturalistic environment. Furthermore, since the contextual background can be manipulated (e.g., driving in a car), multiple stimuli can then be presented in a meaningful way (e.g., traffic lights, traffic signs). This allows direct examination on what 232 determines fear learning and fear generalisation to any given stimulus. VR studies also provide the opportunity for examining how patients approach or avoid fear-related cues since they are immersed in a more interactive environment (e.g., Biedermann et al., 2017;
Glotzbach, Ewald, Andreatta, Pauli & Mühlberger, 2012). By the same token, preliminary evidence has suggested that virtual reality exposure-based therapies are as efficient as traditional exposure-based treatments, especially to patients who refuse to receive traditional exposure (see Gonçalves, Pedrozo, Coutinho, Figueira & Ventura, 2012; Morina, Ijntema,
Meyerbröker & Emmelkamp, 2015).
The current work examined how extinction learning generalises, but primarily on a perceptual dimension. It will be beneficial for future work to further examine the generalisation of extinction beyond perceptual features, for instance, to stimuli that are categorically related. For example, after acquiring fear to a picture of a dog, how effective will the extinction of a novel animal picture that belongs to the same category as a dog (e.g., mammal; a picture of a cat) be in terms of generalising back to the original dog picture?
Given that little to no generalisation decrement was observed to novel exemplars that were categorically related to the threat cues in Experiments 2 and 3, it will be interesting to examine if a similar pattern would be observed in generalisation of extinction learning (but see Vervoort et al., 2014).
Furthermore, the present work has suggested that fear generalisation in humans is a form of inductive reasoning; therefore it will be interesting to see if factors that affect inductive reasoning will also affect generalisation of extinction in a similar way. There is some evidence that stimulus typicality affects inductive reasoning and fear generalisation in the same way (Dunsmoor & Murphy, 2014; Osherson et al., 1990). That is, the more typical the CS+ is of its category, the stronger fear generalises to novel cues. It will be interesting to see if similar findings can be observed in the generalisation of extinction, that is, the more 233 typical the GS presented in extinction, the more fear reduction to CS+ or to other novel cues that belong to the same category of CS+.
Given that the present work provides preliminary evidence that trait anxious individuals display over-generalisation of fear in the presence of threat ambiguity, prospective studies can examine whether over-generalisation of fear is a predictive behavioural marker for the development of anxiety disorders or severity of anxiety symptoms after trauma exposure.
These studies can be done in populations at high risk for trauma exposure, such as soldiers, firefighters and paramedics.
Conclusions
Fear generalisation allows one to utilize previous experiences to respond adaptively to novel, potential threats. However, excessive fear generalisation is considered as maladaptive, since it creates false alarms of perceived threat to a variety of harmless objects or situations.
Preliminary evidence has suggested that over-generalisation of fear is a pathogenic marker of anxiety disorders (Kaczkurkin et al., 2017; Lissek et al., 2010, 2014), highlighting the clinical importance of understanding the process of fear generalisation.
The current thesis highlights the important role of threat ambiguity in modulating the effect of trait anxiety on fear generalisation and extinction. It also suggests that extinction learning to a novel cue that perceptually resembles the threat cue does not effectively generalises to the original threat cue or to another related cue. More generally, the present work highlights the important role of cognitive processes in human fear generalisation, specifically how relational rules and threat beliefs affect how fear generalises. In terms of clinical implications, this thesis suggests that the effectiveness of anxiety treatments may be enhanced by a greater emphasis on modifying patients’ maladaptive beliefs, and providing strategies that help disambiguate threat and safety cues. 234
References
Adler, C. M., Craske, M. G., Kirshenbaum, S., & Barlow, D. H. (1989). ‘Fear of panic’: An
investigation of its role in panic occurrence, phobic avoidance, and treatment
outcome. Behaviour Research and Therapy, 27(4), 391-396.
https://doi.org/10.1016/0005-7967(89)90009-0
Ahmed, O. (2014). Processes underlying generalisation of conditioning in humans. (Doctoral
dissertation, University of New South Wales. http://handle.unsw.edu.au/1959.4/54392
Ahmed, O. & Lovibond, P. F. (2015). The impact of instructions on generalization of
conditioned fear in humans. Behavior Therapy, 46(5), 597-603.
https://doi.org/10.1016/j.beth.2014.12.007
Ahmed, O., & Lovibond, P. F. (2018). Rule-based processes in generalisation and peak shift
in human fear conditioning. Quarterly Journal of experimental Psychology.
https://doi.org/10.1177/1747021818766461
Alloy, L. B., & Tabachnik, N. (1984). Assessment of covariation by humans and animals:
The joint influence of prior expectations and current situational
information. Psychological Review, 91(1), 112-149.
http://dx.doi.org/10.1037/0033-295X.91.1.112
Alvarez, R. P., Johnson, L., & Grillon, C. (2007). Contextual-specificity of short-delay
extinction in humans: Renewal of fear-potentiated startle in a virtual environment.
Learning and Memory, 14(4), 247-253. http://dx.doi.org/10.1101/lm.493707
American Psychiatric Association. (2013). Diagnostic and statistical manual of mental
disorders (DSM-5®). American Psychiatric Publishing, Arlington, VA 235
Amies, P. L., Gelder, M. G., & Shaw, P. M. (1983). Social phobia: A comparative clinical
study. British Journal of Psychiatry, 142, 174-179.
http://dx.doi.org/10.1192/bjp.142.2.174
Amir, N., Elias, J., Klumpp, H., & Przeworski, A. (2003). Attentional bias to threat in social
phobia: Facilitated processing of threat or difficult disengaging attention from threat?
Behaviour Research and Therapy, 41(11), 1325-1335.
https://doi.org/10.1016/S0005-7967(03)00039-1
Anderson, D. S., Cole, J., & McVaugh, W. (1968). Variations in unsingaled inescapable
preshock as determinants of responses to punishment. Journal of Comparative and
Physiological Psychology, 65(3), 1-17. https://dx.doi.org/ 10.1037/h0025838
Anderson, O. D., & Liddell, H. S. (1935). Observations on experimental neurosis in sheep.
Archives of Neurology & Psychiatry, 34(2), 330-354.
https://doi.org/10.1001/archneurpsyc.1935.02250200090008
Andrade, L., Gorenstein, C., Vieira Filho, A. H., Tung, T. C., & Artes, R. (2001).
Psychometric properties of the Portuguese version of the State-Trait Anxiety
Inventory applied to college students: Factor analysis and relation to the Beck
Depression Inventory. Brazilian Journal of medical and Biological Research, 34(3),
367-374. http://dx.doi.org/10.1590/S0100-879X2001000300011
Annau, Z., & Kamin, L. J. (1961). The conditioned emotional response as a function of
intensity of the US. Journal of Comparative and Physiological Psychology, 54, 428-
432. http://dx.doi.org/10.1037/h0042199
Andreatta, M., & Pauli, P. (2017). Learning mechanisms underlying threat absence and threat
relief: Influences of trait anxiety. Neurobiology of Learning and Memory, 145, 105-
113. https://doi.org/10.1016/j.nlm.2017.09.005 236
Antony, M. M., Bieling, P. J., Cox, B. J., Enns, M. W., & Swinson, R. P. (1998).
Psychometric properties of the 42-item and 21-item versions of the Depression
Anxiety Stress Scales in clinical groups and a community sample. Psychological
Assessment, 10, 176-181. http://dx.doi.org/10.1037/1040-3590.10.2.176
Arena, J.G., & Hobbs, S.H. (1995). Reliability of psychophysiological responding as a
function of trait anxiety. Biofeedback and Self-regulation, 20, 19-37.
http://dx.doi.org/10.1007/BF01712764
Arnaudova, I., Krypotos, A., Effting, M., Kindt, M., & Beckers, T. (2017). Fear shades of
grey: Individual differences in fear responding towards generalization stimuli.
Cognition and Emotion, 31(6), 1181-1196.
http://dx.doi.org/10.1080/02699931.2016.1204990
Asmundson, G. J. G., & Stein, M. B. (1994). Selective processing of social threat in patients
with generalized social phobia: Evaluation using a dot-probe paradigm. Journal of
Anxiety Disorders, 8(2), 107-117. http://dx.doi.org/10.1016/0887-6185(94)90009-4
Augustson, E. M., & Dougher, M. J. (1997). The transfer of avoidance evoking functions
through stimulus equivalence classes. Journal of Behavior Therapy and Experimental
Psychiatry, 28(3), 181-191. https://doi.org/10.1016/S0005-7916(97)00008-6
Baas, J. M., Nugent, M., Lissek, S., Pine, D. S., & Grillon, C. (2004). Fear conditioning in
virtual reality contexts: A new tool for the study of anxiety. Biological Psychiatry,
55(11), 1056-1060. https://doi.org/10.1016/j.biopsych.2004.02.024
Baas, J. M. P., van Ooijen, L., Goudriaan, A., & Kenemans, J. L. (2008). Failure to condition
to a cue is associated with sustained contextual fear. Acta Psychologica, 127(3), 581-
592. https://doi.org/10.1016/j.actpsy.2007.09.009 237
Balooch, S. B., Neumann, D. L., & Boschen, M. J. (2012). Extinction treatment in multiple
contexts attenuates ABC renewal in humans. Behaviour Research and Therapy, 10,
604-609. http://dx.doi.org/10.1016/j.brat.2012.06.003
Bandarian-Balooch, S., Neumann, D. L., & Boschen, M. J. (2015). Exposure treatment in
multiple contexts attenuates return of fear via renewal in high spider fearful
individuals. Journal of Behavior Therapy and Experimental Psychiatry, 47, 138-144.
http://dx.doi.org/10.1016/j.jbtep.2014.12.006
Bandelow, B., & Michaelis, S. (2015). Epidemiology of anxiety disorders in the 21st century.
Dialogues in Clinical Neuroscience, 17(3), 327-335.
Bar-Haim, Y., Lamy, D., Pergamin, L., Bakermans-Kranenburg, M. J., & van IJzendoorn, M.
H. (2007). Threat-related attentional bias in anxious and nonanxious individuals: A
meta-analytic study. Psychological Bulletin, 133(1), 1-24.
http://dx.doi.org/10.1037/0033-2909.133.1.1
Baron, A. (1973). Postdiscrimination gradients of human subjects on a tone continuum.
Journal of Experimental Psychology, 101(2), 337-342.
http://dx.doi.org/10.1037/h0035206
Barrett, J., & Armony, J. L. (2009). Influence of trait anxiety on brain activity during the
acquisition and extinction of aversive conditioning. Psychological Medicine, 39, 255-
265. http://dx.doi.org/10.1017/S0033291708003516
Baum, M. (1966). Rapid extinction of an avoidance response following a period of response
prevention in the avoidance apparatus. Psychological Reports, 18(1), 59-64.
https://doi.org/10.2466/pr0.1966.18.1.59 238
Baum, M. (1969). Extinction of an avoidance response motivated by intense fear: Social
facilitation of the action of response prevention (flooding) in rats. Behaviour Research
and Therapy, 7(1), 57-62. https://doi.org/10.1016/0005-7967(69)90049-7
Baum, M. (1970). Extinction of avoidance responding through response prevention
(flooding). Psychological Bulletin, 74(4), 276-284.
http://dx.doi.org/10.1037/h0029789
Beck, A. T. (1970). Cognitive therapy: Nature and relation to behavior therapy. Behavior
Therapy, 1(2), 184-200. http://dx.doi.org/10.1016/S0005-7894(70)80030-2
Beck, A. T. (1976). Cognitive theory and the emotional disorders. New York: International
Universities Press.
Beck, A. T. (1979). Cognitive therapy and the emotional disorders. Penguin.
Beck, A. T., & Clark, D. A. (1988). Anxiety and depression: An information processing
perspective. Anxiety Research, 1, 23-36. https://doi.org/10.1080/10615808808248218
Beck, A. T., & Emery, G. (1985). Anxiety Disorders and Phobias: A Cognitive Perspective.
New York: Basic Books.
Beck, A. T., Epstein, N., Brown, G., & Steer, R. A. (1988). An inventory for measuring
clinical anxiety: Psychometric properties. Journal of Consulting and Clinical
Psychology, 56(6), 893-897. http://dx.doi.org/10.1037/0022-006X.56.6.893
Beck, A. T., Steer, R. A., & Brown, G. K. (1996). Manual for the Beck Depression
Inventory-II. San Antonio, TX: Psychological Corporation.
Beck, A. T, Ward, C. H., Mendelson, M., Mock, J., & Erbaugh, J. (1961). An inventory for
measuring depression. Archives of General Psychiatry, 4(6), 561-571.
http://dx.doi.org/10.1001/archpsyc.1961.01710120031004 239
Beckers, T., Krypotos, A., Boddez, Y., Effting, M., & Kindt, M. (2013). What’s wrong with
fear conditioning? Biological Psychology, 92, 90-96.
https://doi.org/10.1016/j.biopsycho.2011.12.015
Bennett, C. H., Wills, S. J., Wells, J. O., & Mackintosh, N. J. (1994). Reduced generalization
following preexposure: Latent inhibition of common elements or a difference in
familiarity? Journal of Experimental Psychology: Animal Behavior Processes, 20,
232-239. http://dx.doi.org/10.1037/0097-7403.20.3.232
Bennett, M. P., Meulders, A., Baeyens, F., & Vlaeyen, J. W. (2015a). Words putting pain in
motion: The generalization of pain-related fear within an artificial stimulus category.
Frontiers in Psychology, 6:520. https://doi.org/10.3389/fpsyg.2015.00520
Bennett, M., Vervoort, E., Boddez, Y., Hermans, D., & Baeyens, F. (2015b). Perceptual and
conceptual similarities facilitate the generalization of instructed fear. Journal of
Behavior Therapy and Experimental Psychiatry, 48, 149-155.
https://doi.org/10.1016/j.jbtep.2015.03.011
Biedermann, S. V., Biedermann, D. G., Wenzlaff, F., Kurjak, T., Nouri, S., Auer, M. K., et al.
(2017). An elevated plus-maz in mixed reality for studying human anxiety-related
behavior. BMC Biology, 15, 125. http://dx.doi.org/10.1186/s12915-017-0463-6
Biferno, M. A., & Dawson, M. E. (1977). The onset of contingency awareness and
electrodermal classical conditioning: An analysis of temporal relationship during
acquisition and extinction. Psychophysiology, 14, 164-171.
http://dx.doi.org/10.1111/j.1469-8986.1977.tb03370.x
Birbaumer, N., Grodd, W., Diedrich, O., Klose, U., Erb, M., Lotze, M., Schneider, F., Weiss,
U., & Flor, H. (1998). fMRI reveals amygdala activation to human faces in social
phobics. NeuroReport: An International Journal for the Rapid Communication of 240
Research in Neuroscience, 9(6), 1223-1226.
http://dx.doi.org/10.1097/00001756-199804200-00048
Bishop, S., & Forster, S. (2013). Trait anxiety, neuroticism, and the brain basis of
vulnerability to affective disorder. In J. Armony & P. Vuilleumier (Eds.), The
Cambridge handbook of human affective neuroscience (pp. 553-574). New York, NY,
US: Cambridge University Press. http://dx.doi.org/10.1017/CBO9780511843716.031
Blackman, D. (1966). Response rate and conditioned suppression. Psychological Reports,
19(3), 170-174. https://doi.org/10.2466/pr0.1966.19.3.687
Blackwell, H. R., & Schlosberg, H. (1943). Octave generalization, pitch discrimination, and
loudness thresholds in the white rat. Journal of Experimental Psychology, 33(5), 407-
419. http://dx.doi.org/10.1037/h0057863
Blechert, J., Michael, T., Vriends, N., Margraf, J., & Wilhelm, F. H. (2007). Fear
conditioning in posttraumatic stress disorder: Evidence for delayed extinction of
autonomic, experiential, and behavioural responses. Behaviour Research and
Therapy, 45(9), 2019–2033. https://doi.org/10.1016/j.brat.2007.02.012
Blough, D. S. (1975). Steady state and a quantitative model of operant generalization and
discrimination. Journal of Experimental Psychology: Animal Behavior Processes,
1(1), 3-21. http://dx.doi.org/10.1037/0097-7403.1.1.3
Boddez, Y., Baeyens, F., Luyten, L., Vansteenwegen, D., Hermans, D., & Beckers, T. (2013)
Rating data are underrated: Validity of US expectancy in human fear conditioning.
https://doi.org/10.1016/j.jbtep.2012.08.003
Boddez, Y., Vervliet, B., Baeyens, F., Lauwers, S., Hermans, D., & Beckers, T. (2012).
Expectancy bias in a selective conditioning procedure: Trait anxiety increases the 241
threat value of a blocked stimulus. Journal of Behavior Therapy and Experimental
Psychiatry, 43, 832-837. https://doi.org/10.1016/j.jbtep.2011.11.005
Boelen, P. A., & Reijntjes, A. (2008). Intolerance of uncertainty and social anxiety. Journal
of Anxiety Disorders, 23(1), 130–135. http://dx.doi.org/10.1016/j.janxdis.2008.04.007
Bolles, R. C., Warren, J. A. (1965). The acquisition of bar press avoidance as a function of
shock intensity. Psychonomic Science, 3, 297. https://doi.org/10.3758/BF03343145
Bouton, M. E. (1984). Differential control by context in the inflation and reinstatement
paradigms. Journal of Experimental Psychology: Animal Behavior Processes, 10(1),
56-74. http://dx.doi.org/10.1037/0097-7403.10.1.56
Bouton, M. E. (1986). Slow reacquisition following the extinction of conditioned
suppression. Learning and Motivation, 17(1), 1-15.
https://doi.org/10.1016/0023-9690(86)90017-2
Bouton, M. E. (1993). Context, time, and memory retrieval in the interference paradigms of
Pavlovian learning. Psychological Bulletin, 114(1), 80-99.
http://dx.doi.org/10.1037/0033-2909.114.1.80
Bouton, M. E. (2007). Learning and behavior: A contemporary synthesis. Sunderland, MA,
US: Sinauer Associates.
Bouton, M. E., & Bolles, R. C. (1979). Contextual control of the extinction of conditioned
fear. Learning and Motivation, 10(4), 445-466.
https://doi.org/10.1016/0023-9690(79)90057-2
Bouton, M. E., & King, D. A. (1983). Contextual control of the extinction of conditioned
fear: Tests for the associative value of the context. Journal of Experimental
Psychology: Animal Behavior Processes, 9(3), 248-265.
http://dx.doi.org/10.1037/0097-7403.9.3.248 242
Bouton, M. E., & Peck, C. A. (1989). Context effects on conditioning, extinction, and
reinstatement in an appetitive conditioning preparation. Animal Learning & Behavior,
17(2), 188-198. http://dx.doi.org/10.3758/BF03207634
Bouton, M. E., & Ricker, S. T. (1994). Renewal of extinguished responding in a second
context. Animal Learning & Behavior, 22(3), 317-324.
https://doi.org/10.3758/BF03209840
Boyle, S., Roche, B., Dymond, S., & Hermans, D. (2016). Generalisation of fear and
avoidance along a semantic continuum, Cognition and Emotion, 30(2), 340-352.
https://doi.org/10.1080/02699931.2014.1000831
Brainard, D. H. (1997). The Psychophysics Toolbox. Spatial Vision, 10, 433–436.
Branca, A. A. (1957). Semantic generalization at the level of the conditioning experiment.
American Journal of Psychology, 70, 541–549. http://dx.doi.org/10.2307/1419444
Brennan, J. F., & Riccio, D. C. (1973). Stimulus control of avoidance behaviour in rats
following differential or nondifferential Pavlovian training along dimensions of the
conditioned stimulus. Journal of Comparative and Physiological Psychology, 85(2),
313-323. http://dx.doi.org/10.1037/h0035030
Brimer, C. J., & Dockrill, F. J. (1966). Partial reinforcement and the CER. Psychonomic
Science, 5, 185-186. https://doi.org/10.3758/BF03328344
Britton, J. C., Lissek, S., Grillon, C., Maxine, A., Norcross, B. S., & Pine, D. S. (2011).
Development of anxiety: The role of treat appraisal and fear learning. Depression and
Anxiety, 28(1), 5-17. https://dx.doi.org/10.1002/da.20733.
Broadbent, D. E., & Broadbent, M. (1988). Anxiety and attentional bias: State and trait.
Cognition and Emotion, 2, 165-183. http://dx.doi.org/10.1080/02699938808410922 243
Bronzaft, A., Hayes, R., Welch, L., & Koltuv, M. (1960). Relationship between PGR and
measures of extraversion, ascendance, and neuroticism. The Journal of Psychology:
Interdisciplinary and Applied, 50, 193-195.
http://dx.doi.org/10.1080/00223980.1960.9916435
Brotsky, S. J., & Keller, W. H. (1971). Semantic conditioning and generalization of the
galvanic skin response: Locus of mediation in classical conditioning. Journal of
Experimental Psychology, 89(2), 383-389. http://dx.doi.org/10.1037/h0031172
Brown, J. S. (1961). The motivation of behaviour. New York: McGraw-Hill.
Brown, J. S., Kalish, H. I., & Farber, I. E. (1951). Conditioned fear as revealed by magnitude
of startle response to an auditory stimulus. Journal of Experimental Psychology,
41(5), 317-328. http://dx.doi.org/10.1037/h0060166
Brown, P. L., & Jenkins, H. M. (1968). Auto-shaping of the pigeon’s key-peck. Journal of
Experimental Analysis of Behavior, 11(1), 1-8. https://doi.org/10.1901/jeab.1968.11-1
Brown, T. A., Chorpita, B. F., Korotitsch, W., & Barlow, D. H. (1997). Psychometric
properties of the Depression Anxiety Stress Scales (DASS) in clinical samples.
Behaviour Research and Therapy, 35, 79-89.
https://doi.org/10.1016/S0005-7967(96)00068-X
Burdick, C. K., & James, J. P. (1970). Spontaneous recovery of conditioned suppression of
licking by rats. Journal of Comparative and Physiological Psychology, 72(3), 467-
470. http://dx.doi.org/10.1037/h0029749
Burstein, K. R., Epstein, S., & Smith, B. (1967). Primary stimulus generalization of the GSR
as a function of objective and subjective definition of the stimulus dimension. Journal
of Experimental Psychology, 74(1), 124-131. http://dx.doi.org/10.1037/h0024512 244
Butler, A. C., Chapman, J. E., Forman, E. M., & Beck, A. T. (2006). The empirical status of
cognitive-behavioral therapy: A review of meta-analyses. Clinical Psychology
Review, 26(1), 17-31. https://doi.org/10.1016/j.cpr.2005.07.003
Butler, G., & Mathews, A. (1983). Cognitive processes in anxiety. Advances in Behaviour
Research and Therapy, 5, 51-62. http://dx.doi.org/10.1016/0146-6402(83)90015-2
Butler, G., & Mathews, A. (1987). Anticipatory anxiety and risk perception. Cognitive
Therapy and Research, 11, 551-555. https://doi.org/10.1007/BF01183858
Byrne, A., & Eysenck, M. W. (1993). Individual differences in positive and negative
interpretive biases. Personality and Invidividual Differences, 14(6), 849-851.
https://doi.org/10.1016/0191-8869(93)90100-H
Caci, H., Baylem F. J., Dossios, C., Robert, P., & Boyer, P. (2003). The Spielberger trait
anxiety inventory measures more than anxiety. European Psychiatry, 18, 394-400.
https://doi.org/10.1016/j.eurpsy.2003.05.003
Cahill, S. P., & Foa, E. B. (2007). Psychological theories of PTSD. In M. J. Friedman, T. M.
Keane, & P. A. Resick (Eds.), Handbook of PTSD: Science and Practice. (pp. 55-77).
New York: The Guilford Press.
Calvo, M. G., & Eysenck, M W. (1995). Interpretive bias in evaluation anxiety. Ansiedad y
Estres, 1, 5-20. https://dx.doi.org/X1130527414590368
Calvo, M. G., & Ramos, P. M. (1989). Effects of test anxiety on motor learning: The
processing efficiency hypothesis. Anxiety, Stress & Coping, 2(1), 45-55.
https://doi.org/10.1080/08917778908249325
Carey, S. (1985). Conceptual change in childhood. Cambridge, MA: MIT Press. 245
Carr, A. T. (1974). Compulsive neurosis: A review of the literature. Psychological Bulletin,
81(5), 311-318. http://dx.doi.org/10.1037/h0036473
Cattell, R. B., & Scheier, I. H. (1958). The nature of anxiety: A review of thirteen multi-
variate analyses comprising 814 variables. Psychological Reports, 4, 351-388.
http://dx.doi.org/10.2466/PR0.4.3.351-388
Cattell, R. B., & Scheier, I. H. (1961). The meaning and measurement of neuroticism and
anxiety. New York: Ronald Press.
Caul, W. F., & Miller, R. E. (1968). Effects of shock probability on heart rate of rats during
classical conditioning. Physiology & Behavior, 3(6), 865-869.
https://doi.org/10.1016/0031-9384(68)90169-8
Chambers, J. A., Power, K. G., & Durham, R. C. (2004). The relationship between trait
vulnerability and anxiety and depressive diagnoses at long-term follow-up of
Generalized Anxiety Disorder. Journal of Anxiety Disorders, 18(5), 587-607.
https://doi.org/10.1016/j.janxdis.2003.09.001
Chan, C. K. Y., & Lovibond, P. F. (1996). Expectancy bias in trait anxiety. Journal of
Abnormal Psychology, 105(4), 637-647.
http://dx.doi.org/10.1037/0021-843X.105.4.637
Chen, J. T. H., & Lovibond, P. F. (2016). Intolerance of uncertainty is associated with
increased threat appraisal and negative affect under ambiguity but not uncertainty.
Behavior Therapy, 47(1), 42-53. https://doi.org/10.1016/j.beth.2015.09.004
Chumbley, J. R., Flandin, G., Bach, D. R., Daunizeau, J., Fehr, E., Dolan, R. J., & Friston, K.
J. (2012). Learning and generalization under ambiguity: An fMRI study. PLOS
Computational Biology 8(1): e1002346. https://doi.org/10.1371/journal.pcbi.1002346 246
Clark, D. M., Salkovskis, P. M., Gelder, M., Koehler, K., Martin, M., Anastasiades, P.,
Hackmann, A., Middleton, H., & Jeavons, A. (1988). Tests of a cognitive theory of
panic. In I. Hand & H-U. Wittchen (Eds.), Panic and phobias (Vol. 2). Berlin:
Springer.
Chater, N. (2003). How much can we learn from double dissociations? Cortex, 39, 167-169.
http://dx.doi.org/10.1016/S0010-9452(08)70093-5
Clark, L. A., & Watson, D. (1991). Tripartite model of anxiety and depression: Psychometric
evidence and taxonomic implications. Journal of Abnormal Psychology, 100(3), 316-
336. http://dx.doi.org/10.1037/0021-843X.100.3.316
Clark, L. A., Watson, D., & Mineka, S. (1994). Temperament, personality, and the mood and
anxiety disorders. Journal of Abnormal Psychology, 103(1), 103-116.
Clark, R. E., & Squire, L. R. (1998). Classical conditioning and brain systems: The role of
awareness. Science, 280, 77-81. http://dx.doi.org/10.1016/S1364-6613(02)02041-7
Colgan, D. M. (1970). Effects of instructions on the skin resistance response. Journal of
Experimental Psychology, 86(1), 108-112. http://dx.doi.org/10.1037/h0030011
Cook, S. A., & Harris, R. E. (1937). The verbal conditioning of the galvanic skin
reflex. Journal of Experimental Psychology, 21(2), 202-210.
http://dx.doi.org/10.1037/h0063197
Corcoran, D. W. J. (1964). The relation between introversion and salivation. American
Journal of Psychology, 77, 298-300. http://dx.doi.org/10.2307/1420140
Craske, M. G. (1999). Anxiety disorders: psychological approaches to theory and treatment.
Boulder, CO: Westview Press. 247
Craske, M. G., Liao, B., Brown, L., & Vervliet, B. (2012). Role of inhibition in exposure
therapy. Journal of Experimental Psychopathology, 3, 322-345.
http://dx.doi.org/10.5127/jep.026511
Craske, M. G., & Mystkowski, J. L. (2006). Exposure Therapy and Extinction: Clinical
Studies. In M. G. Craske, D. Hermans, & D. Vansteenwegen (Eds.), Fear and
learning: From basic processes to clinical implications (pp. 217-233). Washington,
DC, US: American Psychological Association. http://dx.doi.org/10.1037/11474-011
Cross, D. V., & Lane, H. L. (1962). On the discriminative control of concurrent responses:
The relations among response frequency, latency, and topography in auditory
generalization. Journal of Experimental Analysis of Behaviour, 5, 487-496.
https://doi.org/10.1901/jeab.1962.5-487
Culler, E. A. (1938). Recent advances in some concepts of conditioning. Psychological
Review, 45, 134-153. http://dx.doi.org/10.1037/h0057483
Davey, G. C. L. (1987). An integration of human and animal models of Pavlovian
conditioning: Associations, cognitions, and attributions. In G. C. Davey (Ed.),
Cognitive processes and Pavlovian conditioning in humans (pp. 83-114). Chichester:
John Wiley and Sons.
Davey, G. C. L. (1989). Dental phobias and anxieties: Evidence for conditioning processes in
the acquisition and modulation of a learned fear. Behaviour Research and Therapy,
27(1), 51-58. https://doi.org/10.1016/0005-7967(89)90119-8
Davey, G. C. L. (1997). A conditioning model of phobias. In G. C. L. Davey (Ed.), Phobias:
A handbook of theory, research, and treatment (pp. 301–322). Chichester, England:
Wiley. 248
Davey, G. C., & Singh, J. (1988). The Kamin "blocking" effect and electrodermal
conditioning in humans. Journal of Psychophysiology, 2(1), 17-25.
Davis, M. (1990). Animal models of anxiety based on classical conditioning: The conditioned
emotional response (CER) and the fear-potentiated startle effect. Pharmacology &
Therapeutics, 47, 147-165. https://doi.org/10.1016/0163-7258(90)90084-F
Davis, M. (1992). The role of the amygdala in fear and anxiety. Annual Review of
Neuroscience, 15, 353-375.
Davis, M., Falls, W. A., & Gewirtz, J. (2000) Neural Systems Involved in Fear Inhibition:
Extinction and Conditioned Inhibition. In M. S., Myslobodsky, I. Weiner (Eds.)
Contemporary Issues in Modeling Psychopathology. Neurobiological Foundation of
Aberrant Behaviors, (Vol 1., pp. 113-141). Springer, Boston, MA.
Dawkins, K., Furnham, A. (1989). The colour naming of emotional words. British Journal of
Psychology, 80(3), 383-389. https://doi.org/10.1111/j.2044-8295.1989.tb02328.x
Dawson, M. E. (1970). Cognition and conditioning: Effects of masking the CS-UCS
contingency on human GSR classical conditioning. Journal of Experimental
Psychology, 85, 389-396. http://dx.doi.org/10.1037/h0029715
Dawson, M. E., & Biferno, M. A. (1973). Concurrent measurement of awareness and
electrodermal classical conditioning. Journal of Experimental Psychology, 101(1), 55-
62. http://dx.doi.org/10.1037/h0035524
Dawson, M. E., & Reardon, D. P. (1973). Construct validity of recall and recognition
postconditioning measures of awareness. Journal of Experimental Psychology, 98,
308-315. http://dx.doi.org/10.1037/h0034372 249
Dawson, M. E., & Schell, A. M. (1985). Information processing and human autonomic
classical conditioning. In P. K. Ackles et al. (Eds), Advances in psychophysiology,
Vol 1 (pp.89-165). JAI Press, Greenwich, CT.
De Houwer, J. (2009). The propositional approach to associative learning as an alternative for
association formation models. Learning & Behavior, 37(1), 1–20.
http://dx.doi.org/10.3758/LB.37.1.1
De Houwer, J., Beckers, T., & Glautier, S. (2002). Outcome and cue properties modulate
blocking. The Quarterly Journal of Experimental PsychologyA: Human Experimental
Psychology, 55A(3), 965-985. https://doi.org/10.1080/02724980143000578
De Jongh, A., Muris, P., Ter Horst, G., & Duyx, M. P. M. A. (1995). Acquisition and
maintenance of dental anxiety: The role of conditioning experiences and cognitive
factors. Behaviour Research and Therapy, 33(2), 205-210.
https://doi.org/10.1016/0005-7967(94)P4442-W
Del-Ben, C. M., Vilela, J. A., Hetem, L. A., Guimarães, F. S., Graeff, F. G., & Zuardi, A. W.
(2001). Do panic patients process unconditioned fear vs. conditioned anxiety
differently than normal subjects? Psychiatry Research, 104(3), 227–237.
https://doi.org/10.1016/S0165-1781(01)00312-2
Derakshan, N., & Eysenck, M. W. (1997). Interpretive biases for one’s own behaviour and
physiology in high trait-anxious individuals and repressors. Journal of personality
and Social Psychology, 73(4), 816-825. http://dx.doi.org/10.1037/0022-3514.73.4.816
Derakshan, N., & Eysenck, M. W. (1998). Working memory capacity in high trait-anxious
and repressor groups. Cognition and Emotion, 12(5), 697-713.
http://dx.doi.org/10.1080/026999398379501 250
De Ruiter, C., & Brosschot, J. F. (1994). The emotional Stroop interference effect in anxiety:
Attentional bias or cognitive avoidance? Behaviour Research and Therapy, 32, 315–
319. https://doi.org/10.1016/0005-7967(94)90128-7
Desiderato, O. (1964). Generalization of acquired fear as a function of CS intensity and
number of acquisition trials. Journal of Experimental Psychology, 67, 41-47.
http://dx.doi.org/10.1037/h0045113
Dibbets, P., van den Broek, A., & Evers, E. A. T. (2014). Fear conditioning and extinction in
anxiety- and depression-prone persons. Memory, 23(3), 350-364.
https://doi.org/10.1080/09658211.2014.886704
Dilger, S., Straube, T., Mentzel, H. J., Fitzek, C., Reichenbach, J. R., Hecht, H., et al. (2003).
Brain activation to phobia-related pictures in spider phobic humans: an event-related
functional magnetic resonance imaging study. Neuroscience Letters, 348, 29–32.
http://dx.doi.org/10.1016/S0304-3940(03)00647-5
Dirikx, T., Hermans, D., Vansteenwegen, D., Baeyens, F., & Eelen, P. (2007). Reinstatement
of conditioned responses in human differential fear conditioning. Journal of Behavior
Therapy and Experimental Psychiatry, 38(3), 237-251.
https://doi.org/10.1016/j.jbtep.2006.04.001
Dimmick, F. L., Ludlow, N., & Whiteman, A. (1939). A study of ‘experimental neurosis’ in
cats. Journal of Comparative Psychology, 28(1), 39-43.
http://dx.doi.org/10.1037/h0058856
Doll, T. J., & Thomas, D. R. (1967). Effects of discrimination training on stimulus
generalization for human subjects. Journal of Experimental Psychology, 75(4), 508-
512. http://dx.doi.org/10.1037/h0025116 251
Donner, N. C., & Lowry, C. A. (2013). Sex differences in anxiety and emotional behaviour.
Pflugers Archiv, 465(5), 601-626. http://dx.doi.org/10.1007/s00424-013-1271-7
Dougher, M. J., Augustson, E., Markham, M. R., Greenway, D. E., & Wulfert, E. (1994). The
transfer of respondent eliciting and extinction functions through stimulus equivalence
classes. Journal of the Experimental Analysis of Behavior, 62(3), 331–351.
http://doi.org/10.1901/jeab.1994.62-331
Dugas, M. J., Gagnon, F., Ladouceur, R., & Freeston, H. (1998). Generalized anxiety
disorder: A preliminary test of a conceptual model. Behaviour Research and Therapy,
36(2), 215–226. http://dx.doi.org/10.1016/S0005-7967(97)00070-3
Dugas, M. J., Hedayati, M., Karavidas, A., Buhr, K., Francis, K., & Phillips, N. A. (2005).
Intolerance of uncertainty and information processing: Evidence of biased recall and
interpretations. Cognitive Therapy and Research, 29, 57–70.
http://dx.doi.org/10.1007/s10608-005-1648-9
Dugas, M. J., & Robichaud, M. (2007). Cognitive-behaviour treatment for generalized
anxiety disorder: From science to practice. New York: Routledge.
Duits, P., Cath, D. C., Lissek, S., Hox, J. J., Hamm, A. O., Engelhard, I. M., ... Baas, J. M. P.
(2015). Updated meta-analysis of classical fear conditioning in the anxiety disorders.
Depression and Anxiety, 32(4), 239–253. https://dx.doi.org/10.1002/da.22353.
DuPont, R. L., Rice, D. P., Miller, L. S., Shiraki, S. S., Rowland, C. R., & Harwood, H. J.
(1996). Economic costs of anxiety disorders. Anxiety, 2, 167−172.
http://dx.doi.org/10.1002/(SICI)1522-7154(1996)2:4<167::AID-ANXI2>3.0.CO;2-L
Dunn, B. D., Billotti, D., Murphy, V., & Dalgleish, T. (2009). The consequences of effortful
emotion regulation when processing distressing material: A comparison of 252
suppression and acceptance. Behaviour Research and Therapy, 47(9), 761-773.
http://dx.doi.org/10.1016/j.brat.2009.05.007
Dunsmoor, J. E., Campese, V. D., Ceceli, A. O., LeDoux, J. E., & Phelps, E. A. (2015).
Novelty-facilitated extinction: Providing a novel outcome in place of an expected
threat diminishes recovery of defensive responses. Biological Psychiatry, 78, 203-
209. http://dx.doi.org/10.1016/j.biopsych.2014.12.008
Dunsmoor, J. E., Martin, A., & LaBar, K. S. (2012). Role of conceptual knowledge in
learning and retention of conditionedfear. Biological Psychology, 89, 300–305.
http://dx.doi.org/10.1016/j.biopsycho.2011.11.002
Dunsmoor, J. E., & Murphy, G. L. (2014). Stimulus typicality determines how broadly fear is
generalized. Psychological Science, 25(9), 1816-1821.
https://doi.org/10.1177/0956797614535401
Dunsmoor, J. E., & Paz, R. (2015). Fear generalization and anxiety: Behavioral and neural
mechanisms. Biological Psychiatry, 78(5), 336-343.
https://doi.org/10.1016/j.biopsych.2015.04.010
Dyal, J. A., & Goodman, E. D. (1966). Fear conditioning as a function of CS duration during
acquisition and suppression tests. Psychonomic Science, 4(7), 249-250.
http://dx.doi.org/10.3758/BF03342279
Dymond, S., Dunsmoor, J. E., Vervliet, B., Roche, B., & Hermans, D. (2015). Fear
generalization in humans: Systematic Review and Implications for anxiety disorder
research. Behavior Therapy, 46(5), 561-582.
https://doi.org/10.1016/j.beth.2014.10.001
Dymond, S., & Roche, B. (2009). A contemporary behavior analysis of anxiety and
avoidance. The Behavior Analyst, 32(1), 7–27. https://doi.org/10.1007/BF03392173 253
Dymond, S., Schlund, M. W., Roche, B., Whelan, R., Richards, J., & Davies, C. (2011).
Inferred threat and safety: Symbolic generalization of human avoidance learning.
Behaviour Research and Therapy, 49(10), 614-621.
https://doi.org/10.1016/j.brat.2011.06.007
Eberly, R. E., Harkness, A. R., & Engdahl, B. E. (1991). An adaptational view of trauma
response as illustrated by the prisoner of war experience. Jounal of Traumatic Stress,
4(3), 363-380. http://dx.doi.org/10.1002/jts.2490040305
Effting, M., & Kindt, M. (2007). Contextual control of human fear associations in a renewal
paradigm. Behaviour Research and Therapy, 45(9), 2002-2018.
https://doi.org/10.1016/j.brat.2007.02.011
Ekman, P. (1992). An argument for basic emotions. Cognition & Emotion 6, 169–200.
http://dx.doi.org/10.1080/02699939208411068
Ellis, W. R. III (1970). Role of stimulus sequences in stimulus discrimination and stimulus
generalization. Journal of Experimental Psychology, 83, 155-163.
http://dx.doi.org/10.1037/h0028587
Ellsberg, D. (1961). Risk, ambiguity, and the savage axioms. The Quarterly Journal of
Economics, 75(4), 643-669. http://www.jstor.org/stable/1884324
Ellsberg, D. (2001). Risk, ambiguity and decision. New York: Garlang Publishing Inc.
Estes, W. K., & Skinner, B. F. (1941). Some quantitative properties of anxiety. Journal of
Experimental Psychology, 29(5), 390-400. http://dx.doi.org/10.1037/h0062283
Eysenck, H. J. (1947). Dimensions of Personality. New York, Praeger.
Eysenck, H. J. (1957). The Dynamics of Anxiety and Hysteria. Praeger, New York. 254
Eysenck, H. J. (1959). The differentiation between normal and various neurotic groups on the
Maudsley personality inventory. British Journal of Psychology, 50(2), 176-177.
https://doi.org/10.1111/j.2044-8295.1959.tb00694.x
Eysenck, H. J. (1967). The biological basis of personality. Springfield, Illinois.
Eysenck, H. J., & Eysenck, S. B. G. (1975). Manual of the Eysenck Personality
Questionnaire. London: Hodder and Stoughton.
Eysenck, M. W. (1997). Anxiety and cognition: A unified theory. Psychology Press.
Eysenck, M. W. (1989a). Personality, stress arousal, and cognitive processes in stress
transactions. In R. W. J. Neufeld (Ed.), Advances in the Investigation of Psychological
Stress. New York: Wiley.
Eysenck, M. W. (1989b). Stress, anxiety, and intelligent performance. In D. Vickers and P. L.
Smith (Eds.), Human Information pProcessing: Measures, Mechanisms and Models.
Amsterdam: North-Holland.
Eysenck, M. W., & Calvo, M. G. (1992). Anxiety and performance: The processing
efficiency theory. Cognition and Emotion, 6(6), 409-434.
http://dx.doi.org/10.1080/02699939208409696
Eysenck, M. W., Derakshan, N., Santos, R., & Calvo, M. G. (2007). Anxiety and cognitive
performance: Attentional control theory. Emotion, 7(2), 336-353.
http://dx.doi.org/10.1037/1528-3542.7.2.336
Eysenck, M. W., MacLeod, C., & Mathews, A. (1987). Cognitive functioning and anxiety.
Psychological Research, 49, 189-195. https://doi.org/10.1007/BF00308686 255
Fahrenberg, J., & Foerster, F. (1982). Covariation and consistency of activation
parameters. Biological Psychology, 15(3-4), 151-169.
http://dx.doi.org/10.1016/0301-0511(82)90039-4
Fahrenberg, J., Walschburger, P., Foerster, F., Myrtek, M., & Müller, W. (1983). An
evaluation of trait, state, and reaction aspects of activation
processes. Psychophysiology, 20(2), 188-195.
http://dx.doi.org/10.1111/j.1469-8986.1983.tb03286.x
Fathi-Ashtiani, A., Ejei, J., Khodapanahi, M-K., & Tarkhorani, H. (2007). Relationship
between self-concept, self-esteem, anxiety, depression and academic achievement in
adolecents. Journal of Applied Sciences, 7, 995-1000.
http://dx.doi.org/10.3923/jas.2007.995.1000
Feather, B. W. (1965). Semantic generalization of classically conditioned responses: A
review. Psychological Bulletin, 63(6), 425-441. http://dx.doi.org/10.1037/h0022003
Fetzner, M. G., Horswill, S. C., Boelen, P. A., & Carleton, R. N. (2013). Intolerance of
uncertainty and PTSD symptoms: Exploring the construct relationship in a
community sample with a heterogeneous trauma history. Cognitive Therapy and
Research, 37(4), 725-734. https://doi.org/10.1007/s10608-013-9531-6
Foa, E. B., Steketee, G., & Rothbaum, B. O. (1989). Behavioral/cognitive conceptualizations
of post-traumatic stress disorder. Behavior Therapy, 20(2), 155-176.
http://dx.doi.org/10.1016/S0005-7894(89)80067-X
Fox, E., Russo, R., Bowles, R., & Dutton, K. (2001). Do threatening stimuli draw or hold
visual attention in subclinical anxiety? Journal of Experimental Psychology: General,
130(4), 681-700. http://dx.doi.org/10.1037/0096-3445.130.4.681
Freud, S. (1920). A general introduction to psychoanalysis. New York: Boni and Liveright. 256
Friedman, H., & Guttman, N. (1965). A further analysis of effects of discrimination training
on stimulus generalization gradients. In D. I. Mostofsky (Ed.), Stimulus
Generalization (pp.255-267). Standford: Stanford University Press.
Frijda, N.H. (1986) The Emotions. Cambridge University Press, Cambridge.
Fuhrer, M. J., & Baer, P. E. (1969). Cognitive processes in differential GSR conditioning:
Effects of a masking task. American Journal of Psychology, 82, 168-180.
http://dx.doi.org/10.2307/1421240
Furedy, J. J., & Schiffmann, K. (1973). Concurrent measurement of autonomic and cognitive
processes in a test of the traditional discriminative control procedure for Pavlovian
electrodermal conditioning. Journal of Experimental Psychology, 100, 210-217.
http://dx.doi.org/10.1037/h0035461
Gantt, W. H. (1942). The origin and development of nervous disturbances experimentally
produced. The American Journal of Psychiatry, 98(4), 475-481.
https://doi.org/10.1176/ajp.98.4.475
Ganz, L., & Riesen, A. H. (1962). Stimulus generalization to hue in the darkreared macaque.
Journal of Comparative and Physiological Psychology, 55(1), 92-99.
http://dx.doi.org/10.1037/h0044987
Gazendam, F. J., Kamphuis, J. H., Kindt, M. (2013). Deficient safety learning characterizes
high trait anxious individuals. Biological Psychology, 92(2), 342-352.
https://doi.org/10.1016/j.biopsycho.2012.11.006
Gencoz, F., Gencoz, T., & Joiner, T. E., Jr. (2000). Physiological hyperarousal as a specific
correlate of symptoms of anxiety among young psychiatric inpatients. Social 257
Behavior and Personality, 28(4), 409-412.
http://dx.doi.org/10.2224/sbp.2000.28.4.409
Genheimer, H., Andreatta, M., Asan, E., & Pauli, P. (2017). Reinstatement of contextual
conditioned anxiety in virtual reality and the effects of transcutaneous vagus nerve
stimulation in humans. Scientific Reports, 7, 17886.
http://dx.doi.org/10.1038/s41598-017-18183-3
Gershuny, B. S., & Sher, K. J. (1998). The relation between personality and anxiety: Findings
from a 3-year prospective study. Journal of Abnormal Psychology, 107(2), 252-262.
http://dx.doi.org/10.1037/0021-843X.107.2.252
Ghirlanda, S., & Enquist, M. (2003). A century of generalization. Animal Behaviour, 66, 15-
36. http://dx.doi.org/10.1006/anbe.2003.2174
Ginsburg, G. S., Becker, E. M., Keeton, C. P., Sakolsky, D., Piacentini, J., Albano, A. M.,…
Kendall, P. C. (2014). Naturalistic follow-up of youths treated for pediatric anxiety
disorders. JAMA Psychiatry, 71(3), 310-318.
http://dx.doi.org/10.1001/jamapsychiatry.2013.4186
Glees, P., Cole, J., Whitty, C. W. M., & Cairns, H. (1950). The effects of lesions in the
cingular gyrus and adjacent areas in monkeys. Journal of Neurology, Neurosurgery
and Psychiatry, 13(3), 178-190.
Gloster, A. T., Rhoades, H. M., Novy, D., Klotsche, J., Senior, A., Kunik, M et al. (2008).
Psychometric properties of the Depression Anxiety and Stress Scale-21 in older
primary care patients. Journal of Affective Disorders, 110(3), 248-259.
https://doi.org/10.1016/j.jad.2008.01.023
Glotzbach, E., Ewald, H., Andreatta, M., Pauli, P., & Mühlberger, A. (2012). Contextual fear
conditioning predicts subsequent avoidance behaviour in a virtual reality environment. 258
Cognition and Emotion, 26, 1256-1272.
http://dx.doi.org/10.1080/02699931.2012.656581
Glotzbach-Schoon, E., Andreatta, M., Mühlberger, A., & Pauli, P. (2013). Context
conditioning in virtual reality as a model for pathological anxiety. E-Neuroforum, 4,
63-70. http://dx.doi.org/10.1007/s13295-013-0047-z
Glover, E. M., Jovanovic, T., Mercer, K. B., Kerley, K., Bradley, B., Ressler, K. J., &
Norrholm, S. D. (2012). Estrogen levels are associated with extinction deficits in
women with Posttraumatic Stress Disorder. Biological Psychiatry, 72(1), 19-24.
https://doi.org/10.1016/j.biopsych.2012.02.031
Goette, L., Bendahan, S., Thoresen, J., Hollis, F., & Sandi, C. (2015). Stress pulls us apart:
Anxiety leads to differences in competitive confidence under stress.
Psychoneuroendocrinology, 54, 115-123.
https://doi.org/10.1016/j.psyneuen.2015.01.019
Goldberg, A. E. (2006). Constructions at work: The nature of generalization in language.
Oxford: Oxford University Press.
Goldstein, G., van Kammen, W., Shelly, C., Miller, D. H., & van Kammon, D. P. (1987).
Survivors of imprisonment in Pacific theater during World War II. American Journal
of Psychiatry, 144, 1210-1213. http://dx.doi.org/10.1176/ajp.144.9.1210
Goldstein, M. L. (1960). Acquired drive strength as a joint function of shock intensity and
number of acquisition trials. Journal of Experimental Psychology, 60, 349-358.
http://dx.doi.org/10.1037/h0040798
Gonçalves, R., Pedrozo, A. L., Coutinho, E. S. F., Figueira, I., & Ventura, P. (2012) Efficacy
of virtual reality exposure therapy in the treatment of PTSD: A systematic review.
PLoS ONE, 7(12): e48469. https://doi.org/10.1371/journal.pone.0048469 259
Gormezano, I., Schneiderman, N., Deaux, E., & Fuentes, I. (1962). Nictitating membrane:
Classical conditioning and extinction in the albino rabbit. Science, 138(3536), 33-34.
http://dx.doi.org/10.1126/science.138.3536.33
Gottfried, J. A., Dolan, R. J. (2004). Human orbitofrontal cortex mediates extinction learning
while accessing conditioned representations of value. Nature Neuroscience, 7, 1144–
1152. http://dx.doi.org/10.1038/nn1314
Graham, B. M., & Daher, M. (2016). Estradiol and Progesterone have opposing roles in the
regulation of fear extinction in female rats. Neuropsychopharmacology, 31(3), 774-
780. http://dx.doi.org/10.1038/npp.2015.202
Gray, J. A. (1981). A critique of Eysenck’s theory of personality, In H. J. Eysenck (Eds.). A
model for personality (pp. 246-276). Berlin: Springer
Gray, J. A. (1982). The neuropsychology of anxiety. Oxford: Clarendon.
Gray, J. A., & McNaughton, N. (2000). The neuropsychology of anxiety: An enquiry into the
functions of the septo-hippocampal system. Oxford: Oxford University Press.
Greco, J. A., & Liberzon, I. (2016). Neuroimaging of fear-associated learning.
Neurospychopharmacology, 41, 320-324. http://dx.doi.org/10.1038/npp.2015.255
Greenberg, P. E., Sisitsky, T., Kessler, R. C., Finkelstein, S. N., Berndt, E. R., Davidson, J.
R., ..., & Fyer, A. J. (1999). The economic burden of anxiety disorders in the 1990s.
Journal of Clinical Psychiatry, 60, 427-435. http://dx.doi.org/10.4088/JCP.v60n0702
Grillon, C., & Ameli, R. (2001). Conditioned inhibition of fear-potentiated startle and skin
conductance in humans. Psychophysiology, 38, 807-815.
https://doi.org/10.1111/1469-8986.3850807 260
Grillon, C., Baas, J. M. P., Cornwell, B., & Johnson, L. (2006). Context conditioning and
behavioral avoidance in a virtual reality environment: Effect of predictability.
Biological Psychiatry, 60(7), 752-759. https://doi.org/10.1016/j.biopsych.2006.03.072
Grillon, C., & Morgan III, C. A. (1999). Fear-potentiated startle conditioning to explicit and
contextual cues in Gulf War veterans with posttraumatic stress disorder. Journal of
Abnormal Psychology, 108(1), 134-142.
http://dx.doi.org/10.1037/0021-843X.108.1.134
Grim, P. F., & White S. H. (1965). Effects of stimulus change upon the GSR and reaction
time. Journal of Experimental Psychology, 69, 276-281.
http://dx.doi.org/10.1037/h0021739
Grings, W. W., Schell, A. M., & Carey, C. A. (1973). Verbal control of an autonomic
response in a cue reversal situation. Journal of Experimental Psychology, 99(2), 215-
221. http://dx.doi.org/10.1037/h0034653
Guthrie, R. M., & Bryant, R. A. (2006). Extinction learning before trauma and subsequent
posttraumatic stress. Psychosomatic Medicine, 68(2),307-311.
http://dx.doi.org/10.1097/01.psy.0000208629.67653.cc
Guttman, N. (1953). Operant conditioning, extinction, and periodic reinforcement in relation
to concentration of sucrose used as reinforcing agent. Journal of Experimental
Psychology, 46(4), 213-224. http://dx.doi.org/10.1037/h0061893
Guttman, N., & Kalish, H. I. (1956). Discriminability and stimulus generalization. Journal of
Experimental Psychology, 51, 79-88. http://dx.doi.org/10.1037/h0046219
Haaker, J., Lonsdorf, T., Schumann, D., Brassen, S., Bunzeck, N., Gamer, M., & Kalisch, R.
(2015). Deficient inhibitory processing in trait anxiety: Evidence from context- 261
dependent fear learning, extinction recall and renewal. Biological Psychology, 111,
65-72. https://doi.org/10.1016/j.biopsycho.2015.07.010
Haberlandt, K., & Bingham, G. (1978). Verbs contribute to the coherence of brief narratives:
Reading related and unrelated sentence triples. Journal of Verbal Learning and
Verbal Behavior, 17, 419-425. http://dx.doi.org/10.1016/S0022-5371(78)90247-5
Haddad, A. D. M., Pritchett, D., Lissek, S., & Lau, J. Y. F. (2012). Trait anxiety and fear
resposnes to safety cues: Stimulus generalization or sensitization? Journal of
Psychopathology and Behavioral Assessment, 34(3), 323-331.
http://dx.doi.org/10.1007/s10862-012-9284-7
Hadwin, J., Frost, S., French, C. C., & Richards, A. (1997). Cognitive processing and trait
anxiety in typically developing children: Evidence for an interpretation bias. Journal
of Abnormal Psychology, 106(3), 486-490.
http://dx.doi.org/10.1037/0021-843X.106.3.486
Hagenaars, M. A. (2012). Anxiety symptoms influence the effect of post-trauma interventions
after analogue trauma. Journal of Experimental Psychopathology, 3(2), 209-222.
http://dx.doi.org/10.5127/jep.020311
Haney, J. N. (1973). Approach-avoidance reactions by repressors and sensitizers to ambiguity
in a structured free-association task. Psychological Reports, 33 97-98.
http://dx.doi.org/10.2466/pr0.1973.33.1.97
Hanson, H. M. (1959). Effects of discrimination training on stimulus generalization. Journal
of Experimental Psychology, 58(5), 321-334. http://dx.doi.org/10.1037/h0042606
Harris, C, A., & D’Eon, J. L. (2008). Psychometric properties of the Beck Depresion
Inventory-Second edition (BDI-II) in individuals with chronic pain. Pain, 137(3),
609-622. http://dx.doi.org/10.1016/j.pain.2007.10.022 262
Hart, J.D. (1974). Physiological responses of anxious and normal subjects to simple signal
and non-signal auditory stimuli. Psychophysiology, 11, 443-451.
http://dx.doi.org/10.1111/j.1469-8986.1974.tb00570.x
Hayes, B. K., Kurniawan, H., & Newell, B. R. (2011). Rich in vitamin C or just a convenient
snack? Multiple-category reasoning with cross-classified foods. Memory & Cognition,
39, 92-106. http://dx.doi.org/10.3758/s13421-010-0022-7
Hayes, B. K., & Newell, B. R. (2009). Induction with uncertain categories: When do people
consider the category alternatives? Memory and Cognition, 37, 730-743.
http://dx.doi.org/10.3758/MC.37.6.730
Hayes, B. K., & Thompson, S. P. (2007). Causal relations and feature similarity in children’s
inductive reasoning. Journal of Experimental Psychology: General, 136, 470-
484. http://dx.doi.org/10.1037/0096-3445.136.3.470
Hearst, E., & Koresko, M. B. (1968). Stimulus generalization and amount of prior training on
variable-interval reinforcement. Journal of Comparative Physiological Psychology,
66(1), 1133-138. http://dx.doi.org/10.1037/h0025992
Hebert, J. A. (1970). Context effects in the generalization of a successive discrimination in
human subjects. Canadian Journal of Psychology, 24, 271-275.
http://dx.doi.org/10.1037/h0082863
Heit, E. (2000). Properties of inductive reasoning. Psychonomic Bulletin & Review, 7(4),
569-592. https://doi.org/10.3758/BF03212996
Henry, J. D., & Crawford, J. R. (2005). The short-form version of the Depression Anxiety
Stress Scales (DASS-21): Construct validity and normative data in a large non-
clinical sample. British Journal of Clinical Psychology, 44, 227–239.
https://doi.org/10.1348/014466505X29657. 263
Hermann, C., Ziegler, S., Birbaumer, N., & Flor, H. (2002). Psychophysiological and
subjective indicators of aversive Pavlovian conditioning in generalized social phobia.
Biological Psychiatry, 52(4), 328-337.
https://doi.org/10.1016/S0006-3223(02)01385-9
Hermans, D., Dirikx, T., Vansteenwegenin, D., Baeyens, F., Van den Bergh, O., & Eelen, E.
(2005). Reinstatement of fear responses in human aversive conditioning. Behaviour
Research and Therapy, 43(4), 533-551. https://doi.org/10.1016/j.brat.2004.03.013
Heussen, D., Voorspoels, W., Verheyen, S., Storms, G., & Hampton, J. A. (2011). Raising
argument strength using negative evidence: A constraint on models of induction.
Memory & Cognition, 39, 1496-1507. https://doi.org/10.3758/s13421-011-0111-2
Hirschfeld, R. M. A., & Klerman, G. L. (1979). Personality attributes and affective disorders.
American Journal of Psychiatry, 136, 67-70. https://doi.org/10.1176/ajp.136.1.67
Hodges, W.F., & Spielberger, C.D. (1966). The effects of threat of shock on heart rate for
subjects who differ in manifest anxiety and fear of shock. Psychophysiology, 2, 287-
294. http://dx.doi.org/10.1111/j.1469-8986.1966.tb02656.x
Hofmann, S. G. (2008). Cognitive processes during fear acquisition and extinction in animals
and humans: Implications for exposure therapy of anxiety disorders. Clinical
Psychology Review, 28, 199-210. http://dx.doi.org/10.1016/j.cpr.2007.04.009
Hofmann, S. G., Asnaani, A., Vonk, I. J. J., Sawyer, A. T., & Fang, A. (2012). The effeicacy
of cognitive behavioral therapy: A review on meta-analyses. Cognitive Therapy and
Research,36, 427-440. http://dx.doi.org/10.1007/s10608-012-9476-1
Hofmann, S. G., & Smits, J. A. J. (2008). Cognitive-behavioral therapy for adult anxiety
disorders: A meta-analysis of randomized placebo-controlled trials. Journal of
Clinical Psychiatry, 69(4), 621-632. 264
Holmes, N. M., & Westbrook, R. F. (2014). ABA renewal is greater when extinction occurs
in the same context as cue pre-exposure. Journal of Experimental Psychology: Animal
Learning and Cognition, 40(3), 369-379. http://dx.doi.org/10.1037/xan0000024
Homa, D., Sterling, S., & Trepel, L. (1981). Limitations of exemplar-based generalization
and the abstraction of categorical information. Journal of Experimental Psychology:
Human Learning and Memory, 7(6), 418-439.
http://dx.doi.org/10.1037/0278-7393.7.6.418
Honig, W. K., Thomas, D. R., & Guttman, N. (1959). Differential effects of continuous
extinction and discrimination training on the generalization gradient. Journal of
Experimental Pschology, 58, 145-152. http://dx.doi.org/10.1037/h0048484
Honig, W. K., Boneau, C. A., Burstein, K. R., & Pennypacker, H. S. (1963). Positive and
negative generalization gradients obtained after equivalent training conditions.
Journal of Comparative and Physiological Psychology, 56(1), 111-116.
http://dx.doi.org/10.1037/h0048683
Hope, D. A., Rapee, R. M., Heimberg, R. G., & Dombeck, M. (1990). Representations of the
self in social phobia Vulnerability to social threat. Cognitive Therapy and Research,
14, 177-189. http://dx.doi.org/10.1007/BF01176208
Hovland, C. I. (1937). The generalization of conditioned responses: I. The sensory
generalization of conditioned response with varying frequencies of tone. Journal of
General Psychology, 17, 125-148. https://doi.org/10.1080/00221309.1937.9917977
Huff, N. C., Hernandez, J. A., Blanding, N. Q., & LaBar, K. S. (2009). Delayed extinction
attenuates conditioned fear renewal and spontaneous recovery in humans. Behavioral
Neuroscience, 123(4), 843-843. http://dx.doi.org/10.1037/a0016511 265
Huff, R. C., Sherman, J. E., & Cohn, M. (1975). Some effects of response-independent
reinforcement in auditory generalization gradients. Journal of the Experimental
Analysis of Behavior, 23, 81-86. https://doi.org/10.1901/jeab.1975.23-81
Hugdahl, K. (1981). The three-systems-model of fear and emotion: A critical
examination. Behaviour Research and Therapy, 19(1), 75-85.
http://dx.doi.org/10.1016/0005-7967(81)90114-5
Hull, C. L. (1934a). The concept of the habit-family hierarchy and maze learning: Part I.
Psychological Review, 41, 33-54. http://dx.doi.org/10.1037/h0070758
Hull, C. L. (1934b). The concept of the habit-family hierarchy and maze learning: Part II.
Psychological Review, 41, 134-152. http://dx.doi.org/10.1037/h0072855
Hull, C. L. (1943). Principles of Behavior. New York: Appleton Century Crofts.
Hull., C. L. (1949). Stimulus intensity dynamism (V) and stimulus generalization.
Psychological Review, 56, 67-76. http://dx.doi.org/10.1037/h0058051
Hull, C. L. (1952). A Behaviour System. New Haven: Yale University Press.
Humphreys, L. G. (1939). Generalization as a function of method of reinforcement. Journal
of Experimental Psychology, 25(4), 361-372. http://dx.doi.org/10.1037/h0057941
Indovina, I., Robbins, T. W., Nunez-Elizalde, A. O., Dunn, B. D., & Bishop, S. (2011). Fear-
conditioning mechanisms associated with trait vulnerability to anxiety in humans.
Neuron, 69, 563-571. http://dx.doi.org/10.1016/j.neuron.2010.12.034
Jenkins, H. M., & Harrison, R. H. (1960). Effect of discrimination training on auditory
generalization. Journal of Experimental Psychology, 59(4), 246-253.
http://dx.doi.org/10.1037/h0041661 266
Johnson-Laird, P. N. (2000). Thinking: Reasoning. In A. Kazdin (Ed.), Encyclopedia of
psychology (Vol. 8, pp. 75–79). Washington, DC: American Psychological
Association.
Joiner, T. E., Jr., Catanzaro, S. J., & Laurent, J. (1996). Tripartite structure of positive and
negative affect, depression, and anxiety in child and adolescent psychiatric
inpatients. Journal of Abnormal Psychology, 105(3), 401-409.
http://dx.doi.org/10.1037/0021-843X.105.3.401
Joiner, T. E., Jr., Steer, R. A., Beck, A. T., Schmidt, N. B., Rudd, M. D., & Catanzaro, S. J.
(1999). Physiological hyperarousal: Construct validity of a central aspect of the
tripartite model of depression and anxiety. Journal of Abnormal Psychology, 108(2),
290-298. http://dx.doi.org/10.1037/0021-843X.108.2.290
Jones, F., & McLaren, I. P. L. (1999). Rules and associations. In M. Hahn & M. C. Stoness
(Eds.), Proceedings of the 21st Annual Conference of the Cognitive Science Society
(pp. 240–245). Mahwah, NJ: Erlbaum.
Jorm, A. F., Christensen, H., Henderson, A. S., Jacomb, P. A., Korten, A. E., & Rodgers, B.
(2000). Predicting anxiety and depression from personality: Is there a synergistic
effect of neuroticism and extraversion? Journal of Abnormal Psychology, 109(1),
145-149. http://dx.doi.org/10.1037/0021-843X.109.1.145
Jovanovic, T., Norrholm, S. D., Blanding, N. Q., Davis, M., Duncan, E., Bradley, B., &
Ressler, K. J. (2010). Impaired fear inhibition is a biomarker of PTSD but not
depression. Depression and Anxiety, 27(3), 244-251.
http://dx.doi.org/10.1002/da.20663
Jovanovic, T., Norrholm, S. D., Fennell, J, E., Keyes, M., Fiallos, A. M., Myers, K. M.,
Davis, M., & Duncan, E. J. (2009). Posttraumatic stress disorder may be associated 267
with impaired fear inhibition: Relation to symptom severity. Psychiatry Research,
167(1), 151-160. http://dx.doi.org/10.1016/j.psychres.2007.12.014
Kaczkurkin, A. N., Burton, P. C., Chazin, S. M., Manbeck, A. B., Espen-Sturges, T., Cooper,
S. E., Sponheim, S. R., & Lissek, S. (2017). Neural substrates of overgeneralized
conditioned fear in PTSD. The American Journal of Psychiatry, 174(2), 125-134.
https://doi.org/10.1176/appi.ajp.2016.15121549
Kaczkurkin, A. N., Lissek, S. (2013). Generalization of conditioned fear and obsessive-
compulsive traits. Journal of Pszchology and Pszchotherapy, S7, 003.
http://dx.doi.org/10.4172/2161-0487.S7-003
Kalish, C. W. & Lawson, C. A. (2007). Negative evidence and inductive generalisation.
Thinking & Reasoning, 13, 394–425. https://doi.org/10.1080/13546780701273402
Kamin, L. J. (1969). Predictability, surprise attention and conditioning. In B. A. Campbell &
R. M. Church, (Eds.), Punishment and aversive behaviour, (pp.279-96). New York:
Appleton-Century-Crofts.
Kamin, L. J., Brimer, C. J., & Black, A. H. (1963). Conditioned suppression as a monitor of
fear of the CS in the course of avoidance training. Journal of Comparative and
Physiological Psychology, 56, 497-501. http://dx.doi.org/10.1037/h0047966
Kaufman, S. C. (2013). Opening up openness to experience: A four-factor model and
relations to creative achievement in the arts and sciences. The Journal of Creative
Behavior, 47(4), 233-255. http://dx.doi.org/10.1002/jocb.33
Keenan, J. M., Baillet, S. D., & Brown, P. (1984). The effects of causal cohesion on
comprehension and memory. Journal of Verbal Learning and Verbal Behavior, 23,
115-126. http://dx.doi.org/10.1016/S0022-5371(84)90082-3 268
Kelley, D., Brown, C.C., & Shaffer, J.W. (1970). A comparison of physiological and
psychological measurements on anxious patients and normal controls.
Psychophysiology, 6, 429-441. http://dx.doi.org/10.1111/j.1469-8986.1970.tb01753.x
Kelly, D., Richardson, A., & Mitchell-Heggs, N. (1973). Stereotactic limbic leucotomy:
Neurophysiological aspects and operative technique. British Journal of Psychiatry,
123, 133-140. http://dx.doi.org/10.1192/bjp.123.2.133
Kendler, K. S., Myers, J., Prescott, C. A., & Neale, M. C. (2001). The genetic epidemiology
of irrational fears and phobias in men. Archive of General Psychiatry, 58(3), 257-265.
http://dx.doi.org/10.1001/archpsyc.58.3.257
Kendler, K. S., Walters, E. E., Neale, M. C., Kessler, R. C., Heath, A. C., & Eaves, L. J.
(1995). The structure of the genetic and environmental risk factors for sic major
psychiatric disorders in women: Phobia, generalized anxiety disorder, panic disorder,
bulimia, major depression and alcoholism. Archive of General Psychiatry, 52(5), 374-
383. http://dx.doi.org/10.1001/archpsyc.1995.03950170048007
Kennedy, B. L., Schwab, J. J., Morris, R. L., & Beldia, G. (2001). Assessment of state and
trait anxiety in subjects with anxiety and depressive disorders. Psychiatric Quarterly,
72, 263-276. https://doi.org/10.1023/A:1010305200087
Kessler, R. C., & Greenberg, P. E. (2002). The economic burden of anxiety and stress
disorders. Neuropsychopharmacology: The fifth generation of progress, 67, 982-92.
Kimmel, H. D., & Bevill, M. J. (1996). Blocking and unconditional response diminution in
human classical autonomic conditioning. Integrative Physiological and Behavioral
Science, 31, 18–43. https://doi.org/10.1007/BF02691479
Kinder, A., & Shanks, D. R. (2001). Amnesia and the declarative/ nondeclarative distinction:
A recurrent network model of classification, recognition and repetition priming. 269
Journal of Cognitive Neuroscience, 13, 648-669.
https://doi.org/10.1162/089892901750363217
Kindt, M., Soeter, M. (2014). Fear inhibition in high trait anxiety. PLOS ONE, 9(1), e84642.
https://doi.org/10.1371/journal.pone.0086462
Knight, D. C., Smith, C. N., Cheng, D. T., Stein, E. A., & Helmstetter, F. J. (2004).
Amygdala and hippocampal activity during acquisition and extinction of human fear
conditioning. Cognitive, Affective & Behavioral Neuroscience, 4, 317–325.
http://dx.doi.org/10.3758/CABN.4.3.317
Knyazev, G.G., Slobodskaya, H.R., & Wilson, G.D. (2002). Psychophysiological correlates
of behavioural inhibition and activation. Personality and Individual Differences, 33,
647-660. http://dx.doi.org/10.1016/S0191-8869(01)00180-5
Koerner, N., & Dugas, M. J. (2008).An investigation of appraisals in individuals vulnerable
to excessive worry: The role of intolerance of uncertainty. Cognitive Therapy and
Research, 32, 619–638. http://dx.doi.org/10.1007/s10608-007-9125-2
Koster, E. H. W., Crombez, G., Verschuere, B., van Damme, S., & Wiersema, J. R. (2006).
Components of attentional bias to threat in high trait anxiety: Facilitated engagement,
impaired disengagement, and attentional avoidance. Behaviour Research and
Therapy, 44(12), 1757-1771. https://doi.org/10.1016/j.brat.2005.12.011
Kovach, J. K. (1971). Interaction of innate and acquired: Color preferences and early
exposure learning in chicks. Journal of Comparative and Physiological Psychology,
75(3), 386-398. http://dx.doi.org/10.1037/h0030940
Kroes, M. C. W., Dunsmoor, J. E., Mackey, W. E., McClay, M., & Phelps, E. A. (2017).
Context conditioning in humans using commercially available immersive virtual
reality. Scientific Reports, 7, 8640. http://dx.doi.org/s41598-017-08184-7 270
Kurtz, K. H., & Siegel, A. (1966). Conditioned fear and magnitude of startle response: A
replication and extension. Journal of Comparative and Physiological Psychology,
62(1), 8-14. http://dx.doi.org/10.1037/h0023485
LaBerge, D. (1961). Generalization gradients in a discrimination situation. Journal of
Experimental Psychology, 62(1), 88-94. http://dx.doi.org/10.1037/h0043035
Lacey, J. I., & Smith, R. L. (1954). Conditioning and generalization of unconscious anxiety.
Science, 120, 1045-1052. http://dx.doi.org/10.1126/science.120.3130.1045
Lader, M. H., & Wing, L. (1964). Habitation of the psycho-galvanic reflex in patients with
anxiety states and in normal subjects. Journal of Neurology, Neurosurgery &
Psychiatry, 27(3), 210-218. http://dx.doi.org/10.1136/jnnp.27.3.210
Lal, H., & Emmett-Oglesby, M. W. (1983). Behavioral analogues of anxiety. Animal models.
Neuropharmacology, 2212B), 1423-1441.
https://doi.org/10.1016/0028-3908(83)90111-9
Lang, P. J., Bradley, M. M., Cuthbert, B. N. (1998). Emotion, motivation, and anxiety: Brain
mechanisms and psychophysiology. Biological Psychiatry 44, 1248–1263
https://doi.org/10.1016/S0006-3223(98)00275-3
Lang, P. J., Davis, M., & Öhman, A. (2000). Fear and anxiety: Animal models and human
cognitive psychophysiology. Journal of Affective Disorders, 61(3), 137-159.
https://doi.org/10.1016/S0165-0327(00)00343-8
Lang, P. J., Geer, J., & Hnatiow, M. (1963). Semantic generalization of conditioned
autonomic responses. Journal of Experimental Psychology, 65(6), 552-558.
http://dx.doi.org/10.1037/h0049131
Larson, C. L., Schaefer, H. S., Siegle, G. J., Jackson, C. A., Anderle, M. J., & Davidson, R.
J. (2006). Fear is fast in phobic individuals: Amygdala activation in response to fear- 271
relevant stimuli. Biological. Psychiatry, 60, 410–417.
http://dx.doi.org/10.1016/j.biopsych.2006.03.079
Lashley, K. S., & Wade, M. (1946). The Pavlovian theory of generalization. Psychological
Review, 53(2), 72-87. http://dx.doi.org/10.1037/h0059999
Lawrence, C. (1973). Generalization along the dimension of sound intensity in
pigeons. Animal Learning & Behavior, 1(1), 60-64.
http://dx.doi.org/10.3758/BF03199002
Lazarus, R. D. (1966). Psychological stress and the coping process. New York:McGraw-Hill.
Lazarus, R. D. (1981). The stress and coping paradigm. In C. Eisdorfer, D. Cohen, A.
Kleinmen & P. Maxim (Eds.), Models for clinical psychopathology. New York:
Spectrum.
Lazarus, R. D. (1991). Emotion and adaptation. Oxford: Oxford University Press.
Lebron-Milad, K., Abbs, B., Milad, M. R., Linnman, C., Rougemount- Bücking, A., Zeidan,
M. A., Holt, D. J., & Goldstein, J. M. (2012). Sex differences in the neurobiology of
fear conditioning and extinction: A preliminary fMRI study of shared sex differences
with stress-arousal circuitry. Biology of Mood & Anxiety Disorders, 2:7.
https://doi.org/10.1186/2045-5380-2-7
LeDoux, J. E. (2000). Emotion circuits in the brain. Annual Review of Neuroscience, 23,
155–184. http://ds.doi.org/10.1146/annurev.neuro.23.1.155
LeDoux, J.E. (2012). Rethinking the emotional brain. Neuron, 73, 653–676.
http://dx.doi.org/10.1016/j.neuron.2012.02.004 272
Ledgerwood, L., Richardson, R., & Cranney, J. (2003). Effects of D-cycloserine on
extinction of conditioned freezing. Behavioral Neuroscience, 117(2), 341-349.
http://dx.doi.org/10.1037/0735-7044.117.2.341
Lee, Y. C., Chatterton, M. L., Magnus, A., Mohebbi, M., Le, L. K., & Mihalopoulos, C.
(2017). Cost of high prevalence mental disorders: Findings from the 2007 Australian
National Survey of Mental Health and Wellbeing. Australian & New Zealand Journal
of Psychiatry, 51(12), 1198-1211. https://doi.org/10.1177/0004867417710730
Lee, J. C., Hayes, B. K., & Lovibond, P. F. (2018). Peak shift and rules in human
generalization. Journal of Experimental Psychology: Learning, Memory and
Cognition. http://dx.doi.org/10.1037/xlm0000558
Lenaert, B., Boddez, Y., Griffith, J. W., Vervliet, B., Schruers, K., & Hermans, D. (2014).
Aversive learning and generalization predict subclinical levels of anxiety: A six-
month longitudinal study. Journal of Anxiety Disoders, 28, 747-753.
https://doi.org/10.1016/j.janxdis.2014.09.006.
Levey, A., & Martin, I. (1991). Personality and conditioning. In H. J. Eysenck (Ed.), A model
for personality (pp. 123-168). Berlin: Springer.
Lissek, S., Biggs, A. L., Rabin, S. J., Cornwell, B. R., Alvarez, R. P., Pine, D. S., & Grillon,
C. (2008b). Generalization of conditioned fear-potentiated startle in humans;
Experimental validation and clinical relevance. Behaviour and Research and Therapy,
46, 678-687. https://doi.org/10.1016/j.brat.2008.02.005
Lissek, S., Kaczkurkin, A. N., Rabin, S., Geraci, M., Pine, D. S., & Grillon, C. (2014).
Generalized anxiety disorder is associated with over-generalization of classically
conditioned fear. Biological Psychiatry, 75, 909-915.
http://dx.doi.org/10.1016/j.biopsych.2013.07.025 273
Lissek, S., Levenson, J., Biggs, A. L., Hohnson, L. L., Ameli, R., Pine, D. S., & Grillon, C.
(2008a). Elevated fear conditioning to socially relevant unconditioned stimuli in social
anxiety disorder. American Journal of Psychiatry, 165(1), 124-132.
https://doi.org/10.1176/appi.ajp.2007.06091513
Lissek, S., Pine, D. S., & Grillon, C. (2006). The strong situation: A potential impediment to
studying the psychobiology and pharmacology of anxiety disorders. Biological
Psychology, 72(3), 265–270. https://doi.org/10.1016/j.biopsycho.2005.11.004
Lissek, S., Powers, A. S., McClure, E. B., Phelps, E. A., Woldenhawariat, G., Grillon, C., et
al. (2005). Classical fear conditioning in the anxiety disorders: A meta-analysis.
Behaviour Research and Therapy, 43(11), 1391–1424.
https://doi.org/10.1016/j.brat.2004.10.007.
Lissek, S., Rabin, S., Heller, R. E., Lukenbaugh, D., Geraci, M., Pine, D. S., & Grillon, C.
(2010). Overgeneralization of conditioned fear as a pathogenic marker of panic
disorder. American Journal of Psychiatry, 167, 47-55.
https://doi.org/10.1176/appi.ajp.2009.09030410
Lissek, S., Rabin, S. J., McDowell, D. J., Dvir, S., Bradford, D. E., Geraci, M., Pine, D. S., &
Grillion, C. (2009). Impaired discriminative fear-conditioning resulting from elevated
fear responding to learned safety cues among individuals with panic disorder.
Behaviour Research and Therapy, 47(2), 111-118.
https://doi.org/10.1016/j.brat.2008.10.017
Littman, R. A. (1949). Conditioned generalization of the galvanic skin reaction to tones.
Journal of Experimental Psychology, 39(6), 868-882.
http://dx.doi.org/10.1037/h0055577 274
Livesey, E. J., & McLaren, I. P. L. (2009). Discrimination and generalization along a simple
dimension: Peak shift and rule-governed responding. Journal of Experimental
Psychology: Animal Behavior Processes, 35(4), 554-565.
http://dx.doi.org/10.1037/a0015524
Livesey, E. J., & McLaren, I. P. L. (2018). Revisiting peak shift on an artificial dimension:
Effects of stimulus variability on generalisation. Quarterly Journal of Experimental
Psychology. https://doi.org/10.1177/1747021817739832
Lolordo, V. M. (1967). Similarity of conditioned fear responses based upon different aversive
events. Journal of Comparative and Physiological Psychology, 64(1), 154-158.
http://dx.doi.org/10.1037/h0024809
Lommen, M. J. J., Engelhard, I. M., & van den Hout, M. A. (2010). Neuroticism and
avoidance of ambiguous stimuli; Better safe than sorry? Personality and Individual
Differences, 49, 1001-1006. https://doi.org/10.1016/j.paid.2010.08.012
Lopez, A., Gelman, S. A., Gutheil, G., & Smith, E. E. (1992). The development of category-
based induction. Child Development, 63(5), 1070-1090.
https://doi.org/10.1111/j.1467-8624.1992.tb01681.x
Lovejoy, E., & Russell, D. G. (1967). Suppression of learning about a heard cue by the
presence of an easy cue. Psychonomic Science, 8, 365-366.
https://doi.org/10.3758/BF03332242
Lovibond, P. F. (1992). Tonic and phasic electrodermal measures of human aversive
conditioning with long duration stimuli, Psychophysiology, 29(6), 621-632.
https://doi.org/10.1111/j.1469-8986.1992.tb02037.x 275
Lovibond, P. F. (1998). Long-term stability of depression, anxiety, and stress syndromes.
Journal of Abnormal Psychology, 107(3), 520-526.
http://dx.doi.org/10.1037/0021-843X.107.3.520
Lovibond, P. F. (2004). Cognitive processes in extinction. Learning and Memory, 11, 495-
500. http://dx.doi.org/10.1101/lm.79604
Lovibond, P. F. (2011). Learning and anxiety: A cognitive perspective. In T. R. Schachtman,
& S. Reilly (Eds.), Associative learning and conditioning theory: Human and non-
human applications (pp. 104-120). Oxford University Press.
Lovibond, P.F., Been, S. L., Mitchell, C. J., Bouton, M. E., & Frohardt, R. (2003). Forward
and backward blocking of causal judgment is enhanced by additivity of effect
magnitude, Memory and Cognition, 31. 133 – 142.
https://doi.org/10.3758/BF03196088
Lovibond, P. F., Davis, N, R., & Flaherty, A. S. (2000). Protection from extinction in human
fear conditioning. Behaviour Research and Therapy, 38, 967-983.
http://dx.doi.org/10.1016/S0005-7967(99)00121-7
Lovibond, S. H., & Lovibond, P. F. (1995a). Manual for the Depression Anxiety Stress
Scales. (2nd Ed.) Sydney Psychology Foundation.
Lovibond, P. F., & Lovibond, S. H. (1995b). The structure of negative emotional states:
Comparison of the Depression Anxiety Stress Scales (DASS) with the Beck
Depression and Anxiety Inventories. Behaviour Research and Therapy, 33(3), 335-
343. https://doi.org/10.1016/0005-7967(94)00075-U
Lovibond, P. F., Mitchell, C. J., Minard, E., Brady, A., & Menzies, R. G. (2009). Safety
behaviours preserve threat beliefs: Protection from extinction of human fear 276
conditioning by an avoidance response. Behaviour Research and Therapy, 47(8), 716-
720. https://doi.org/10.1016/j.brat.2009.04.013
Lovibond, P. F., & Shanks, D. R. (2002). The role of awareness in Pavlovian conditioning:
Empirical evidence and theoretical implications. Journal of Experimental Psychology:
Animal Behavior Processes, 28(1), 3-26. http://dx.doi.org/10.1037/0097-7403.28.1.3
Lovibond, P.F., Siddle, D.A.T., & Bond, N. (1988). Insensitivity to stimulus validity in
human Pavlovian conditioning. Quarterly Journal of Experimental Psychology, 40B,
377–410. https://doi.org/10.1080/14640748808402331
Lubow, R. E. (1973). Latent inhibition. Psychological Bulletin, 79(6), 398-407.
http://dx.doi.org/10.1037/h0034425
Lucas, K., Luck, C. C., & Lipp, O. V. (2018). Novelty-facilitated extinction and the
reinstatement of conditional human fear. Behaviour Research and Therapy, 109, 68-
74. https://doi.org/10.1016/j.brat.2018.08.002
Lykken, D. T., & Venables, P. H. (1971). Direct measurement of skin conductance: A
proposal for standardization. Psychophyiology, 8(5), 656-672.
https://doi.org/10.1111/j.1469-8986.1971.tb00501.x
Lyon, D. O. (1963). Frequency of reinforcement as a parameter of conditioned suppression.
Journal of the Experimental Analysis of Behavior, 6, 95-98.
https://doi.org/10.1901/jeab.1963.6-95
Mackintosh, N. J. (1974). The psychology of animal learning. London: Academic Press.
Mackintosh, N. J., & Honig, W. K. (1970). Blocking and enhancement of stimulus control in
pigeons. Journal of Comparative and Physiological Psychology, 73, 78-85.
http://dx.doi.org/10.1037/h0030021 277
Mackintosh, N. J., Kaye, H., & Bennett, C. H. (1991). Perceptual learning in flavour aversion
conditioning. Quarterly Journal of Experimental Psychology, 43B, 297-322.
https://doi.org/10.1080/14640749108401272
MacLean, E. L. (2016). Unraveling the evolution of uniquely human cognition. PNAS,
113(23), 6348-6354. https://doi.org/10.1073/pnas.1521270113
MacLeod, C. (1990). Mood disorders and cognition. In M. W. Eysenck (Ed.), Cognitive
psychology: An international review. Chichester: Wiley.
MacLeod, C., & Cohen, I. L. (1993). Anxiety and the interpretation of ambiguity: A text
comprehension study. Journal of Abnormal Psychology, 102(2), 238-247.
http://dx.doi.org/10.1037/0021-843X.102.2.238
MacLeod, C., & Donnellan, A. M. (1993). Individual differences in anxiety and the
restriction of working memory capacity. Personality and Individual Differences,
15(2), 163-173.http://dx.doi.org/10.1016/0191-8869(93)90023-V
MacLeod, C., & Mathews, A. (1988). Anxiety and the allocation of attention to threat. The
Quarterly Journal of Experimental Psychology A: Human Experimental Psychology,
40(4-A), 653-670. http://dx.doi.org/10.1080/14640748808402292
MacLeod, C., Mathews, A., & Tata, P. (1986). Attentional bias in emotional disorders.
Journal of Abnormal Psychology, 95, 15-20.
http://dx.doi.org/10.1037/0021-843X.95.1.15
Maes, E., Boddez, Y., Alfei, J. M., Krypotos, A.-M., D'Hooge, R., De Houwer, J., & Beckers,
T. (2016). The elusive nature of the blocking effect: 15 failures to replicate. Journal
of Experimental Psychology: General, 145(9), 49-71.
http://dx.doi.org/10.1037/xge0000200
Magoun, H. W. (1963). The waking brain. Springfield, Illinois 278
Maier, N. R. F. (1940). Studies of abnormal behavior in the rat. IV. Abortive behavior and its
relation to the neurotic attack. Journal of Experimental Psychology, 27(4), 369-393.
http://dx.doi.org/10.1037/h0053935
Malt, B. C., Ross, B. H., & Murphy, G. L. (1995). Predicting features for members of natural
categories when categorization is uncertain. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 21(3), 646-661.
http://dx.doi.org/10.1037/0278-7393.21.3.646
Maltzman, I., Langdon, B., & Feeney, D. (1970). Semantic generalization without prior
conditioning. Journal of Experimental Psychology, 83, 73.
http://dx.doi.org/10.1037/h0028536
Maltzman, I., Langdon, B., Pendery, M., & Wolff, C. (1977). Galvanic skin response-
orienting reflex and semantic conditioning and generalization with different
unconditioned stimuli. Journal of Experimental Psychology: General, 106(2), 141-
171. http://dx.doi.org/10.1037/0096-3445.106.2.141
Maren, S., & Fanselow, M. S. (1996). The amygdala and fear conditioning: Has the nut been
cracked? Neuron, 16, 237-240. https://doi.org/10.1016/S0896-6273(00)80041-0
Martin-Soelch, C., Stocklin, M., Dammann, G., Opwis, K., & Seifritz, E. (2006). Anxiety
trait modulates psychophysiological reactions, but not habituation processes related to
affective auditory stimuli. International Journal of Psychophysiology, 61, 87-97.
http://dx.doi.org/10.1016/j.ijpsycho.2005.07.009
Mathews, A., & MacLeod, C. (1985). Selective processing of threat cues in anxiety states.
Behaviour Research and Therapy, 23, 563–569.
http://dx.doi.org/10.1016/0005-7967(85)90104-4 279
Mathews, A., & MacLeod, C. (1986). Discrimination of threat cues without awareness in
anxiety states. Journal of Abnormal Psychology, 95(2), 131-138.
http://dx.doi.org/10.1037/0021-843X.95.2.131
Mattia, J. I., Heimberg, R. G., & Hope, D. A. (1993). The revised Stroop color-naming task
in social phobics. Behaviour Research and Therapy, 31(3), 305-313
.http://dx.doi.org/10.1016/0005-7967(93)90029-T
Mauss, I. B., Levenson, R. W., McCarter, L., Wilhelm, F. H., & Gross, J. J. (2005). The tie
that binds? Coherence among emotional experience, behaviour, and autonomic
physiology. Emotion, 5(2), 175-190. http://dx.doi.org/10.1037/1528-3542.5.2.175
McAllister, W. R., & McAllister, D. E. (1962). Role of the CS and of apparatus cues in the
measurement of acquired fear. Psychological Reports, 11, 749-756.
http://dx.doi.org/10.2466/pr0.1962.11.3.749
McAllister, W. R., & McAllister, D. E. (1971). Behavioral measurement of conditioned fear.
In F. R. Brush (Eds.), Aversive conditioning and learning (pp. 105-182).London:
Academic Press.
McEvoy, P. M., Grove, R., & Slade, T. (2011). Epidemiology of anxiety disorders in the
Australian general population: Findings of the 2007 Australian National Survey of
mental health and wellbeing. Australian and New Zealand Journal of Psychiatry, 45,
957-967. https://doi.org/10.3109/00048674.2011.624083
McLaren, I. P. L., Forrest, C. L. D., McLaren, R. P., Jones, F. W., Aitken, M. R. F., &
Mackintosh, N. J. (2014). Associations and propositions: The case for a dual-process
account of learning in humans. Neurobiology of Learning and Memory, 108, 185-195.
http://dx.doi.org/10.1016/j.nlm.2013.09.014 280
McLaren, I. P. L., Green, R. E. A. & Mackintosh, N. J. (1994) Animal learning and the
implicit/explicit distinction. In: Implicit and explicit learning of languages, (Ed.). N.
C. Ellis, pp. 313–32. Academic Press.
McLaren, I. P., & Mackintosh, N. J. (2002). Associative learning and elemental
representation: II. Generalization and discrimination. Animal Learning and Behavior,
30(3), 177-200. https://doi.org/10.3758/BF03192828
McLaughlin, K. A., Costello, E. J., LeBlanc, W., Sampson, N. A., & Kessler, R. C. (2012).
Socioeconomic status and adolescent mental disorders. American Journal of Public,
102(9), 1742-1750. http://dx.doi.org/10.2105/AJPH.2011.300477
McLean, C. P., Asnaani, A., Litz, B. T., & Hofmann, S. G. (2011). Gender differences in
anxiety disorders: Prevalence, course of illness, comorbidity and burden of illness.
Journal of Psychiatric Research, 45(8), 1027-1035.
http://dx.doi.org/10.1016/j.jpsychires.2011.03.006
McNally, R. J., Riemann, B. C., Louro, C. E., Lukach, B. M., & Kim, E. (1992). Cognitive
processing of emotional information in panic disorder. Behaviour Research and
Therapy, 30(2), 143-149. https://doi.org/10.1016/0005-7967(92)90137-6
McNally, R. J., Riemann, B. C., & Kim, E. (1990). Selective processing of threat cues in
panic disorder. Behaviour Research and Therapy, 28, 407-412.
http://dx.doi.org/10.1016/0005-7967(90)90160-K
McNew, J. J., & Thompson, R. (1966). Role of the limbic system in active and passive
avoidance conditioning in the rat. Journal of Comparative and Physiological
Psychology, 61(2), 173-180. http://dx.doi.org/10.1037/h0023127 281
McReynolds, P., Acker, M., & Brackbill, G. (1966). On the assessment of anxiety: IV. By
measures of basal conductance and palmar sweat. Psychological Reports, 19(2), 347-
356. http://dx.doi.org/10.2466/pr0.1966.19.2.347
Mecke, E. (1983). Absence of changes in colour discrimination ability of goldfish when
reared in monochromatic light. Annales Zoologici Fennici, 20(3), 239-244.
Mendlowicz, M. V., & Stein, M. B. (2000). Quality of life in individuals with anxiety
disorders. American Journal of Psychiatry, 157, 669-682.
http://dx.doi.org/10.1176/appi.ajp.157.5.669
Menzies, R. G., & Clark, J. C. (1995). Danger experiences and insight in acrophobia.
Behaviour Research and Therapy, 33, 215-221.
https://doi.org/10.1016/0005-7967(94)P4443-X
Merz, C. J., & Wolf, O. T. (2016). Sex differences in stress effects on emotional learning.
Journal of Neuroscience Research, 95(1), 93-105. https://doi.org/10.1002/jnr.23811
Meulders, A., Vandael, K., & Vlaeyen, J. W. S. (2017). Generalization of pain-related fear
based on conceptual knowledge. Behavior Therapy, 48(3), 295-310.
https://doi.org/10.1016/j.beth.2016.11.014
Michael, T., Blechert, J., Vriends, N., Margraf, J., & Wilhelm, F. H. (2007). Fear
conditioning in panic disorder: Enhanced resistance to extinction. Journal of
Abnormal Psychology, 116(3), 612–617.
http://dx.doi.org/10.1037/0021-843X.116.3.612
Miller, R. R., & Matzel, L. D. (1988). Contingency and relative associative strength. In S. B.
Klein., & R. R. Mowrer. (Eds). Contemporary learning theories: Pavlovian
conditioning and the status of traditional learning theory (pp 61-84). Hillsdale, NJ:
Lawrence Erlbaum Associates. 282
Mineka, S. (1985). Animal models of anxiety-based disorders: Their usefulness and
limitations. In J. Maser & A. Tuma (Eds.), Anxiety and the anxiety disorders (pp.
199–244). Hillsdale, NJ: Erlbaum.
Mineka, S., & Sutton, J. (2006). Contemporary Learning Theory Perspectives on the Etiology
of Fears and Phobias. In M. G. Craske, D. Hermans, & D. Vansteenwegen
(Eds.), Fear and learning: From basic processes to clinical implications (pp. 75-97).
Washington, DC, US: American Psychological Association.
http://dx.doi.org/10.1037/11474-004
Mineka, S., & Zinbarg, R. (2006). A contemporary learning theory perspective on the
etiology of anxiety disorders. It’s not what you thought it was. American
Psychologist, 61(1), 10-26. https://doi.org/10.1037/0003-066X.61.1.10
Mitchell, C. J., De Houwer, J., & Lovibond, P. F. (2009). The propositional nature of human
associative learning. Behavioral and Brain Sciences, 32, 183-246.
http://dx.doi.org/10.1017/S0140525X09000855
Mitchell, C. J., & Lovibond, P. F. (2002). Backward and forward blocking in human
electrodermal conditioning: Blocking requires an assumption of outcome additivity.
The Quarterly Journal of Experimental Psychology, 55B(4), 311-329.
https://doi.org/10.1080/02724990244000025
Mitchell-Heggs, N., Kelly, D., & Richardson, A. (1976). Stereotactic limbic leucotomy – A
follow-up at 16 months. British Journal of Psychiartry, 128, 226-240.
Mitte, K. (2007). Anxiety and risky decision-making: The role of subjective probability and
subjective costs of negative events. Personality and Individual Differences, 42(2),
243-253. https://doi.org/10.1016/j.paid.2006.11.028 283
Moeller, G. (1954). The CS-UCS interval in GSR conditioning. Journal of experimental
Psychology, 48, 162-166. http://dx.doi.org/10.1037/h0060475
Mogg, K., Holmes, A., Garner, M., & Bradley, B. P. (2008). Effects of threat cues on
attentional shifting, disengagement and response slowing in anxious individuals.
Behavioural Research and Therapy, 46(5), 656-667.
http://dx.doi.org/10.1016/j.brat.2008.02.011
Molesworth, C. J., Bowler, D. M., & Hampton, J. A. (2005). The prototype effect in
recognition memory: Intact in autism? Journal of Child Psychology and Psychiatry,
46(6), 661-672. https://doi.org/10.1111/j.1469-7610.2004.00383.x
Moore, J. W. (1972). Stimulus control: Studies of auditory generalisation in rabbits. In A. H.
Black & W. F. Prokasy (Eds.), Classical conditioning II: Current research and theory
(pp. 206-230). New York: Appleton-Century-Crofts.
Morgan, M. A., Romanski, L. M., & LeDoux, J. E. (1993). Extinction of emotional learning:
contribution of medial prefrontal cortex. Neuroscience Letters 163, 109–113.
https://doi.org/10.1016/0304-3940(93)90241-C
Morina, N., Ijntema, H., Meyerbröker, K., & Emmelkamp, P. M. G. (2015). Can virtual
reality exposure therapy gains be generalized to real-life? A meta-analzsis of studies
applying behavioral assessments. Behaviour Research and Therapy, 74, 18-24.
https://doi.org/10.1016/j.brat.2015.08.010
Morriss, J., Christakou, A., & van Reekum, C. M. (2016). Nothing is safe: Intolerance of
uncertainty is associated with compromised fear extinction learning. Biological
Psychology, 121(B), 187-193. https://doi.org/10.1016/j.biopsycho.2016.05.001 284
Morse, W. H., & Skinner, B. F. (1958). Some factors involved in the stimulus control of
operant behavior. Journal of Experimental Analysis of Behavior, 1(1), 103-107.
https://doi.org/10.1901/jeab.1958.1-103
Moruzzi, G., & Magoun, H. W. (1949). Brain stem reticular formation and activation of the
EEG. Electroencephalography and Clinical Neurophysiology, 1(4), 455-473.
https://doi.org/10.1016/0013-4694(49)90219-9
Mowrer, O. H. (1939). A stimulus-response analysis of anxiety and its role as a reinforcing
agent. Psychological Review, 46(6), 553-565. http://dx.doi.org/10.1037/h0054288
Mowrer, O. H. (1947). On the dual nature of learning – a reinterpretation of “conditioning”
and “problem solving”. Harvard Educational Review, 17, 102-148.
Mowrer, O. H. (1960). Learning theory and behavior. New York: Wiley.
Mowrer, O. H., & Solomon, R. L. (1954). Contiguity vs drive-reduction in conditioned fear:
The proximity and abruptness of drive-reduction. American Journal of Psychology,
67, 15-25. http://dx.doi.org/10.2307/1418068
Mineka, S., & Öhman, A. (2002). Phobias and preparedness: The selective, automatic, and
encapsulated nature of fear. Biological Psychiatry, 52(10), 927-937.
https://doi.org/10.1016/S0006-3223(02)01669-4
Mitchell, C. J., & Lovibond, P. F. (2002). Backward and forward blocking in human
electrodermal conditioning: Blocking requires an assumption of outcome additivity.
Quarterly Journal of experimental Psychology, 55(B), 311-329.
https://doi.org/10.1080/02724990244000025
Mitchell, C. J., De Houwer, J., & Lovibond, P. F. (2009). The propositional nature of human
associative learning. Behavioral and Brain Sciences, 32, 183-246.
http://dx.doi.org/10.1017/S0140525X09000855 285
Mitchell, C. J., Lovibond, P. F., & Condoleon, M. (2005). Evidence for deductive reasoning
in blocking of causal judgments. Learning and Motivation, 36(1), 77-87.
https://doi.org/10.1016/j.lmot.2004.09.001
Murphy, G. L. (2002). The big book of concepts. Cambridge, MA: MIT Press.
Murphy, G. L., & Ross, B. H. (1999). Induction with cross-classified categories. Memory &
Cognition, 27(6), 1024-1041. http://dx.doi.org/10.3758/BF03201232
Murphy, G. L., & Ross, B. H. (2010). Category vs. Object Knowledge in Category-based
Induction. Journal of Memory and Language, 63(1), 1–17.
http://doi.org/10.1016/j.jml.2009.12.002
Mwinyi, J., Pisanu, C., Castelao, E., Stringhini, S., Preisig, M., & Schioth, H. B. (2017).
Anxiety disorders are associated with low socioeconomic status in women but not in
men. Women’s Health Issues, 27(3), 302-307.
https://doi.org/10.1016/j.whi.2017.01.001
Myers, A. K. (1964). Discriminated operant avoidance learning in Wistar and G-4 rats as a
function of type of warning stimulus. Journal of Comparative and Physiological
Psychology, 58(3), 453-455. http://dx.doi.org/10.1037/h0040504
Myers, L. B. (2009). The importance of the repressive coping style: Findings from 30 years
of research. Anxiety, Stress & Coping, 23(1), 3-17
https://doi.org/10.1080/10615800903366945
Mystkowski, J. L., Craske, M. G., & Echiverri, A. M. (2002). Treatment context and return of
fear in spider phobia. Behavior Therapy, 33(3), 399-416.
https://doi.org/10.1016/S0005-7894(02)80035-1 286
Neary, R.S., & Zuckerman, M. (1976). Sensation seeking, trait and state anxiety, and the
electrodermal orienting response. Psychophysiology, 13, 205-211.
http://dx.doi.org/10.1111/j.1469-8986.1976.tb00098.x
Nelson, K., & Nelson, A. P. (1990). Category production in response to script and category
cues by kindergarten and second-grade children. Journal of Applied Developmental
Psychology, 11(4), 431-446. http://dx.doi.org/10.1016/0193-3973(90)90019-G
Nesse, R. M., & Klaas, R. (1994). Risk perception by patients with anxiety disorders. The
Journal of Nervous and Mental Disease, 182(8), 465-470.
Neumann, D. L. (2006). The effects of physical context changes and multiple extinction
contexts on two forms of renewal in a conditioned suppression task with humans.
Learning and Motivation, 37(2), 149-175. https://doi.org/10.1016/j.lmot.2005.06.004
Neumann, D. L., Lipp, O. V., & Cory, S. E. (2007). Conducting extinction in multiple
contexts does not necessarily attenuate the renewal of shock expectancy in a fear-
conditioning procedure with humans. Behaviour Research and Therapy, 45(2), 385-
394. https://doi.org/10.1016/j.brat.2006.02.001
Newell, B. R., & Dunn, J. C. (2008). Dimensions in data: Testing psychological models using
state-trace analysis. Trends in Cognitive Sciences, 12, 285-290.
https://doi.org/10.1016/j.tics.2008.04.009
Ng, A., & Lovibond, P. F. (2017). Intentions matter: Avoidance intentions regulate anxiety
via outcome expectancy. Behaviour Research and Therapy, 96, 57-65.
https://doi.org/10.1016/j.brat.2017.03.007
Nicholson, J. N., & Gray, J. A. (1971). Behavioural contrast and peak shift in children.
British Journal of Psychology, 62(3), 367-373.
https://doi.org/10.1111/j.2044-8295.1971.tb02047.x 287
Nielsen, T. C., & Petersen, K. (1976). Electrodermal correlates of extraversion, trait anxiety
and schitophrenisrn. Journal of Psychology, 17(1), 73-80.
https://doi.org/10.1111/j.1467-9450.1976.tb00214.x
Norrholm, S. D., Jovanovic, T., Vervliet, B., Myers, K. M., Davis, M., Rothbaum, B. O.,
Duncan, E. J. (2006). Conditioned fear extinction and reinstatement in a human fear-
potentiated startle paradigm. Learning and Memory, 13(6), 681-685.
http://dx.doi.org/10.1101/lm.393906
Norrholm, S. D., Vervliet, B., Jovanovic, T., Boshoven, W., Myers, K. M., Davis, M.,
Rothbaum, B., & Duncan, E. J. (2008). Timing of extinction relative to acquisition: A
parametric analysis of fear extinction in humans. Behavioral Neuroscience, 122(5),
1016-1030. http://dx.doi.org/10.1037/a0012604
Norrholm, S. D., Jovanovic, T., Olin, I. W., Sands, L. A., Karapanou, I., Bradley, B., et al.
(2011). Fear extinction in traumatized civilians with posttraumatic stress disorder:
Relation to symptom severity. Biological Psychiatry, 69(6), 556–563.
https://doi.org/10.1016/j.biopsych.2010.09.013
Norton, P. J., & Price, E. C. (2007). A meta-analytic review of adult cognitive-behavioral
treatment outcome across the anxiety disorders. The Journal of Nervous and Mental
Disease, 195(6), 521-531. http://dx.doi.org/10.1097/01.nmd.0000253843.70149.9a
Notterman, J. M., Schoenfeld, W. N., & Bersh, P. J. (1952). Conditioned heary rate response
in human beings during experimental anxiety. Journal of Comparative and
Physiological Psychology, 45, 1-8. http://dx.doi.org/10.1037/h0060870
Obsessive Compulsive Cognitions Working Group (2005). Psychometric validation of the
Obsessive Beliefs Questionnaire and the Interpretation of Instructions Inventory: Part 288
2: Factor analyses and testing of a brief version. Behaviour Research and Therapy,
43, 1527-1542. https://doi.org/10.1016/j.brat.2004.07.010
O'Gorman, J. G. (1977). Individual differences in habituation of human physiological
responses: A review of theory, method, and findings in the study of personality
correlates in non-clinical populations. Biological Psychology, 5(4), 257-318.
http://dx.doi.org/10.1016/0301-0511(77)90017-5
Öhman, A., Ellstrom, P. E., & Bjorkstrand, P. A. (1976). Electrodermal resposnes and
subjective estimates of UCS probability in a long interstimulus interval conditioning
paradigm. Psychophysiology, 13, 121-127.
http://dx.doi.org/10.1111/j.1469-8986.1976.tb00086.x
Öhman, A., Flykt, A., & Esteves, F. (2001). Emotion drives attention: detecting the snake in
the grass. Journal of Experimental Psychology: General, 130, 466–478.
http://dx.doi.org/10.1037/0096-3445.130.3.466
Öhman, A., & Mineka, S. (2001). Fears, phobias, and preparedness: toward an evolved
module of fear and fear learning. Psychological Review, 108, 483–522.
http://dx.doi.org/10.1037/0033-295X.108.3.483
Öhman, A., & Soares, J. J. (1993). On the automatic nature of phobic fear: Conditioned
electrodermal responses to masked fear-relevant stimuli. Journal of Abnormal
Psychology, 102(1), 121-132. http://dx.doi.org/10.1037/0021-843X.102.1.121
Olatunji, B. O., Cisler, J. M., & Tolin, D. F. (2007). Quality of life in the anxiety disorders: A
meta-analytic review. Clinical Psychology Review, 27, 572-581.
https://doi.org/10.1016/j.cpr.2007.01.015 289
Olsson, A., Nearing, K. I., & Phelps, E. A. (2007). Learning fears by observing others: The
neural systems of social fear transmission. Scan, 2, 3-11.
http://dx.doi.org/10.1093/scan/nsm005
Olsson, A., & Phelps, E. A. (2004). Learned fear of “Unseen” faces after Pavlovian,
observational and instructed fear. Psychological Science, 15(12), 822-828.
https://doi.org/10.1111/j.0956-7976.2004.00762.x
Orr, S. P., Metzger, L. J., Lasko, N. B., Macklin, M. L., Peri, T., & Pitman, R. K. (2000). De
novo conditioning in trauma-exposed individuals with and without posttraumatic stress
disorder. Journal of Abnormal Psychology, 109(2), 290-298.
http://dx.doi.org/10.1037/0021-843X.109.2.290
Osherson, D.N., Smith, E. E., Wilkie, O., Lopez, A., & Shafir, E. (1990). Category-based
induction. Psychological Review, 97,185
http://dx.doi.org/10.1037//0033-295X.97.2.185
Öst, L, G., Ferebee, I., & Furmark, T. (1997a). One-session group therapy of spider phobia:
Direct versus indirect treatments. Behaviour Research and Therapy, 35(8), 721-732.
https://doi.org/10.1016/S0005-7967(97)00028-4
Öst, L. G., Brandberg, M., & Alm, T. (1997b). One versus five sessions of exposure in the
treatment of flying phobia. Behaviour Research and Therapy, 35(11), 987-996.
https://doi.org/10.1016/S0005-7967(97)00077-6
Otto, M. W., Leyro, T. M., Christian, K., Deveney, C. M., Reese, H., Pollack, M. H., & Orr,
S. P. (2007). Prediction of “fear” acquisition in healthy control participants in a de
novo fear-conditioning paradigm.Behaviour Modification, 31(1), 32-51.
http://dx.doi.org/10.1177/0145445506295054 290
Pallesen, S., Nordhus, I. H., Carlstedt, B., Thayer, J. F., & Johnsen, T. B. (2006). A
Norwegian adaptation of the Penn State Worry Questionnaire: Factor structure,
reliability, validity and norms. Scandinavian Journal of Psychology, 47(4), 281-291.
http://dx.doi.org/10.1111/j.1467-9450.2006.00518.x
Papini, M. R., & Bitterman, M. E. (1990). The role of contingency in classical
conditioning. Psychological Review, 97(3), 396-403.
http://dx.doi.org/10.1037/0033-295X.97.3.396
Parkinson, B. (1994). Emotion. In A. M. Colman (Ed.), Companion encyclopaedia of
psychology (Vol. 2). London: Routledge.
Parkinson, B. (1995). Ideas and realities of emotion. London: Routledge.
Parsons, R. G., & Ressler, K. J. (2013). Implications of memory modulation for post-
traumatic stress and fear disorders. Nature Neuroscience, 16(2), 146–153.
http://doi.org/10.1038/nn.3296
Paterson, R. J., & Neufeld, R. W. (1987). Clear danger: Situational determinants of the
appraisal of threat. Psychological Bulletin, 101(3), 404-416.
http://dx.doi.org/10.1037/0033-2909.101.3.404
Pavlov, I. P. (1927). Conditioned reflexes: An investigation of the physiological activity of
the cerebral cortex. Oxford: Oxford University Press.
Pavlov, I. P. (1932). The reply of a physiologist to psychologists. Psychological Review,
39(2), 91-127. http://dx.doi.org/10.1037/h0069929
Pearce, J. M. (1987). A model for stimulus generalization in Pavlovian conditioning.
Psychological Review, 94, 61-73. http://dx.doi.org/10.1037/0033-295X.94.1.61 291
Pearce, J. M. (1994). Similarity and discrimination: A selective review and a connectionist
model. Psychological Review, 101, 587-607.
http://dx.doi.org/10.1037/0033-295X.101.4.587
Pearce, J. M. (2002). Evaluation and development of a connectionist theory of configural
learning. Animal Learning & Behavior, 30(2), 73-95.
https://doi.org/10.3758/BF03192911
Peastrel, A. L. (1964). Studies in efficiency: Semantic generalization in schizophrenia. The
Journal of Abnormal and Social Psychology, 69(4), 444-449.
http://dx.doi.org/10.1037/h0040656
Pellon, R., Montano, J.M.G., & Sanchez, P. (1995). Blocking and electrodermal conditioning
in humans. Psicologica, 16, 321–329.
Peri, T., Ben-Shakhar, G., Orr, S. P., & Shalev, A. Y. (2000). Psychophysiologic assessment
of aversive conditioning in posttraumatic stress disorder. Biological Psychiatry, 47(6),
512–519. https://doi.org/10.1016/S0006-3223(99)00144-4
Perruchet, P. (1985). A pitfall for the expectancy theory of human eyelid conditioning.
Pavlovian Journal of Biological Sciences, 20, 163-170.
https://doi.org/10.1007/BF03003653
Peterson, N. (1962). Effects of monochromatic rearing on the control of responding by
wavelength. Science, 136, 774-775. http://dx.doi.org/ 10.1126/science.136.3518.774
Phelps, E. A., O’Connor, K. J., Gatenby, J. C., Gore, J. C., Grillon, C., & Davis, M. (2001).
Activation of the left amygdala to a cognitive representation of fear. Nature
Neuroscience, 4, 437–441. http://dx.doi.org/10.1038/86110 292
Phelps, E. A., Delgado, M. R., Nearing, K. I., & LeDoux, J. E. (2004). Extinction learning in
humans: Role of the amygdala and vmPFC. Neuron, 43(6), 897-905.
https://doi.org/10.1016/j.neuron.2004.08.042
Pierrel, R., & Sherman, J. G. (1960). Generalization of auditory intensity following
discrimination training. Journal of the Experimental Analysis of Behavior, 3, 313-322.
http://dx.doi.org/10.1901/jeab.1960.3-313
Pitman, R. K., & Orr, S. P. (1986). Test of the conditioning model of neurosis: Differential
aversive conditioning of angry and neutral facial expressions in anxiety disorder
patients. Journal of Abnormal Psychology, 95(3), 208–213.
http://dx.doi.org/10.1037/0021-843X.95.3.208
Pittig, A., Treanor, M., LeBeau, R. T., & Craske, M. G. (2018). The role of associative fear
and avoidance learning in anxiety disorders: Gaps and directions for future research.
Neuroscience & Behavioral Reviews, 88, 117-140.
https://doi.org/10.1016/j.neubiorev.2018.03.015
Pittig, A., van den Berg, L., & Vervliet, B. (2015). The key role of extinction learning in
anxiety disorders: Behavioral strategies to enhance exposure-based treatments.
Current Opinion in Psychiatry, 29(1), 39-47.
http://dx.doi.org/10.1097/YCO.0000000000000220
Plaisted, K., O’Riordan, M., & Baron-Cohen, S. (1998). Enhanced visual search for a
conjunctive target in Autism: A research note. Journal of Child Psychology and
Psychiatry, 39(5), 777-783. https://doi.org/10.1111/1469-7610.00376
Podlesnik, C. A., & Miranda-Dukoski, L. (2015). Stimulus generalization and operant context
renewal. Behavioural Processes, 119, 93-98.
https://doi.org/10.1016/j.beproc.2015.07.015 293
Pole, N., Neylan, T. C., Otte, C., Henn-Hasse, C., Metzler, T. J., & Marmar, C. R. (2009).
Prospective prediction of PTSD symptoms using fear potentiated auditory startle
responses. Biological Psychiatry, 65(3), 235-240.
http://dx.doi.org/10.1016/j.biopsych.2008.07.015
Powell, G. E. (1979). Brain and Personality. Saxon House, London.
Premack, D. (1995). Cause/induced motion; intention/spontaneous motion. In J. P. Changeux,
& J. Chavaillon, J. (Eds.), Origins of the human brain, (pp. 286-308). Oxford: Oxford
University Press.
Pushkarskaya, H., Liu, X., Smithson, M., & Joseph, J. E. (2010). Beyond risk and ambiguity:
Deciding under ignorance. Cognitive, Affective & Behavioral Neuroscience, 10(3),
382-391. http://dx.doi.org/10.3758/CABN.10.3.382
Quirk, G. J., Garcia, R., & Gonzalez-Lima, F. (2006). Prefrontal mechanisms in extinction of
conditioned fear. Biological Psychiatry 60, 337–343.
http://dx.doi.org/10.1016/j.biopsych.2006.03.010
Quirk, G. J., & Mueller, D. (2008). Neural mechanisms of extinction learning and retrieval.
Neuropsychopharmacology Review, 33, 56-72.
http://dx.doi.org/10.1038/sj.npp.1301555
Rachman, S. (1990). Fear and courage (2nd ed.). New York: Freeman
Rapee, R. M. (1995). Descriptive psychopathology of social phobia. In R. G. Heimberg, M.
R. Liebowitz, D. A. Hope, & F. R. Schneier, (Eds.), Social Phobia: Diagnosis,
assessment, and treatment (pp. 41-66). New York, NY: Guilford Press.
Rapee, R. M., & Lim, L. (1992). Discrepancy between self- and observer ratings of
performance in social phobics. Journal of Abnormal Psychology, 101(4), 728-731.
http://dx.doi.org/10.1037/0021-843X.101.4.728 294
Rauch, S. L., Whalen, P. J., Shin, L. M., McLnerney, S. C., Macklin, M. L., Lasko, N. B.,
Orr, S. P., & Kitman, R. K. (2000). Exaggerated amygdala response to masked facial
stimuli in posttraumatic stress disorder: A functional MRI study. Biological
Psychiatry, 47(9), 769-776. https://doi.org/10.1016/S0006-3223(00)00828-3
Razran, G. H. S. (1939). A quantitative study of meaning by a conditioned salivary technique
(Semantic conditioning). Science, 90, 89–90.
http://dx.doi.org/10.1126/science.90.2326.89-a
Razran, G. H. S. (1949). Stimulus generalisation of conditioned responses. Psychological
Bulletin, 46, 337–365. http://dx.doi.org/10.1037/h0060507
Razran, G. H. S. (1955). Conditioning and perception. Psychological Review, 62(2), 83-95.
http://dx.doi.org/10.1037/h0046875
Reberg, D. (1972). Compound tests for excitation in early acquisition and after prolonged
extinction of conditioned suppression. Learning and Motivation, 3(3), 246-258.
https://doi.org/10.1016/0023-9690(72)90021-5
Reiss, S. (1980). Pavlovian conditioning and human fear: An expectancy model. Behavior
Therapy, 11(3), 380-396. https://doi.org/10.1016/S0005-7894(80)80054-2
Rescorla, R. A. (1968). Probability of shock in the presence and absence of CS in fear
conditioning. Journal of Comparative and Phsuological Psycho;pgy. 66(1), 1-5.
http://dx.doi.org/10.1037/h0025984
Rescorla, R. A. (1969a). Conditioned inhibition of fear resulting from negative CS-US
contingencies. Journal of Comparative and Physiological Psychology, 67(4), 504-
509. http://dx.doi.org/10.1037/h0027313
Rescorla, R. A. (1969b). Pavlovian conditioned inhibition. Psychological Bulletin, 72(2), 77-
94. http://dx.doi.org/10.1037/h0027760 295
Rescorla, R. (1974). Effect of inflation of the unconditioned stimulus value following
conditioning. Journal of Comparative and Physiological Psychology, 86, 101–106.
http://dx.doi.org/10.1037/h0035964
Rescorla, R. A. (1992). Inhibitory associations between S and R in extinction. Animal
Learning & Behavior, 21(4), 327-336. https://doi.org/10.3758/BF03197998
Rescorla, R. A. (2003). Protection from extinction. Animal Learning & Behavior, 31(2), 124-
132. https://doi.org/10.3758/BF03195975
Rescorla, R. A., & Heth, C. D. (1975). Reinstatement of fear to an extinguished conditioned
stimulus. Journal of Experimental Psychology: Animal Behavior Processes, 1(1), 88-
96. http://dx.doi.org/10.1037/0097-7403.1.1.88
Rescorla, R. A., & Skucy, J. C. (1969). Effect of response-independent reinforcers during
extinction. Journal of Comparative and Physiological Psychology, 67(3), 381-389.
http://dx.doi.org/10.1037/h0026793
Rescorla, R. A, & Wagner, AR. (1972). A theory of Pavlovian conditioning: variations in the
effectiveness of reinforcement and non reinforcement. In AH. Black & W.F. Prokasy
(eds.), Classical conditioning II: current research and theory (pp. 64-99) New York:
Appleton-Century-Crofts.
Richards, A., French, C. C., Johnson, W., Naparstek, J., & Williams, J. (1992). Effects of
mood manipulation and anxiety on performance of an emotional Stroop task. British
Journal of Psychology, 83(4), 479-491.
https://doi.org/10.1111/j.2044-8295.1992.tb02454.x
Riess, B. F. (1940). Semantic conditioning involving the galvanic skin reflex. Journal of
Experimental Psychology, 26(2), 238-240. http://dx.doi.org/10.1037/h0063000 296
Rips, L. J. (1975). Inductive judgments about natural categories. Journal of Verbal Learning
and Verbal Behavior, 14, 665–681. https://doi.org/10.1016/S0022-5371(75)80055-7
Rips, L. J. (1999). Deductive reasoning. In R. A. Wilson & F. C. Keil (Eds.), The MIT
Encyclopedia of the cognitive sciences (pp. 225–226). Cambridge, MA: MIT Press.
Rilling, M. (1977). Stimulus control and inhibitory processes. In W. K. Honig, & J. E. R.
Staddon (Eds.), Handbook of opernt behavior (pp. 432-480). Englewood Cliffs, NJ:
Prentice-Hall.
Rodriguez, B. I., Craske, M. G., Mineka, S., & Hladek, D. (1999). Context-specificity of
relapse: Effects of therapist and environmental context on return of fear. Behaviour
Research and Therapy, 37(9), 845-862.
https://doi.org/10.1016/S0005-7967(98)00106-5
Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D. M., & Boyes-Braem, P. (1976). Basic
objects in natural categories. Cognitive Psychology, 8(3), 382-439.
http://dx.doi.org/10.1016/0010-0285(76)90013-X
Ross, L. E., & Nelson, M. N. (1973). The role of awareness in differential conditioning.
Psychophysiology, 10, 91-94. http://dx.doi.org/10.1111/j.1469-8986.1973.tb01089.x
Rozin, P., Dow, S., Moscovitch, M., & Rajaram, S. (1998). What causes humans to begin and
end a meal? A role for memory for what has been eaten, as evidenced by a study of
multiple meal eating in amnesic patients. Psychological Science, 9(5), 392-396.
http://dx.doi.org/10.1111/1467-9280.00073
Santibáñez-H, G., & Schroeder, H. (1988). Does heart rate differentiate neurotic from normal
people in a conditional reflex test? The Pavlovian Journal of Biological Science,
23(1), 15-21. https://doi.org/10.1007/BF02910540 297
Scheveneels, S., Boddez, Y., Vervliet, B., & Hermans, D. (2016). The validity of laboratory-
based treatment research: Bridging the gap between fear extinction and exposure
treatment. Behaviour Research and Therapy, 86, 87-94.
https://doi.org/10.1016/j.brat.2016.08.015
Schmaltz, L. W., & Theios, J. (1972). Acquisition and extinction of a classically conditioned
response in hippocampectomized rabbits (Oryctolagus cuniculus). Journal of
Comparative and Physiological Psychology, 79(2), 328-333.
http://dx.doi.org/10.1037/h0032531
Schneider, F., Weiss, U., Kessler, C., Muller-Gartner, H. W., Posse, S., Salloum, J. B.,
Grodd, W., Hommelmann, F., Gaebel, W., & Birbraumer, N. (1999), Subcortical
correlates of differential conditioning of aversive emotional reactions in social phobia.
Biological Psychiatry, 45(7), 863-871.
https://doi.org/10.1016/S0006-3223(98)00269-8
Schwartz, G. E. (1995). Psychobiology of repression and health: A systems approach. In J. L.
Singer (Ed.), The John D. and Catherine T. MacArthur Foundation series on mental
health and development. Repression and dissociation: Implications for personality
theory, psychopathology, and health (pp. 405-434). Chicago, IL, US: University of
Chicago Press.
Schweckendiek, J., Klucken, T., Merz, C. J., Tabbert, K, Walter, B., Ambach, W., Vatil, D.,
& Stark, R. (2011). Weaving the (neuronal) web: Fear learning in spider phobia.
NeuroImage, 54(1), 681-688. https://doi.org/10.1016/j.neuroimage.2010.07.049
Sehlmeyer, C., Dannlowski, U., Schöning, S., Kugel, H., Pyka, M., Pfleiderer, B.,
Zwitserlood, P., Schiffbauer, H., Heindel, W., Arolt, V., & Konrad, C. (2011). Neural 298
correlates of trait anxiety in fear extinction. Psychological Medicine, 41(4), 789-798.
https://doi.org/10.1017/S0033291710001248
Seligman, M .E. (1968). Chronic fear produced by unpredictable electric shock, Journal of
Comparative and Physiological Psychology, 66(2), 402-411.
http://dx.doi.org/10.1037/h0026355
Shanks, D. R. (2007) Associationism and cognition: Human contingency learning at 25.
Quarterly Journal of Experimental Psychology 60:291–309.
http://dx.doi.org/10.1080/17470210601000581
Shanks, D. R., & Darby, R. J. (1998). Feature- and rule-based generalization in human
associative learning. Journal of Experimental Psychology: Animal Behavior
Processes, 24(4), 405-415. http://dx.doi.org/10.1037/0097-7403.24.4.405
Shekhar, A., McCann, U. D., Meaney, M. J., Blanchard, D. C., Davis, M., Frey, K. A.,…, &
Winsky, L. (2001). Summary of a National Institute of Mental Health workshop:
Developing animal models of anxiety disorders. Psychopharmacology, 157(4), 327-
339. https://doi.org/10.1007/s002130100859
Sherman, J. E. (1978). US inflation with trace and simultaneous fear conditioning. Animal
Learning & Behavior, 6(4), 463-468. https://doi.org/10.3758/BF03209645
Shiban, Y., Schelhorn, I., Pauli, P., & Mühlberger, A. (2015). Effect of combined multiple
contexts and multiple stimuli exposure in spider phobia: A randomized clinical trial in
virtual reality. Behaviour Research and Therapy, 71, 45-53.
http://dx.doi.org/10.1016/j.brat.2015.05.014
Shore, J. H., Vollmer, W. M., & Tatum, E. L. (1989). Community patterns of posttraumatic
stress disorders. Journal of Nervous and Mental Disease, 177(11), 681-685.
http://dx.doi.org/10.1097/00005053-198911000-00004 299
Shou, Y., & Smithson, M. (2015). Adapting to an Uncertain World: Cognitive Capacity and
Causal Reasoning with Ambiguous Observations. PLoS ONE, 10(10), e0140608.
http://doi.org/10.1371/journal.pone.0140608
Shurtleff, D., & Aryes, J. J. B. (1980). One-trial backward excitatory fear conditioning in
rats: Acquisition, retention, extinction, and spontaneous recovery. Animal Learning &
Behavior, 9(1), 65-74. https://doi.org/10.3758/BF03212027
Sidman, M. (1971). Reading and auditory-visual equivalences. Journal of Speech & Hearing
Research, 14(1), 5-13. http://dx.doi.org/10.1044/jshr.1401.05
Sidman, M. (1994). Equivalence relations and behavior: A research story. Boston: Authors
Cooperative.
Siegel, S., Hearst, E., George, N., & O’Neal, E. (1968). Generalization gradients obtained
from individual subjects following classical conditioning. Journal of Experimental
Psychology, 78(1), 171-174. http://dx.doi.org/10.1037/h0026178
Skre, I., Onstad, S., Torgersen, S., Lygren, S., & Kringlen, E. (1993). A twin study of DSM-
III-R anxiety disorders. Acta Psychiatrica Scandinavica, 88(2), 85-92.
https://doi.org/10.1111/j.1600-0447.1993.tb03419.x
Sloutsky, V. M., & Fisher, A. V. (2004). Induction and Categorization in Young Children: A
Similarity-Based Model. Journal of Experimental Psychology: General, 133(2), 166-
188. http://dx.doi.org/10.1037/0096-3445.133.2.166
Smith, B. H. (1993). Merging mechanism and adaptation: An ethological approach to
learning and generalization. In D. R. Papaj, & A. C. Lewis, (Eds.), Insect Learning
(pp.126-157). New York: Spring US. 300
Smith, C. A., & Lazarus, R. D. (1993). Appraisal components, core relational themes, and the
emotions. Cognition and Emotion, 7, 233-269.
https://doi.org/10.1080/02699939308409189
Smith, E. E., & Medin, D. L. (1981). Categories and concepts. Cambridge, MA: Harvard
University Press.
Soeter, M., & Kindt, M. (2013). High trait anxiety: A challenge for disrupting fear memory
reconsolidation. PLoS ONE, 8(11): e75239.
https://doi.org/10.1371/journal.pone.0075239
Sokol, N., & Lovibond, P. F. (2012). Cross-US reinstatement of human conditioned fear:
Return of old fears or emergence of new ones? Behaviour Research and Therapy,
50(5), 313-322. https://doi.org/10.1016/j.brat.2012.02.005
Sokolov, E. N. (1960). Neuronal models and the orienting influence. In M. A. Brazier, (Eds.),
The central nervous system and behavior: III. New York, Macy Foundation.
Solomon, R. L., & Wynne, L. C. (1953). Traumatic avoidance learning: Acquisition in
normal dogs. Psychological monographs, 67, 1-19.
http://dx.doi.org/10.1037/h0093649
Sotres-Bayon, F., Cain, C. K., LeDoux, J. E. (2006). Brain mechanisms of fear extinction:
Historical perspectives on the contribution of prefrontal cortex. Biological Psychiatry
60, 329–336. https://doi.org/10.1016/j.biopsych.2005.10.012
Spence, K. W. (1937). The differential response in animals to stimuli varying in a single
dimension. Psychological Review, 44, 430-444. http://dx.doi.org/10.1037/h0062885
Spence, K. W. (1945). An experimental test of the continuity and non-continuity theories of
discrimination learning. Journal of Experimental Psychology, 35(4), 253-266.
http://dx.doi.org/10.1037/h0060794 301
Spence, K, W., & Runquist, W. N. (1958). Temporal effects of conditioned fear on the eyelid
reflex. Journal of Experimental Psychology, 55, 613-616.
http://dx.doi.org/10.1037/h0049214
Spence, K. W., & Taylor, J. A. (1951). Anxiety and strength of UCS as determiners of the
amount of eyelid conditioning. Journal of Experimental Psychology, 42, 183-188.
http://dx.doi.org/10.1037/h0061580
Spielberger, C. D. (1966). Theory and research on anxiety. In C. D. Spielberger (Ed.),
Anxiety and Behaviour (pp.3-20). New York: Academic Press.
Spielberger, C. D. (1972a). Anxiety as an Emotional State. In C. D. Spielberger (Ed.),
Anxiety: Current Trends in Theory and Research (Vol. 1, pp. 23-49). New York:
Academic Press. http://dx.doi.org/10.1016/B978-0-12-657401-2.50009-5
Spielberger, C. D. (1972b). Conceptual and methodological issues in anxiety research. In C.
D. Spielberger (Ed.), Anxiety: Current trends in Theory and Research (Vol. 2. New
York: Academic Press.
Spielberger, C. D. (1975a). The measurement of state and trait anxiety: Conceptual and
methodological issues. In L. Levi (Ed.), Emotions-their parameters and measurement.
New York: Raven Press.
Spielberger, C. D. (1975b). Anxiety: State-trait process. In C. D. Spielberger & I. G. Sarason
(Eds.), Stress and Anxiety (Vol.1, pp. 115-141). Washington, DC:
Hermisphere/Wiley.
Spielberger, C. D. (1985).Anxiety, cognition and affect: A state-trait perspective. . In A. H.
Tuma & J. D. Maser (Eds.), Anxiety and the anxiety disorders (pp.171-182).
Hillsdale, NJ: Erlbaum. 302
Spielberger, C. D., Gorsuch, R. L., & Lushene, R. E. (1970). Manual for the state-trait
anxiety inventory. Palo Alto, CA: Consulting Psychologist Press.
Spielberger, C. D., Gorsuch, R. L., Lushene, R., Vagg, P. R., & Jacobs, G. A. (1983). Manual
for the State-Trait Anxiety Inventory. Palo Alto, CA: Consulting Psychologists Press.
Staats, A. W., Staats, C. K., & Crawford, H. L. (1962). First-order conditioning of meaning
and the parallel conditioning of a GSR. Journal of General Psychology, 67, 159-167.
http://dx.doi.org/10.1080/00221309.1962.9711540
Stöber, J. (1997). Trait anxiety and pessimistic appraisal of risk and chance. Personality and
Individual Differences, 22(4), 465-476.
https://doi.org/10.1016/S0191-8869(96)00232-2
Stopa, L., & Clark, D. M. (1993). Cognitive processes in social phobia. Behaviour Research
and Therapy, 31(3), 255-267. http://dx.doi.org/10.1016/0005-7967(93)90024-O
Storch, E. A., Roberti, J. W., & Roth, D. A. R. (2004). Factor structure, concurrent validity,
and internal consistency of the Beck Depression Inventory – Second edition in a
sample of college students. Depression and Anxiety, 19, 187-189.
http://dx.doi.org/10.1002/da.20002
Stroop, J. R. (1935). Studies of interference in serial verbal reactions. Journal of
Experimental Psychology, 18, 643–662. http://dx.doi.org/10.1037/h0054651
Tellegen, A. (1985). Structures of mood and personality and their relevance to assessing
anxiety, with an emphasis on self-report. In A. H. Tuma & J. D. Maser (Eds.), Anxiety
and the anxiety disorders (pp.681-706). Hillsdale, NJ: Erlbaum.
Terrace, H. S. (1964). Wavelength generalization after discrimination training with and
without errors. Science, 144, 78-80. http://dx.doi.org/10.1126/science.144.3614.78 303
Thayer, J. F., Friedman, B. H., Borkovec, T. D., Johnsen, B. H., & Molina, S. (2000). Phasic
heart period reactions to cued threat and nonthreat stimuli in generalized anxiety
disorder. Psychophysiology, 37(3), 361-368.
https://doi.org/10.1111/1469-8986.3730361
The MathWorks Inc (2014). MATLAB – the language of technical computing. Version
R2014b (9.5). Natick, Massachusetts.
Thomas, D. R. (1993). A model for adaptation-level effects on stimulus
generalization. Psychological Review, 100(4), 658-673.
http://dx.doi.org/10.1037/0033-295X.100.4.658
Thomas, D. R., & Decapito, A. (1966). Role of stimulus labeling in stimulus
generalization. Journal of Experimental Psychology, 71(6), 913-915.
http://dx.doi.org/10.1037/h0023212
Thomas, D. R., Lusky, M., & Morrison, S. (1992). A comparison of generalization functions
and frame of reference effects in different training paradigms. Perception &
Psychophysics, 51(6), 529-540. https://doi.org/10.3758/BF03211650
Thomas, D. R., & Mitchell, K. (1962). Instructions and stimulus categorizing in a measure of
stimulus generalization. Journal of the Experimental Analysis of Behavior, 5(3), 375-
381. http://dx.doi.org/10.1901/jeab.1962.5-375
Thomas, D. R., Mood, K., Morrison, S., & Wiertelak, E. (1991). Peak shift revisited: A test of
alternative interpretations. Journal of Experimental Psychology: Animal Behavior
Processes, 17(2), 130-140. http://dx.doi.org/10.1037/0097-7403.17.2.130
Thomas, D. R., & Switalski, R. W. (1966). Comparison of stimulus generalization following
variable-ratio and variable-interval training. Journal of Experimental Psychology,
71(2), 236-240. http://dx.doi.org/10.1037/h0022880 304
Thorndike, E. L. (1898). Animal intelligence: An experimental study of the associative
processes in animals. The Psychological Review: Monograph Supplements, 2(4), i-
109. http://dx.doi.org/10.1037/h0092987
Tolman, E. C. (1932). Purposive behaviour in animals and men. New York: Century.
Torgersen, S. (1983). Genetic factors in anxiety disorders. Archives of General Psychiatry,
40(10), 1085-1089. http://dx.doi.org/10.1001/archpsyc.1983.01790090047007
Torrents-Rodas, D., Fullana, M. A., Bonillo, A., Caseras, X., Andion, O., & Torrubia, R.
(2013). No effect of trait anxiety on differential fear conditioning or fear
generalization. Biological Psychology, 92(2), 185-190.
https://doi.org/10.1016/j.biopsycho.2012.10.006
Treanor, M., Stapes-Bradley, L. K., & Craske, M. G. (2015). Learning theory and cognitive-
behavior therapy. In D. S. Pine, K. J. Ressler, & B. O. Rothbaum (Eds.). Primer on
anxiety disorders: Translational perspectives on diagnosis and treatment (pp. 355–
366). Oxford University Press.
Vansteenwegen, D., Hermans, D., Vervliet, B., Francken, G., Beckers, T., Baeyens, F., &
Eelen, P. (2005). Return of fear in a human differential conditioning paradigm caused
by a return to the original acquisition context. Behaviour Research and Therapy,
43(3), 323-336. https://doi.org/10.1016/j.brat.2004.01.001
Vansteenwegen, D., Vervliet, B., Iberico, C., Baeyens, F., Van den Bergh, O., & Hermans, D.
(2007). The repeated confrontation with videotapes of spiders in multiple contexts
attenuates renewal of fear in spider-anxious students. Behaviour Research and
Therapy, 45, 1169-1179. http://dx.doi.org/10.1016/j.brat.2006.08.023 305
Vervliet, B., Baker, A., & Craske, M. G. (2012). Fear conditioning in animals and humans. In
N. M. Seel, & D. Quinones (Eds.). Encyclopedia of the sciences of learning (pp.
1273–1276). New York: Springer Science + Business Media.
Vervliet, B., & Geens, M. (2014). Fear generalization in humans: Impact of feature learning
on conditioning and extinction. Neurobiology of Learning and Memory, 113, 143-148.
https://doi.org/10.1016/j.nlm.2013.10.002
Vervliet, B., & Indekeu, E. (2015). Low-cost avoidance behaviors are resistant to fear
extinction in humans. Frontiers in Behavioral Neuroscience, 24.
https://doi.org/10.3389/fnbeh.2015.00351
Vervliet, B., Kindt, M., Vansteenwegen, D., & Hermans, D. (2010). Fear generalization in
humans: Impact of verbal instructions. Behaviour Research and Therapy, 48(1), 38-
43. https://doi.org/10.1016/j.brat.2009.09.005
Vervliet, B., & Raes, F. (2013). Criteria of validity in experimental psychopathology:
Application to models of anxiety and depression. Psychological Medicine, 43, 2241-
2244. https://doi.org/ doi:10.1017/S0033291712002267
Vervliet, B., Vansteenwegen, D., Baeyens, F., Hermans, D., & Eelen, P. (2005). Return of
fear in a human differential conditioning paradigm caused by a stimulus change after
extinction. Behaviour Research and Therapy, 43, 357-371.
http://dx.doi.org/10.1016/j.brat.2004.02.005
Vervliet, B., Vansteenwegen, D., & Eelen, P. (2004). Generalization of extinguished skin
conductance responding in human fear conditioning. Learning & Memory, 11, 555-
558. http://dx.doi.org/10.1101/lm.77404 306
Vervoort, E., Vervliet, B., Bennett, M., & Baeyens, F. (2014). Generalization of Human Fear
Acquisition and Extinction within a Novel Arbitrary Stimulus Category. PLoS ONE
9(5): e96569. https://doi.org/10.1371/journal.pone.0096569
Voorspoels, W., Navarro, D. J., Perfors, A., Ransom, K., & Storms, G. (2015). How do
people learn from negative evidence? Non-monotonic generalizations and sampling
assumptions in inductive reasoning. Cognitive Psychology, 81, 1–25.
https://doi.org/10.1016/j.cogpsych.2015.07.001
Wagner, A. R. (1961). Effects of amount and percentage of reinforcement and number of
acquisition trials on conditioning and extinction. Journal of Experimental Psychology,
62(3), 234-242. http://dx.doi.org/10.1037/h0042251
Warner, L. H. (1932). The association span of the white rat. The Pedagogical Seminary and
Journal of Genetic Psychology, 41, 57-90.
http://dx.doi.org/10.1080/08856559.1932.9944143
Waters, A. M., Kershaw, R., & Lipp, O. V. (2018). Multiple fear-related stimuli enhance
physiological arousal during extinction and reduce physiological arousal to novel
stimuli and the threat conditioned stimulus. Behaviour Research and Therapy, 106,
28-36. https://doi.org/10.1016/j.brat.2018.04.005
Watson, D., & Clark, L, A. (1984). Negative affectivity: The disposition to experience
aversive emotional states. Psychological Bulletin, 96(3), 465-490.
http://dx.doi.org/10.1037/0033-2909.96.3.465
Watson, D., Clark, L. A., & Carey, G. (1988). Positive and negative affectivity and their
relation to anxiety and depressive disorders. Journal of Abnormal Psychology, 97(3),
346-353. http://dx.doi.org/10.1037/0021-843X.97.3.346 307
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief
measures of positive and negative affect: The PANAS scales. Journal of Personality
and Social Psychology, 54(6), 1063-1070.
http://dx.doi.org/10.1037/0022-3514.54.6.1063
Watson, D., & Tellegen, A. (1985). Toward a consensual structure of mood. Psychological
Bulletin, 98(2), 219-235. http://dx.doi.org/10.1037/0033-2909.98.2.219
Watson, D., Weber, K., Assenheimer, J. S., Clark, L. A., Strauss, M. E., & McCormick, R. A.
(1995a). Testing a tripartite model: I. Evaluating the convergent and discriminant
validity of anxiety and depression symptom scales. Journal of Abnormal Psychology,
104(1), 3-14. http://dx.doi.org/10.1037/0021-843X.104.1.3
Watson, D., Clark, L. A., Weber, K., Assenheimer, J. S., Strauss, M. E., & McCormick, R. A.
(1995b). Testing a tripartite model: II. Exploring the symptom structure of anxiety and
depression in student, adult, and patient samples. Journal of Abnormal Psychology,
104(1), 15-25. http://dx.doi.org/10.1037/0021-843X.104.1.15
Watson, J. B., & Morgan, J. J. B. (1917). Emotional reactions and psychological
experimentation. The American Journal of Psychology, 28(2), 163-174.
http://dx.doi.org/10.2307/1413718
Watson, J. B., & Rayner, R. (1920). Conditioned emotional reactions. Journal of
Experimental Psychology, 3(1), 1-14. http://dx.doi.org/10.1037/h0069608
Wegerer, M., Blechert, J., Kerschbaum, H., & Wihelm, F. H. (2013). Relationship between
fear conditionability and aversive memories: Evidence from a novel conditioned-
intrusion paradigm. PLoSONE, 8(11), e79025.
https://doi.org/10.1371/journal.pone.0079025 308
Weidemann, G., & Antees, C. (2012). Parallel acquisition of awareness and differential delay
eyeblink conditioning. Learning & Memory, 19, 201-210.
http://dx.doi.org/10.1101/lm.024851.111
Weidemann, G., & Lovibond, P. F. (2016). The role of US recency in the Perruchet effect in
eyeblink conditioning. Biological Psychology, 119, 1-10.
https://doi.org/10.1016/j.biopsycho.2016.06.007
Weidemann, G., Satkunarajah, M., & Lovibond, P. F. (2016). I think, Therefore eyeblink:
The importance of contingency awareness in conditioning. Psychological Science,
27(4), 467-475. https://doi.org/10.1177/0956797615625973
Weinberger, D. A., Schwartz, G. E., & Davidson, R. J. (1979). Low-anxious, high-anxious,
and repressive coping styles: Psychometric patterns and behavioral and physiological
responses to stress. Journal of Abnormal Psychology, 88(4), 369-380.
http://dx.doi.org/10.1037/0021-843X.88.4.369
Weiss, J. M., Krieckhaus, E. E., & Conte, R. (1968). Effects of fear conditioning on
subsequent avoidance behaviour and movement, Journal of Comparative and
Physiological Psychology, 65(3), 413-421. http://dx.doi.org/10.1037/h0025832
Westbrook, R. F., Iordanova, M., McNally, G., Richardson, R., & Harris, J. A. (2002).
Reinstatement of fear to an extinguished conditioned stimulus: Two roles for
context. Journal of Experimental Psychology: Animal Behavior Processes, 28(1), 97-
110. http://dx.doi.org/10.1037/0097-7403.28.1.97
White, C. T., & Schlosberg, H. (1952). Degree of conditioning of the GSR as a function of
the time of delay. Journal of Experimental Psychology, 43, 357-362.
http://dx.doi.org/10.1037/h0061520 309
Whitty, C. W. M. (1955). Effects of anterior cingulectomy in man. Proceedings of the Royal
Society of Medicine, 48(6), 463-469. https://doi.org/10.1177/003591575504800619
Williams, P. (2002). What Australians eat for breakfast: An analysis of data from the 1995
National Nutrition Survey. Nutrition & Dietetics, 59(2), 103-112.
Willis, R. D., & Lundin, R. W. (1966). Conditioned suppression in the rat as a function of
shock reinforcement schedule. Psychonomic Science, 6, 107-108.
http://dx.doi.org/10.3758/BF03327980
Wickens, D. D., Schroder, H. M., & Snide, J. D. (1954). Primary stimulus generalization of
the GSR under two conditions. Journal of Experimental Psychology, 47(1), 52.
http://dx.doi.org/10.1037/h0053617
Williams, S. L., Kinney, P. J., & Falbo, J. (1989). Generalization of therapeutic changes in
agoraphobia: The role of perceived self-efficacy. Journal of Consulting and Clinical
Psychology, 57(3), 436-442. http://dx.doi.org/10.1037/0022-006X.57.3.436
Willner, J. A. (1978). Blocking of a taste aversion by prior pairings of exteroceptive stimuli
with illness. Learning and Motivation, 9, 125-140.
https://doi.org/10.1016/0023-9690(78)90016-4
Wills, S., & Mackintosh, N. J. (1998). Peak shift on an artificial dimension. Quarterly Journal
of Experimental Psychology Section B, 51(1b), 1-32.
https://doi.org/10.1080/713932669
Winograd, T. (1983). Language as a Cognitive Process. Volume I: Syntax. Massachuetts,
Addison-Wesley.
Wong, A. H. K., & Lovibond, P. F. (2017). Rule-based gneralisation in single-cue and
differential fear conditioning in humans. Biological Psychology, 129, 111-120.
https://doi.org/10.1016/j.biopsycho.2017.08.056 310
Yehuda, R., Kahana, B., Schmeidler, J., Southwick, S. M., Wilson, S., & Giller, E. L. (1995).
Impact of cumulative lifetime trauma and recent stress on current posttraumatic stress
disorder symptoms in Holocaust survivors. American Journal of Psychiatry, 152,
1815-1818. https://doi.org/10.1176/ajp.152.12.1815
Zbozinek, T. D., & Craske, M. G. (2018). Pavlovian extinction of fear with the original
conditional stimulus, a generalization stimulus, or multiple generalization stimuli.
Behaviour Research and Therapy, 107, 64-75.
https://doi.org/10.1016/j.brat.2018.05.009
Zeaman, D., Deane, G., & Wegner, N. (1954). Amplitude and latency characteristics of the
conditioned heart response. Journal of Psychology, 38, 235-250.
http://dx.doi.org/10.1080/00223980.1954.9712933
Zielinski, K., & Jakubowska, E. (1977). Auditory intensity generalization after CER
differentiation training. Acta Neurobiologiae Experimentalis, 37, 191-205.
Zigmond, A. S., & Snaith, R. P. (1983). The Hospital Anxiety and Depression Scale. Acta
Psychiatrica Scandinavica, 67(6), 361-370.
https://doi.org/10.1111/j.1600-0447.1983.tb09716.x
Zinbarg, R., & Revelle, W. (1989). Personality and conditioning: A test of four
models. Journal of Personality and Social Psychology, 57(2), 301-314.
http://dx.doi.org/10.1037/0022-3514.57.2.301
311
Appendices
Appendix A
Experiment 1
Information Statement and Consent forms……………………………………………… 312
»›Course credit compensation…………………………………………………… 312
»›Financial compensation……………………………………………………….. 314
Post-experimental questionnaire………………………………………………………… 316
Output from statistical analyses…………………………………………………………. 318
312
Approval No HC14003 THE UNIVERSITY OF NEW SOUTH WALES
PARPTICIPANT INFORMATION STATEMENT
The generalisation of emotional learning in humans
Participant Selection and Purpose of Study You are invited to participate in a study on the physiological responses produced by emotional stimuli. We hope to learn more about the cognitive processes involved in learning and anxiety. You were selected as a possible participant in this study because you are enrolled in Psychology 1.
Description of Study and Risks The emotional stimulus will be electric shock, presented through electrodes attached to one finger. The physiological measure recorded will be skin conductance, a sensitive measure of emotional state. If you decide to participate, we will ask you to sample the electric shock and select a level of shock which you personally judge to be definitely uncomfortable but not painful. During the experiment, you will be presented with a series of visual stimuli, and asked to make responses. You may receive up to 10 electric shocks over a testing period of approximately 30 minutes. There will be rest periods between the test trials, during which no shock is given. After the experiment we will also ask you to report your thoughts about the stimuli. If you decide to participate but change your mind during the experiment, please notify us straight away and we will terminate the experiment.
Confidentiality and Disclosure of Information Any information that is obtained in connection with this study and that can be identified with you will remain confidential and will be disclosed only with your permission or except as required by law. If you give us your permission by signing this document, we plan to publish the results in scientific journals. In any publication, information will be provided in such a way that you cannot be identified, for example by presenting only group data.
Recompense to participants You will receive 0.75 hour research credit for your participation.
Your consent Your decision whether or not to participate will not prejudice your future relations with The University of New South Wales. If you decide to participate, you are free to withdraw your consent and to discontinue participation at any time, without disadvantage. For example, if you become excessively anxious at any stage, we ask you to tell us so that we can terminate the experiment immediately. You will still receive your course credit. You are reminded that your involvement in the School of Psychology research participation scheme is voluntary, and that you may obtain the same course marks through an alternative form of assessment.
Inquiries If you have any questions or concerns following your participation, Alex Wong (0404118987; [email protected]) and Professor Peter Lovibond (9385 3830; [email protected]) will be happy to address them.
Complaints may be directed to the Ethics Secretariat, The University of New South Wales, SYDNEY 2052 AUSTRALIA (phone 9385 4234, fax 9385 6648, email [email protected]).
You may keep this information sheet and a copy of the Participant Consent Form. The investigator will keep the other signed copy. Both copies should be signed by you and the investigator.
313
THE UNIVERSITY OF NEW SOUTH WALES Approval No HC14003 PARTICIPANT CONSENT FORM
The generalisation of emotional learning in humans
You are making a decision whether or not to participate. Before that, we ask you to answer the following question concerning any potential heart problem you might have.
DO YOU HAVE A HEART PROBLEM? No Yes
Your signature below indicates that: a) you have read the Participant Information Statement, b) you have answered “No” to the above question about heart problem, and c) you have decided to participate.
…………………………………………………… ……………………………………………… Signature of Research Participant Please PRINT name
…………………………………………………… Date
…………………………………………………… …………………………………………………. Signature of Investigator Please PRINT Name
......
REVOCATION OF CONSENT
The generalisation of emotional learning in humans
I hereby wish to WITHDRAW my consent to participate in the research proposal described above and understand that such withdrawal WILL NOT jeopardise my relationship with the researchers, the School of Psychology or the University of New South Wales.
…………………………………………………… ………………………………………………… Signature of Research Participant Please PRINT name
…………………………………………………… Date
The section for Revocation of Consent should be forwarded to Alex Wong ([email protected]) 314
Approval No HC14003 THE UNIVERSITY OF NEW SOUTH WALES PARPTICIPANT INFORMATION STATEMENT The generalisation of emotional learning in humans
Participant Selection and Purpose of Study You are invited to participate in a study on the physiological responses produced by emotional stimuli. We hope to learn more about the cognitive processes involved in learning and anxiety. You were selected as a possible participant in this study because your responded to an advertisement for research on learning and emotion.
Description of Study and Risks The emotional stimulus will be electric shock, presented through electrodes attached to one finger. The physiological measure recorded will be skin conductance, a sensitive measure of emotional state. If you decide to participate, we will ask you to sample the electric shock and select a level of shock which you personally judge to be definitely uncomfortable but not painful. During the experiment, you will be presented with a series of visual stimuli, and asked to make responses. You may receive up to 10 electric shocks over a testing period of approximately 30 minutes. There will be rest periods between the test trials, during which no shock is given. After the experiment we will also ask you to report your thoughts about the stimuli. If you decide to participate but change your mind during the experiment, please notify us straight away and we will terminate the experiment.
Confidentiality and Disclosure of Information Any information that is obtained in connection with this study and that can be identified with you will remain confidential and will be disclosed only with your permission or except as required by law. If you give us your permission by signing this document, we plan to publish the results in scientific journals. In any publication, information will be provided in such a way that you cannot be identified, for example by presenting only group data.
Recompense to participants You will receive $15 to compensate for your time.
Your consent Your decision whether or not to participate will not prejudice your future relations with The University of New South Wales. If you decide to participate, you are free to withdraw your consent and to discontinue participation at any time, without disadvantage. For example, if you become excessively anxious at any stage, we ask you to tell us so that we can terminate the experiment immediately. You will still receive your course credit. You are reminded that your involvement in the School of Psychology research participation scheme is voluntary, and that you may obtain the same course marks through an alternative form of assessment.
Inquiries If you have any questions or concerns following your participantion, Alex Wong (0404118987; [email protected]) and Professor Peter Lovibond (9385 3830; [email protected]) will be happy to address them.
Complaints may be directed to the Ethics Secretariat, The University of New South Wales, SYDNEY 2052 AUSTRALIA (phone 9385 4234, fax 9385 6648, email [email protected]).
You may keep this information sheet and a copy of the Participant Consent Form. The investigator will keep the other signed copy. Both copies should be signed by you and the investigator. 315
THE UNIVERSITY OF NEW SOUTH WALES Approval No HC14003 PARTICIPANT CONSENT FORM
The generalisation of emotional learning in humans
You are making a decision whether or not to participate. Before that, we ask you to answer the following question concerning any potential heart problem you might have.
DO YOU HAVE A HEART PROBLEM? No Yes
Your signature below indicates that: a) you have read the Participant Information Statement, b) you have answered “No” to the above question about heart problem, and c) you have decided to participate.
…………………………………………………… ……………………………………………… Signature of Research Participant Please PRINT name
…………………………………………………… Date
…………………………………………………… …………………………………………………. Signature of Investigator Please PRINT Name
......
REVOCATION OF CONSENT
The generalisation of emotional learning in humans
I hereby wish to WITHDRAW my consent to participate in the research proposal described above and understand that such withdrawal WILL NOT jeopardise my relationship with the researchers, the School of Psychology or the University of New South Wales.
…………………………………………………… ………………………………………………… Signature of Research Participant Please PRINT name
…………………………………………………… Date
The section for Revocation of Consent should be forwarded to Alex Wong ([email protected])
316
Post-experimental questionnaire
1. In the experiment, your expectancy of shock when was presented was ______,
while your expectancy of shock when was presented was ______.
Please explain in detail below why you made those expectancy ratings. If you came up with
any strategies/rules in the experiment, please write these below as well.
______
______
______
______
______
______
______
______
______
______
______
______
______
______
______
______
______317
Please rate the extent to which you consider the following statements to be true:
a) The more the dot was to the right the more likely shock would be presented.
0% 50% 100% |____|____|____|____|____|____|____|____|____|____|
False True
b) The more the dot was to the left, the more likely shock would be presented.
0% 50% 100% |____|____|____|____|____|____|____|____|____|____|
False True
c) The closer the dot was to the middle, the more likely shock would be presented.
0% 50% 100% |____|____|____|____|____|____|____|____|____|____|
False True
d) The shock was equally likely regardless of the position of the dot. 100% 0% 50% |____|____|____|____|____|____|____|____|____|____|
False True
e) Other (please specify)
______
______
0% 50% 100% |____|____|____|____|____|____|____|____|____|____|
False True
318
Acquisition phase (Expectancy data)
B1 = Between groups W1 = Linear trend B1W1 = Interaction between groups and linear trend
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 73.702 1 73.702 0.027 Error 210416.881 78 2697.652 ------Within ------Linear W1 96290.660 1 96290.660 101.632 B1W1 626.856 1 626.856 0.662 Error 73900.750 78 947.446 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 -0.555 3.353 -7.229 6.120 Linear W1 23.048 2.286 18.497 27.600 B1W1 3.719 4.573 -5.384 12.822 ------
319
Acquisition phase (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 3.61030e-7 1 3.61030e-7 0.001 Error 0.049 78 0.001 ------Within ------Linear W1 0.011 1 0.011 17.014 B1W1 0.000 1 0.000 0.034 Error 0.052 78 0.001 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.000 0.003 -0.006 0.006 Linear W1 -0.013 0.003 -0.020 -0.007 B1W1 -0.001 0.006 -0.014 0.012 ------
320
Test phase (Overall data)
B1 = Between groups W1 = Linear trend W2 = Quadratic trend W3 = Comparison between CS+ to all GSs
Expectancy data
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 19173.472 1 19173.472 11.033 Error 135547.884 78 1737.793 ------Within ------Linear W1 18099.101 1 18099.101 23.779 B1W1 281.502 1 281.502 0.370 Error 59369.834 78 761.152 Quad W2 36091.255 1 36091.255 49.027 B1W2 6127.349 1 6127.349 8.323 Error 57420.055 78 736.155 CS+ vs GSs W3 64684.597 1 64684.597 119.940 B1W3 3186.470 1 3186.470 5.908 Error 42065.942 78 539.307 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 10.321 3.107 4.135 16.507 Linear W1 11.651 2.389 6.894 16.408 B1W1 2.906 4.779 -6.607 12.419 Quad W2 -15.976 2.282 -20.518 -11.433 B1W2 13.165 4.563 4.080 22.250 CS+ vs GSs W3 30.160 2.754 24.677 35.643 B1W3 -13.388 5.508 -24.353 -2.423 ------
321
Skin conductance data
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.000 1 0.000 0.396 Error 0.069 78 0.001 ------Within ------Linear W1 0.001 1 0.001 5.167 B1W1 0.000 1 0.000 0.533 Error 0.015 78 0.000 Quad W2 0.004 1 0.004 10.268 B1W2 0.001 1 0.001 2.581 Error 0.027 78 0.000 CS+ vs GSs W3 0.006 1 0.006 13.518 B1W3 0.001 1 0.001 1.429 Error 0.036 78 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 -0.002 0.003 -0.009 0.005 Linear W1 -0.004 0.002 -0.007 0.000 B1W1 0.003 0.003 -0.004 0.009 Quad W2 -0.007 0.002 -0.011 -0.003 B1W2 0.007 0.004 -0.002 0.015 CS+ vs GSs W3 0.010 0.003 0.005 0.016 B1W3 -0.007 0.006 -0.018 0.004 ------
322
Test Phase (Comparison between rule subgroups)
Linear vs Similiarity (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Rule B1 12749.034 1 12749.034 10.256 Error 45995.569 37 1243.123 ------Within ------Linear W1 38688.348 1 38688.348 68.896 B1W1 23232.530 1 23232.530 41.372 Error 20777.296 37 561.549 Quad W2 28835.151 1 28835.151 119.048 B1W2 15756.281 1 15756.281 65.051 Error 8961.919 37 242.214 CS+ vs GSs W3 36955.183 1 36955.183 134.346 B1W3 6514.857 1 6514.857 23.684 Error 10177.768 37 275.075 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Rule B1 13.058 4.078 4.796 21.320 Linear W1 26.430 3.184 19.978 32.882 B1W1 40.962 6.368 28.059 53.866 Quad W2 -22.156 2.031 -26.270 -18.041 B1W2 32.756 4.061 24.527 40.984 CS+ vs GSs W3 35.371 3.052 29.188 41.554 B1W3 -29.702 6.103 -42.069 -17.336 ------
323
Linear vs Similiarity (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Rule B1 0.001 1 0.001 0.452 Error 0.044 37 0.001 ------Within ------Linear W1 3.84461e-7 1 3.84461e-7 0.002 B1W1 0.001 1 0.001 3.876 Error 0.007 37 0.000 Quad W2 0.001 1 0.001 2.631 B1W2 0.003 1 0.003 8.268 Error 0.014 37 0.000 CS+ vs GSs W3 0.002 1 0.002 5.664 B1W3 0.007 1 0.007 17.403 Error 0.014 37 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Rule B1 -0.004 0.006 -0.016 0.008 Linear W1 0.000 0.003 -0.005 0.006 B1W1 0.011 0.005 0.000 0.022 Quad W2 -0.006 0.003 -0.012 0.001 B1W2 0.020 0.007 0.006 0.033 CS+ vs GSs W3 0.009 0.004 0.001 0.017 B1W3 -0.033 0.008 -0.048 -0.017 ------
324
Similarity vs No rule (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Rule B1 35327.473 1 35327.473 20.796 Error 112120.500 66 1698.795 ------Within ------Linear W1 3303.749 1 3303.749 15.640 B1W1 20.202 1 20.202 0.096 Error 13941.305 66 211.232 Quad W2 52317.077 1 52317.077 137.037 B1W2 34110.639 1 34110.639 89.348 Error 25196.960 66 381.772 CS+ vs GSs W3 70889.674 1 70889.674 172.863 B1W3 13418.582 1 13418.582 32.721 Error 27065.971 66 410.090 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Rule B1 -15.528 3.405 -22.327 -8.730 Linear W1 5.517 1.395 2.732 8.303 B1W1 0.863 2.790 -4.708 6.434 Quad W2 -21.319 1.821 -24.955 -17.683 B1W2 -34.429 3.642 -41.701 -27.157 CS+ vs GSs W3 34.996 2.662 29.682 40.310 B1W3 30.452 5.323 19.823 41.080 ------
325
Similarity vs No rule (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Rule B1 0.000 1 0.000 0.070 Error 0.063 66 0.001 ------Within ------Linear W1 0.002 1 0.002 8.723 B1W1 3.80216e-6 1 3.80216e-6 0.021 Error 0.012 66 0.000 Quad W2 0.006 1 0.006 18.534 B1W2 0.002 1 0.002 6.058 Error 0.022 66 0.000 CS+ vs GSs W3 0.012 1 0.012 28.802 B1W3 0.005 1 0.005 12.820 Error 0.026 66 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Rule B1 0.001 0.004 -0.007 0.009 Linear W1 -0.006 0.002 -0.009 -0.002 B1W1 0.001 0.004 -0.007 0.008 Quad W2 -0.010 0.002 -0.014 -0.005 B1W2 -0.011 0.005 -0.020 -0.002 CS+ vs GSs W3 0.015 0.003 0.010 0.021 B1W3 0.020 0.006 0.009 0.032 ------
326
Linear vs No rule (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Rule B1 509.709 1 509.709 0.333 Error 78164.267 51 1532.633 ------Within ------Linear W1 41830.516 1 41830.516 81.937 B1W1 27065.268 1 27065.268 53.015 Error 26036.531 51 510.520 Quad W2 1602.709 1 1602.709 4.186 B1W2 45.951 1 45.951 0.120 Error 19524.584 51 382.835 CS+ vs GSs W3 13394.518 1 13394.518 28.355 B1W3 4.634 1 4.634 0.010 Error 24091.372 51 472.380 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Rule B1 -2.470 4.283 -11.069 6.129 Linear W1 25.998 2.872 20.232 31.765 B1W1 41.825 5.744 30.293 53.357 Quad W2 -4.941 2.415 -9.790 -0.093 B1W2 -1.673 4.830 -11.370 8.024 CS+ vs GSs W3 20.145 3.783 12.550 27.740 B1W3 0.749 7.566 -14.440 15.939 ------
327
Linear vs No rule (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Rule B1 0.000 1 0.000 0.565 Error 0.030 51 0.001 ------Within ------Linear W1 6.67512e-7 1 6.67512e-7 0.004 B1W1 0.001 1 0.001 5.673 Error 0.008 51 0.000 Quad W2 8.44320e-8 1 8.44320e-8 0.000 B1W2 0.001 1 0.001 2.783 Error 0.012 51 0.000 CS+ vs GSs W3 0.000 1 0.000 0.077 B1W3 0.001 1 0.001 3.157 Error 0.017 51 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Rule B1 -0.003 0.004 -0.011 0.005 Linear W1 0.000 0.002 -0.005 0.005 B1W1 0.011 0.005 0.002 0.021 Quad W2 0.000 0.003 -0.005 0.005 B1W2 0.008 0.005 -0.002 0.019 CS+ vs GSs W3 -0.001 0.003 -0.008 0.006 B1W3 -0.012 0.007 -0.026 0.002 ------
328
Overall analysis of the interaction between trait anxiety and rules on fear generalisation
B1 = Between anxiety groups B7 = Interaction between anxiety and B4 B2 = Between Linear and No rule subgroups W1 = Linear trend B3 = Between Similarity and No rule subgroups W2 = Quadratic trend B4 = Between Similarity and Linear subgroups W3 = Comparison between CS+ & GSs B5 = Interaction between anxiety and B2 B6 = Interaction between anxiety and B3 Expectancy data
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 3721.270 1 3721.270 2.789 LvNR B2 304.277 1 304.277 0.228 SvNR B3 30813.912 1 30813.912 23.093 SvL B4 11099.806 1 11099.806 8.319 I_LvNR B5 3804.695 1 3804.695 2.851 I_SvNR B6 6108.338 1 6108.338 4.578 I_SvL B7 14.808 1 14.808 0.011 Error 98740.129 74 1334.326 ------Within ------Linear W1 33388.346 1 33388.346 82.945 B1W1 4.657 1 4.657 0.012 B2W1 23768.302 1 23768.302 59.046 B3W1 0.004 1 0.004 8.71132e-6 B4W1 21052.068 1 21052.068 52.298 B5W1 55.023 1 55.023 0.137 B6W1 367.523 1 367.523 0.913 B7W1 411.379 1 411.379 1.022 Error 29787.721 74 402.537 Quad W2 27010.844 1 27010.844 81.099 B1W2 1149.073 1 1149.073 3.450 B2W2 75.662 1 75.662 0.227 B3W2 30827.594 1 30827.594 92.559 B4W2 12908.431 1 12908.431 38.757 B5W2 71.132 1 71.132 0.214 B6W2 203.429 1 203.429 0.611 B7W2 3.822 1 3.822 0.011 Error 24646.390 74 333.059 CS+ vs GSs W3 43512.542 1 43512.542 118.292 B1W3 49.324 1 49.324 0.134 B2W3 48.095 1 48.095 0.131 B3W3 11839.097 1 11839.097 32.185 B4W3 6725.947 1 6725.947 18.285 B5W3 2078.123 1 2078.123 5.650 B6W3 587.307 1 587.307 1.597 B7W3 681.346 1 681.346 1.852 Error 27220.206 74 367.841 ------
329
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 5.388 3.226 -1.041 11.817 LvNR B2 2.000 4.187 -6.344 10.343 SvNR B3 14.829 3.086 8.680 20.978 SvL B4 12.829 4.448 3.966 21.692 I_LvNR B5 7.071 4.187 -1.273 15.415 I_SvNR B6 6.602 3.086 0.454 12.751 I_SvL B7 -0.469 4.448 -9.332 8.395 Linear W1 18.752 2.059 14.649 22.855 B1W1 -0.443 4.118 -8.648 7.762 B2W1 -41.069 5.345 -51.718 -30.420 B3W1 -0.012 3.939 -7.859 7.836 B4W1 41.057 5.677 29.745 52.370 B5W1 -1.976 5.345 -12.625 8.673 B6W1 3.763 3.939 -4.084 11.611 B7W1 5.739 5.677 -5.573 17.052 Quad W2 -16.377 1.819 -20.001 -12.754 B1W2 6.756 3.637 -0.491 14.003 B2W2 2.250 4.721 -7.156 11.656 B3W2 33.468 3.479 26.536 40.399 B4W2 31.218 5.015 21.226 41.210 B5W2 2.182 4.721 -7.224 11.588 B6W2 2.719 3.479 -4.213 9.650 B7W2 0.537 5.015 -9.454 10.529 CS+ vs GSs W3 29.313 2.695 23.943 34.683 B1W3 -1.974 5.390 -12.714 8.767 B2W3 2.530 6.996 -11.410 16.469 B3W3 -29.248 5.155 -39.520 -18.976 B4W3 -31.778 7.431 -46.585 -16.970 B5W3 -16.628 6.996 -30.568 -2.689 B6W3 -6.514 5.155 -16.787 3.758 B7W3 10.114 7.431 -4.693 24.922 ------
330
Skin conductance data
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.001 1 0.001 0.746 LvNR B2 0.000 1 0.000 0.117 SvNR B3 9.17191e-6 1 9.17191e-6 0.010 SvL B4 0.000 1 0.000 0.154 I_LvNR B5 0.001 1 0.001 0.741 I_SvNR B6 0.001 1 0.001 0.952 I_SvL B7 0.000 1 0.000 0.018 Error 0.067 74 0.001 ------Within ------Linear W1 0.000 1 0.000 0.362 B1W1 7.02567e-9 1 7.02567e-9 0.000 B2W1 0.001 1 0.001 5.730 B3W1 0.000 1 0.000 0.127 B4W1 0.001 1 0.001 4.026 B5W1 0.000 1 0.000 0.584 B6W1 0.000 1 0.000 0.447 B7W1 0.000 1 0.000 1.400 Error 0.014 74 0.000 Quad W2 0.002 1 0.002 5.058 B1W2 0.000 1 0.000 0.864 B2W2 0.000 1 0.000 1.501 B3W2 0.002 1 0.002 5.452 B4W2 0.002 1 0.002 7.691 B5W2 7.54215e-6 1 7.54215e-6 0.023 B6W2 9.22196e-6 1 9.22196e-6 0.029 B7W2 0.000 1 0.000 0.068 Error 0.024 74 0.000 CS+ vs GSs W3 0.002 1 0.002 6.188 B1W3 2.61466e-6 1 2.61466e-6 0.007 B2W3 0.001 1 0.001 2.831 B3W3 0.005 1 0.005 11.997 B4W3 0.006 1 0.006 15.895 B5W3 0.000 1 0.000 0.228 B6W3 0.000 1 0.000 0.044 B7W3 0.000 1 0.000 0.353 Error 0.028 74 0.000 ------
331
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 -0.003 0.004 -0.011 0.005 LvNR B2 0.002 0.005 -0.009 0.012 SvNR B3 0.000 0.004 -0.008 0.007 SvL B4 -0.002 0.006 -0.013 0.009 I_LvNR B5 0.004 0.005 -0.006 0.015 I_SvNR B6 0.004 0.004 -0.004 0.011 I_SvL B7 -0.001 0.006 -0.012 0.010 Linear W1 -0.001 0.002 -0.005 0.003 B1W1 0.000 0.004 -0.008 0.008 B2W1 -0.012 0.005 -0.023 -0.002 B3W1 -0.001 0.004 -0.009 0.006 B4W1 0.011 0.006 0.000 0.022 B5W1 0.004 0.005 -0.006 0.014 B6W1 -0.003 0.004 -0.010 0.005 B7W1 -0.007 0.006 -0.018 0.004 Quad W2 -0.005 0.002 -0.010 -0.001 B1W2 0.004 0.005 -0.005 0.014 B2W2 -0.008 0.006 -0.020 0.005 B3W2 0.011 0.005 0.002 0.020 B4W2 0.018 0.007 0.005 0.031 B5W2 -0.001 0.006 -0.013 0.011 B6W2 0.001 0.005 -0.008 0.010 B7W2 0.002 0.007 -0.011 0.015 CS+ vs GSs W3 0.007 0.003 0.001 0.013 B1W3 0.000 0.006 -0.012 0.011 B2W3 0.013 0.008 -0.002 0.029 B3W3 -0.020 0.006 -0.031 -0.008 B4W3 -0.033 0.008 -0.049 -0.016 B5W3 -0.004 0.008 -0.019 0.012 B6W3 0.001 0.006 -0.010 0.013 B7W3 0.005 0.008 -0.012 0.021 ------
332
Trait anxiety effect between Similarity subgroups (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 95.596 1 95.596 0.060 Error 39880.305 25 1595.212 ------Within ------Linear W1 1079.029 1 1079.029 6.990 B1W1 482.116 1 482.116 3.123 Error 3858.918 25 154.357 Quad W2 64080.637 1 64080.637 226.173 B1W2 233.999 1 233.999 0.826 Error 7083.149 25 283.326 CS+ vs GSs W3 55197.041 1 55197.041 213.310 B1W3 107.062 1 107.062 0.414 Error 6469.122 25 258.765 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 1.299 5.306 -9.629 12.226 Linear W1 5.070 1.918 1.121 9.020 B1W1 -6.778 3.835 -14.677 1.121 Quad W2 -37.939 2.523 -43.135 -32.744 B1W2 4.585 5.045 -5.806 14.977 CS+ vs GSs W3 49.655 3.400 42.653 56.657 B1W3 -4.374 6.800 -18.378 9.630 ------
333
Trait anxiety effect between Similarity subgroups (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.001 1 0.001 0.495 Error 0.038 25 0.002 ------Within ------Linear W1 0.000 1 0.000 1.915 B1W1 0.000 1 0.000 0.899 Error 0.005 25 0.000 Quad W2 0.006 1 0.006 11.711 B1W2 0.000 1 0.000 0.100 Error 0.012 25 0.000 CS+ vs GSs W3 0.012 1 0.012 24.848 B1W3 0.000 1 0.000 0.206 Error 0.012 25 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 -0.005 0.008 -0.021 0.010 Linear W1 -0.004 0.003 -0.011 0.002 B1W1 0.006 0.006 -0.007 0.019 Quad W2 -0.015 0.004 -0.024 -0.006 B1W2 0.003 0.009 -0.015 0.021 CS+ vs GSs W3 0.025 0.005 0.015 0.035 B1W3 -0.005 0.010 -0.025 0.016 ------
334
Trait anxiety effect between Linear subgroups (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 3.139 1 3.139 0.005 Error 6016.529 10 601.653 ------Within ------Linear W1 37826.725 1 37826.725 23.153 B1W1 98.205 1 98.205 0.060 Error 16338.056 10 1633.806 Quad W2 851.820 1 851.820 5.702 B1W2 150.985 1 150.985 1.011 Error 1493.787 10 149.379 CS+ vs GSs W3 3030.222 1 3030.222 10.081 B1W3 595.833 1 595.833 1.982 Error 3005.751 10 300.575 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.362 5.007 -10.794 11.518 Linear W1 46.128 9.587 24.767 67.488 B1W1 4.701 19.173 -38.020 47.421 Quad W2 -6.721 2.815 -12.993 -0.450 B1W2 5.660 5.629 -6.883 18.203 CS+ vs GSs W3 17.877 5.630 5.332 30.422 B1W3 15.855 11.261 -9.236 40.945 ------
335
Trait anxiety effect between Linear subgroups (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.001 1 0.001 1.021 Error 0.005 10 0.001 ------Within ------Linear W1 0.000 1 0.000 2.103 B1W1 0.000 1 0.000 0.582 Error 0.002 10 0.000 Quad W2 0.000 1 0.000 0.537 B1W2 0.000 1 0.000 0.492 Error 0.002 10 0.000 CS+ vs GSs W3 0.000 1 0.000 2.432 B1W3 0.000 1 0.000 0.269 Error 0.002 10 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 -0.007 0.007 -0.022 0.008 Linear W1 0.007 0.005 -0.004 0.017 B1W1 -0.007 0.009 -0.027 0.013 Quad W2 0.003 0.004 -0.007 0.013 B1W2 0.006 0.009 -0.013 0.026 CS+ vs GSs W3 -0.008 0.005 -0.019 0.003 B1W3 0.005 0.010 -0.017 0.028 ------
336
Trait anxiety effect between No rule subgroups (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 19301.303 1 19301.303 14.245 Error 52843.296 39 1354.956 ------Within ------Linear W1 1739.188 1 1739.188 7.072 B1W1 9.523 1 9.523 0.039 Error 9590.747 39 245.917 Quad W2 1441.258 1 1441.258 3.498 B1W2 1810.358 1 1810.358 4.394 Error 16069.454 39 412.037 CS+ vs GSs W3 15095.676 1 15095.676 33.177 B1W3 2744.454 1 2744.454 6.032 Error 17745.333 39 455.009 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 14.504 3.843 6.731 22.276 Linear W1 5.059 1.902 1.211 8.906 B1W1 0.749 3.804 -6.946 8.444 Quad W2 -4.471 2.391 -9.307 0.364 B1W2 10.023 4.782 0.351 19.694 CS+ vs GSs W3 20.407 3.543 13.241 27.573 B1W3 -17.402 7.086 -31.735 -3.070 ------
337
Trait anxiety effect between No rule subgroups (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.000 1 0.000 0.262 Error 0.025 39 0.001 ------Within ------Linear W1 0.001 1 0.001 6.557 B1W1 7.63054e-6 1 7.63054e-6 0.045 Error 0.007 39 0.000 Quad W2 0.001 1 0.001 3.058 B1W2 0.000 1 0.000 0.756 Error 0.010 39 0.000 CS+ vs GSs W3 0.001 1 0.001 2.236 B1W3 0.000 1 0.000 0.098 Error 0.014 39 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.002 0.004 -0.006 0.010 Linear W1 -0.006 0.002 -0.010 -0.001 B1W1 0.001 0.005 -0.008 0.010 Quad W2 -0.004 0.002 -0.009 0.001 B1W2 0.004 0.005 -0.006 0.014 CS+ vs GSs W3 0.005 0.003 -0.002 0.012 B1W3 -0.002 0.007 -0.016 0.012 ------
338
Appendix B
Validation Questionnaires, Experiment 2 and 3
Information Statement and Consent forms 339
»›Course credit compensation 339
»›Financial compensation 341
Stimuli used in Validation Questionnaires 343
Pictures used in the Categorical task in Experiment 2 & 3 346
Post-experimental questionnaire 347
Output from statistical analyses 349
»›Validation Questionnaires 349
»›Experiment 2 353
»›Experiment 3 359
339
Approval No HC14003
THE UNIVERSITY OF NEW SOUTH WALES PARTICIPANT INFORMATION STATEMENT The generalisation of emotional learning in humans
Participant Selection and Purpose of Study You are invited to participate in a study on the physiological responses produced by emotional stimuli. We hope to learn more about the cognitive processes involved in learning and anxiety. You were selected as a possible participant in this study because you are enrolled in Psychology 1.
Description of Study and Risks The emotional stimulus will be electric shock, presented through electrodes attached to one finger. The physiological measure recorded will be skin conductance, a sensitive measure of emotional state. If you decide to participate, we will ask you to sample the electric shock and select a level of shock which you personally judge to be definitely uncomfortable but not painful. The experiment consists of 2 parts – the first part being a categorization task, in which you have to categorize items into the most appropriate category. In the second part of the experiment, you will be presented with some visual stimuli, and asked to make responses. You may receive up to 10 electric shocks over a testing period of approximately 30 minutes. There will be rest periods between the test trials, during which no shock is given. After the experiment we will also ask you to report your thoughts about the stimuli. If you decide to participate but change your mind during the experiment, please notify us straight away and we will terminate the experiment.
Confidentiality and Disclosure of Information Any information that is obtained in connection with this study and that can be identified with you will remain confidential and will be disclosed only with your permission or except as required by law. If you give us your permission by signing this document, we plan to publish the results in scientific journals. In any publication, information will be provided in such a way that you cannot be identified, for example by presenting only group data.
Recompense to participants You will receive 0.75 hour research credit for your participation.
Your consent Your decision whether or not to participate will not prejudice your future relations with The University of New South Wales. If you decide to participate, you are free to withdraw your consent and to discontinue participation at any time, without disadvantage. For example, if you become excessively anxious at any stage, we ask you to tell us so that we can terminate the experiment immediately. You will still receive your course credit. You are reminded that your involvement in the School of Psychology research participation scheme is voluntary, and that you may obtain the same course marks through an alternative form of assessment.
Inquiries If you have any questions or concerns following your participantion, Alex Wong (0404118987; [email protected]) and Professor Peter Lovibond (9385 3830; [email protected]) will be happy to address them.
Complaints may be directed to the Ethics Secretariat, The University of New South Wales, SYDNEY 2052 AUSTRALIA (phone 9385 4234, fax 9385 6648, email [email protected]).
You may keep this information sheet and a copy of the Participant Consent Form. The investigator will keep the other signed copy. Both copies should be signed by you and the investigator. 340
THE UNIVERSITY OF NEW SOUTH WALES Approval No HC14003 PARTICIPANT CONSENT FORM
The generalisation of emotional learning in humans
You are making a decision whether or not to participate. Before that, we ask you to answer the following question concerning any potential heart problem you might have.
DO YOU HAVE A HEART PROBLEM? No Yes
Your signature below indicates that: a) you have read the Participant Information Statement, b) you have answered “No” to the above question about heart problem, and c) you have decided to participate.
…………………………………………………… ……………………………………………… Signature of Research Participant Please PRINT name
…………………………………………………… Date
…………………………………………………… …………………………………………………. Signature of Investigator Please PRINT Name
......
REVOCATION OF CONSENT
The generalisation of emotional learning in humans
I hereby wish to WITHDRAW my consent to participate in the research proposal described above and understand that such withdrawal WILL NOT jeopardise my relationship with the researchers, the School of Psychology or the University of New South Wales.
…………………………………………………… ………………………………………………… Signature of Research Participant Please PRINT name
…………………………………………………… Date
The section for Revocation of Consent should be forwarded to Alex Wong ([email protected])
341
Approval No HC14003
THE UNIVERSITY OF NEW SOUTH WALES PARTICIPANT INFORMATION STATEMENT The generalisation of emotional learning in humans
Participant Selection and Purpose of Study You are invited to participate in a study on the physiological responses produced by emotional stimuli. We hope to learn more about the cognitive processes involved in learning and anxiety. You were selected as a possible participant in this study because you responded to an advertisement for research on learning and emotion.
Description of Study and Risks The emotional stimulus will be electric shock, presented through electrodes attached to one finger. The physiological measure recorded will be skin conductance, a sensitive measure of emotional state. If you decide to participate, we will ask you to sample the electric shock and select a level of shock which you personally judge to be definitely uncomfortable but not painful. The experiment consists of 2 parts – the first part being a categorization task, in which you have to categorize items into the most appropriate category. In the second part of the experiment, you will be presented with some visual stimuli, and asked to make responses. You may receive up to 10 electric shocks over a testing period of approximately 30 minutes. There will be rest periods between the test trials, during which no shock is given. After the experiment we will also ask you to report your thoughts about the stimuli. If you decide to participate but change your mind during the experiment, please notify us straight away and we will terminate the experiment.
Confidentiality and Disclosure of Information Any information that is obtained in connection with this study and that can be identified with you will remain confidential and will be disclosed only with your permission or except as required by law. If you give us your permission by signing this document, we plan to publish the results in scientific journals. In any publication, information will be provided in such a way that you cannot be identified, for example by presenting only group data.
Recompense to participants You will receive $15 to compensate you for your time.
Your consent Your decision whether or not to participate will not prejudice your future relations with The University of New South Wales. If you decide to participate, you are free to withdraw your consent and to discontinue participation at any time, without disadvantage. For example, if you become excessively anxious at any stage, we ask you to tell us so that we can terminate the experiment immediately. You will still receive your payment for participation. You are reminded that your involvement in the School of Psychology research participation scheme is voluntary, and that you may obtain the same course marks through an alternative form of assessment.
Inquiries If you have any questions or concerns about the research, Professor Peter Lovibond (9385 3830; [email protected]) will be happy to address them.
Complaints may be directed to the Ethics Secretariat, The University of New South Wales, SYDNEY 2052 AUSTRALIA (phone 9385 4234, fax 9385 6648, email [email protected]).
You may keep this information sheet and a copy of the Participant Consent Form. The investigator will keep the other signed copy. Both copies should be signed by you and the investigator. 342
THE UNIVERSITY OF NEW SOUTH WALES Approval No HC14003 PARTICIPANT CONSENT FORM
The generalisation of emotional learning in humans
You are making a decision whether or not to participate. Before that, we ask you to answer the following question concerning any potential heart problem you might have.
DO YOU HAVE A HEART PROBLEM? No Yes
Your signature below indicates that: a) you have read the Participant Information Statement, b) you have answered “No” to the above question about heart problem, and c) you have decided to participate.
…………………………………………………… ……………………………………………… Signature of Research Participant Please PRINT name
…………………………………………………… Date
…………………………………………………… …………………………………………………. Signature of Investigator Please PRINT Name
......
REVOCATION OF CONSENT
The generalisation of emotional learning in humans
I hereby wish to WITHDRAW my consent to participate in the research proposal described above and understand that such withdrawal WILL NOT jeopardise my relationship with the researchers, the School of Psychology or the University of New South Wales.
…………………………………………………… ………………………………………………… Signature of Research Participant Please PRINT name
…………………………………………………… Date
The section for Revocation of Consent should be forwarded to Alex Wong ([email protected])
343
Stimuli used in Validation Questionnaires
Bacon Baked beans
Boiled egg Cereal
Cornflake Hashbrown
Oatmeal Pan-fried egg
Pancakes Sausage
Scrambled egg Waffles
344
Apple & Walnut log Apple pie
Baguette Ciabatta roll
Cupcake Custard pie
Finger bun Garlic bread
Hamburger bun Hotcross bun
Jam bunlet Pretzel 345
Bagel Croissant
English muffin Toast
Spaghetti Steak
Aeroplane Spongebob®
346
Pictures used in the Categorical task in Experiment 2 & 3
Breakfast items:
Bakery items:
347
Post-experimental questionnaire
In the experiment, how did you predict if the picture would be followed by a shock?
Similarly, how did you predict if the picture would NOT be followed by a shock?
If you came up with any strategies/rules in the experiment, please write these below as well.
______
______
______
______
______
______
______
______
______
______
______
______
______
______
______
______
______
______
______
348
The pictures presented can actually be categorized into 2 main groups: BREAKFAST and BAKERY. Knowing that now, please rate the extent to which you consider the following statements to be true (Circle the corresponding vertical lines):
a) When pictures in the BREAKFAST category were shown, shock would be presented
0% 50% 100% |____|____|____|____|____|____|____|____|____|____|
False True
b) When pictures in the BAKERY category were shown, shock would be presented
0% 50% 100% |____|____|____|____|____|____|____|____|____|____|
False True
c) When pictures can be simultaneously fitted in both BREAKFAST and BAKERY categories, shock would be presented
0% 50% 100% |____|____|____|____|____|____|____|____|____|____|
False True
d) When pictures were neither in the BREAKFAST nor BAKERY categories, shock would be presented 50% 100% 0% |____|____|____|____|____|____|____|____|____|____|
False True
e) Other (please specify)
______
0% 100% |____|____|____|____|____|____|____|____|____|____| 50% False True
349
Validation Questionnaires
Analyses of the typicality ratings of breakfast and bakery items
W1 = Main effect of category W2 = Main effect of typicality ratings W3 = Interaction between category and typicality ratings
Analysis of Variance Summary Table
Source SS df MS F ------Between 90.957 25 3.638 ------Within ------Category W1 11.422 1 11.422 3.461 Error 82.507 25 3.300 Typicality W2 94.865 1 94.865 33.179 Error 71.479 25 2.859 Interaction W3 2180.532 1 2180.532 329.647 Error 165.369 25 6.615 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Category W1 -0.271 0.145 -0.570 0.029 Typicality W2 0.780 0.135 0.501 1.059 Interaction W3 3.739 0.206 3.315 4.163 ------
350
Analyses of the typicality ratings among breakfast items
W1 = Difference between breakfast typicality ratings and bakery typicality ratings
Analysis of Variance Summary Table
Source SS df MS F ------Between 69.023 25 2.761 ------Within ------W1 1592.514 1 1592.514 248.996 Error 159.894 25 6.396 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------W1 4.519 0.286 3.929 5.108 ------
351
Analyses of the typicality ratings among bakery items
W1 = Difference between breakfast typicality ratings and bakery typicality ratings
Analysis of Variance Summary Table
Source SS df MS F ------Between 104.441 25 4.178 ------Within ------W1 682.884 1 682.884 221.846 Error 76.955 25 3.078 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------W1 -2.959 0.199 -3.368 -2.550 ------
352
Analyses of the typicality ratings of cross-classified items
W1 = Main effect of typicality ratings
Analysis of Variance Summary Table
Source SS df MS F ------Between 66.231 25 2.649 ------Within ------Typicality W1 4.024 1 4.024 3.094 Error 32.518 25 1.301 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Typicality W1 0.321 0.183 -0.055 0.697 ------
353
Experiment 2
Acquisition phase (Expectancy data)
B1 = Between groups W1 = Main effect of CS trial type W2 = Main effect of block W3 = Interaction between CS trial type and block
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 1159.789 1 1159.789 2.168 Error 31031.455 58 535.025 ------Within ------CS type W1 1.10214e+6 1 1.10214e+6 1562.876 B1W1 38.792 1 38.792 0.055 Error 40901.623 58 705.200 Block W2 6929.158 1 6929.158 20.515 B1W2 387.543 1 387.543 1.147 Error 19589.856 58 337.756 Interaction W3 108226.787 1 108226.787 267.618 B1W3 26.402 1 26.402 0.065 Error 23455.649 58 404.408 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 2.200 1.494 -0.791 5.190 CS type W1 67.804 1.715 64.371 71.237 B1W1 -0.805 3.430 -7.671 6.062 Block W2 -5.376 1.187 -7.752 -3.000 B1W2 2.543 2.374 -2.209 7.295 Interaction W3 -21.247 1.299 -23.847 -18.647 B1W3 0.664 2.598 -4.536 5.863 ------
354
Acquisition phase (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.000 1 0.000 0.017 Error 0.040 58 0.001 ------Within ------CS type W1 0.010 1 0.010 40.038 B1W1 0.001 1 0.001 5.565 Error 0.015 58 0.000 Block W2 0.000 1 0.000 0.560 B1W2 0.000 1 0.000 0.162 Error 0.014 58 0.000 Interaction W3 0.001 1 0.001 5.793 B1W3 1.0853e-10 1 1.0853e-10 4.29555e-7 Error 0.015 58 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.000 0.002 -0.003 0.004 CS type W1 0.007 0.001 0.004 0.009 B1W1 0.005 0.002 0.001 0.009 Block W2 -0.001 0.001 -0.003 0.001 B1W2 0.001 0.002 -0.003 0.005 Interaction W3 -0.002 0.001 -0.005 0.000 B1W3 1.34570e-6 0.002 -0.004 0.004 ------
355
Test phase (Expectancy data)
B1 = Between group W1 = Threat trial type W2 = Novelty trial type (CS vs GENs) W3 = Interaction of threat trial type and novelty trial type W4 = Comparison between ambiguous and non-ambiguous exemplars W5 = Comparison between CC and filler exemplars
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 5105.453 1 5105.453 7.007 Error 42257.164 58 728.572 ------Within ------Threat type W1 706045.309 1 706045.309 3287.460 B1W1 4.307 1 4.307 0.020 Error 12456.617 58 214.769 CS vs GEN W2 0.009 1 0.009 0.000 B1W2 0.670 1 0.670 0.016 Error 2359.481 58 40.681 Interaction W3 688.190 1 688.190 18.074 B1W3 7.476 1 7.476 0.196 Error 1602.495 58 27.629 Ambigity W4 17.457 1 17.457 0.032 B1W4 6455.840 1 6455.840 11.800 Error 31731.771 58 547.099 CC vs Filler W5 30655.718 1 30655.718 11.039 B1W5 8034.281 1 8034.281 2.893 Error 161072.429 58 2777.111 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 5.566 2.102 1.357 9.774 Threat type W1 88.621 1.546 85.527 91.715 B1W1 0.438 3.091 -5.750 6.626 CS vs GEN W2 -0.010 0.713 -1.439 1.418 B1W2 -0.183 1.427 -3.040 2.673 Interaction W3 2.935 0.690 1.553 4.316 B1W3 0.612 1.381 -2.152 3.375 Ambigity W4 0.327 1.830 -3.335 3.989 B1W4 -12.569 3.659 -19.894 -5.245 CC vs Filler W5 20.646 6.214 8.207 33.084 B1W5 21.139 12.428 -3.739 46.016 ------
356
Test phase (Skin conductance data)
B1 = Between group W1 = Comparison between GEN+ and GEN-
W2 = Comparison between GEN and CC
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.001 1 0.001 3.897 Error 0.022 58 0.000 ------Within ------GEN type W1 0.005 1 0.005 16.046 B1W1 0.000 1 0.000 1.301 Error 0.017 58 0.000 GEN vs CC W2 0.004 1 0.004 15.360 B1W2 0.000 1 0.000 0.162 Error 0.017 58 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.005 0.003 0.000 0.010 GEN type W1 0.013 0.003 0.006 0.019 B1W1 0.007 0.006 -0.005 0.020 GEN vs CC W2 0.009 0.002 0.004 0.013 B1W2 -0.002 0.004 -0.011 0.007 ------
357
Direct comparison in responding to CC exemplars between groups (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------B1 19295.649 1 19295.649 9.462 Error 120320.528 59 2039.331 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------B1 20.562 6.685 7.186 33.938 ------
358
Direct comparison in responding to CC exemplars between groups (Skin conductance
data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------B1 0.001 1 0.001 4.749 Error 0.012 58 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------B1 0.006 0.003 0.000 0.011 ------
359
Experiment 3
Acquisition phase (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------B1 19.886 1 19.886 0.043 Error 23800.729 52 457.706 ------Within ------CS type W1 917342.209 1 917342.209 729.487 B1W1 2728.092 1 2728.092 2.169 Error 65390.860 52 1257.517 Block W2 4056.453 1 4056.453 14.895 B1W2 171.232 1 171.232 0.629 Error 14161.884 52 272.344 Interaction W3 116625.897 1 116625.897 270.204 B1W3 576.737 1 576.737 1.336 Error 22444.294 52 431.621 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 -0.304 1.457 -3.227 2.619 CS type W1 65.213 2.415 60.368 70.058 B1W1 -7.113 4.829 -16.803 2.577 Block W2 -4.337 1.124 -6.591 -2.082 B1W2 1.782 2.247 -2.728 6.291 Interaction W3 -23.252 1.415 -26.091 -20.414 B1W3 3.270 2.829 -2.407 8.947 ------
360
Acquisition phase (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.000 1 0.000 0.120 Error 0.051 52 0.001 ------Within ------CS type W1 0.005 1 0.005 17.338 B1W1 2.26978e-6 1 2.26978e-6 0.007 Error 0.016 52 0.000 Block W2 0.000 1 0.000 0.668 B1W2 0.000 1 0.000 0.066 Error 0.013 52 0.000 Interaction W3 0.001 1 0.001 7.661 B1W3 0.000 1 0.000 1.689 Error 0.008 52 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.001 0.002 -0.004 0.005 CS type W1 0.005 0.001 0.003 0.007 B1W1 0.000 0.002 -0.005 0.005 Block W2 0.001 0.001 -0.001 0.003 B1W2 0.001 0.002 -0.004 0.005 Interaction W3 -0.002 0.001 -0.004 -0.001 B1W3 -0.002 0.002 -0.005 0.001 ------
361
Test phase (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 5758.611 1 5758.611 6.528 Error 45871.599 52 882.146 ------Within ------Threat type W1 473781.984 1 473781.984 407.791 B1W1 5620.840 1 5620.840 4.838 Error 60414.854 52 1161.824 CS vs GEN W2 7.531 1 7.531 0.030 B1W2 170.770 1 170.770 0.676 Error 13126.936 52 252.441 Interaction W3 2106.004 1 2106.004 14.319 B1W3 183.195 1 183.195 1.246 Error 7647.867 52 147.074 Ambiguity W4 706.139 1 706.139 1.589 B1W4 4862.960 1 4862.960 10.941 Error 23111.965 52 444.461 CC vs Filler W5 13922.102 1 13922.102 5.913 B1W5 17032.283 1 17032.283 7.234 Error 122436.166 52 2354.542 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 6.232 2.439 1.337 11.126 Threat type W1 76.532 3.790 68.927 84.137 B1W1 -16.672 7.580 -31.882 -1.462 CS vs GEN W2 0.324 1.874 -3.436 4.084 B1W2 -3.082 3.748 -10.602 4.438 Interaction W3 5.412 1.430 2.542 8.282 B1W3 3.192 2.860 -2.547 8.932 Ambiguity W4 -2.191 1.738 -5.680 1.297 B1W4 -11.501 3.477 -18.477 -4.524 CC vs Filler W5 14.668 6.032 2.564 26.772 B1W5 32.447 12.064 8.239 56.656 ------
362
Test phase (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.001 1 0.001 0.944 Error 0.030 52 0.001 ------Within ------GEN type W1 0.002 1 0.002 13.923 B1W1 5.89621e-6 1 5.89621e-6 0.037 Error 0.008 52 0.000 GEN vs CC W2 0.000 1 0.000 0.245 B1W2 0.002 1 0.002 6.248 Error 0.018 52 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.003 0.003 -0.003 0.010 GEN type W1 0.009 0.002 0.004 0.014 B1W1 0.001 0.005 -0.009 0.011 GEN vs CC W2 0.001 0.003 -0.004 0.006 B1W2 -0.013 0.005 -0.023 -0.003 ------
363
Follow-up test for HA group (Expectancy data)
1. Group differences between ratings to threat (CS+ & GEN+) and safety (CS- & GEN-) cues
Analysis of Variance Summary Table
Source SS df MS F ------Between 28823.920 25 1152.957 ------Within ------Threat type W1 181378.900 1 181378.900 141.434 Error 32060.658 25 1282.426 CS vs GEN W2 51.386 1 51.386 0.140 Error 9190.271 25 367.611 Interaction W3 1702.674 1 1702.674 8.033 Error 5298.669 25 211.947 Ambiguity W4 4472.005 1 4472.005 8.710 Error 12835.777 25 513.431 CC vs Filler W5 29773.346 1 29773.346 22.254 Error 33447.930 25 1337.917 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Threat type W1 68.196 5.734 56.386 80.006 CS vs GEN W2 -1.217 3.256 -7.924 5.489 Interaction W3 7.008 2.473 1.916 12.101 Ambiguity W4 -7.941 2.691 -13.483 -2.400 CC vs Filler W5 30.891 6.548 17.405 44.378 ------
Analysis of Variance Summary Table
Source SS df MS F ------Between 28823.920 25 1152.957 ------Within ------CS- vs GEN- W1 1172.825 1 1172.825 3.503 Error 8370.392 25 334.816 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------CS- vs GEN- W1 -8.226 4.395 -17.277 0.826 ------364
Follow-up test for LA group (Expectancy data)
1. Group differences between ratings to threat (CS+ & GEN+) and safety (CS- & GEN-) cues
Analysis of Variance Summary Table
Source SS df MS F ------Between 17047.679 27 631.396 ------Within ------Threat type W1 302510.271 1 302510.271 288.062 Error 28354.196 27 1050.155 CS vs GEN W2 129.819 1 129.819 0.890 Error 3936.665 27 145.802 Interaction W3 543.597 1 543.597 6.248 Error 2349.198 27 87.007 Ambiguity W4 967.289 1 967.289 2.541 Error 10276.188 27 380.600 CC vs Filler W5 81.335 1 81.335 0.025 Error 88988.236 27 3295.861 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Threat type W1 84.868 5.000 74.608 95.128 CS vs GEN W2 1.865 1.976 -2.190 5.920 Interaction W3 3.816 1.527 0.683 6.948 Ambiguity W4 3.559 2.232 -1.022 8.140 CC vs Filler W5 -1.556 9.904 -21.877 18.766 ------
Analysis of Variance Summary Table
Source SS df MS F ------Between 17047.679 27 631.396 ------Within ------CS+ vs GEN+ W1 602.357 1 602.357 3.126 Error 5202.835 27 192.698 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------CS+ vs GEN+ W1 5.681 3.213 -0.912 12.273 ------365
Follow-up test for HA group (Expectancy data)
2. Group differences between ratings to CC and filler exemplars
Analysis of Variance Summary Table
Source SS df MS F ------Between 29866.521 25 1194.661 ------Within ------CC vs Filler W1 29773.346 1 29773.346 22.254 Error 33447.930 25 1337.917 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------CC vs Filler W1 30.891 6.548 17.405 44.378 ------
366
Follow-up test for LA group (Expectancy data)
2. Group differences between ratings to CC and filler exemplars
Analysis of Variance Summary Table
Source SS df MS F ------Between 21960.937 27 813.368 ------Within ------CC vs Filler W1 81.335 1 81.335 0.025 Error 88988.236 27 3295.861 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------CC vs Filler W1 -1.556 9.904 -21.877 18.766 ------
367
Follow-up test for HA group (Skin conductance data)
3. Group differences in responding to CC and filler exemplars
Analysis of Variance Summary Table
Source SS df MS F ------Between 0.019 25 0.001 ------Within ------GEN vs Filler W1 0.001 1 0.001 1.817 Error 0.009 25 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------GEN vs Filler W1 -0.005 0.004 -0.013 0.003 ------
368
Follow-up test for LA group (Skin conductance data)
3. Group differences in responding to CC and filler exemplars
Analysis of Variance Summary Table
Source SS df MS F ------Between 0.011 27 0.000 ------Within ------GEN vs Filler W1 0.002 1 0.002 4.961 Error 0.009 27 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------GEN v Filler W1 0.008 0.003 0.001 0.015 ------
369
Categorical test (Expectancy data)
B1 = Between anxiety groups
B2 = Main effect of categorization
B3 = Interaction between groups and categorization
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 16382.421 1 16382.421 9.020 Categorize B2 9636.747 1 9636.747 5.306 Interaction B3 724.151 1 724.151 0.399 Error 34508.311 19 1816.227 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 36.780 12.246 11.148 62.412 Categorize B2 28.209 12.246 2.577 53.841 Interaction B3 -7.733 12.246 -33.365 17.899 ------
370
Categorical test (Skin conductance data)
^
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.003 1 0.003 6.334 Categorize B2 0.000 1 0.000 0.022 Interaction B3 0.000 1 0.000 0.571 Error 0.010 19 0.001 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.020 0.008 0.003 0.037 Categorize B2 0.001 0.008 -0.016 0.018 Interaction B3 0.006 0.008 -0.011 0.023 ------
371
Appendix C
Experiment 4 and 5
Information Statement and Consent forms……………………………………………… 372
»›Course credit compensation…………………………………………………… 372
»›Financial compensation……………………………………………………….. 374
Post-experimental questionnaire………………………………………………………… 376
Output from statistical analyses………………………………………………………….. 377
»›Experiment 4………………………………………………………………….. 377
»›Experiment 5………………………………………………………………….. 389
372
Approval No HC14003
THE UNIVERSITY OF NEW SOUTH WALES PARTICIPANT INFORMATION STATEMENT The generalisation of emotional learning in humans
Participant Selection and Purpose of Study You are invited to participate in a study on the physiological responses produced by emotional stimuli. We hope to learn more about the cognitive processes involved in learning and anxiety. You were selected as a possible participant in this study because you are enrolled in Psychology 1.
Description of Study and Risks The emotional stimulus will be electric shock, presented through electrodes attached to one finger. The physiological measure recorded will be skin conductance, a sensitive measure of emotional state. If you decide to participate, we will ask you to sample the electric shock and select a level of shock which you personally judge to be definitely uncomfortable but not painful. During the experiment, you will be presented with a series of visual stimuli, and asked to make responses. You may receive up to 10 electric shocks over a testing period of approximately 20 minutes. There will be rest periods between the test trials, during which no shock is given. After the experiment we will also ask you to report your thoughts about the stimuli. If you decide to participate but change your mind during the experiment, please notify us straight away and we will terminate the experiment.
Confidentiality and Disclosure of Information Any information that is obtained in connection with this study and that can be identified with you will remain confidential and will be disclosed only with your permission or except as required by law. If you give us your permission by signing this document, we plan to publish the results in scientific journals. In any publication, information will be provided in such a way that you cannot be identified, for example by presenting only group data.
Recompense to participants You will receive 0.5 hour research credit for your participation.
Your consent Your decision whether or not to participate will not prejudice your future relations with The University of New South Wales. If you decide to participate, you are free to withdraw your consent and to discontinue participation at any time, without disadvantage. For example, if you become excessively anxious at any stage, we ask you to tell us so that we can terminate the experiment immediately. You will still receive your course credit. You are reminded that your involvement in the School of Psychology research participation scheme is voluntary, and that you may obtain the same course marks through an alternative form of assessment.
Inquiries If you have any questions or concerns following your participantion, Alex Wong (0404118987; [email protected]) and Professor Peter Lovibond (9385 3830; [email protected]) will be happy to address them.
Complaints may be directed to the Ethics Secretariat, The University of New South Wales, SYDNEY 2052 AUSTRALIA (phone 9385 4234, fax 9385 6648, email [email protected]).
You may keep this information sheet and a copy of the Participant Consent Form. The investigator will keep the other signed copy. Both copies should be signed by you and the investigator.
373
THE UNIVERSITY OF NEW SOUTH WALES Approval No HC14003 PARTICIPANT CONSENT FORM
The generalisation of emotional learning in humans
You are making a decision whether or not to participate. Before that, we ask you to answer the following question concerning any potential heart problem you might have.
DO YOU HAVE A HEART PROBLEM? No Yes
Your signature below indicates that: a) you have read the Participant Information Statement, b) you have answered “No” to the above question about heart problem, and c) you have decided to participate.
…………………………………………………… ……………………………………………… Signature of Research Participant Please PRINT name
…………………………………………………… Date
…………………………………………………… …………………………………………………. Signature of Investigator Please PRINT Name
......
REVOCATION OF CONSENT
The generalisation of emotional learning in humans
I hereby wish to WITHDRAW my consent to participate in the research proposal described above and understand that such withdrawal WILL NOT jeopardise my relationship with the researchers, the School of Psychology or the University of New South Wales.
…………………………………………………… ………………………………………………… Signature of Research Participant Please PRINT name
…………………………………………………… Date
The section for Revocation of Consent should be forwarded to Alex Wong ([email protected])
374
Approval No HC14003 THE UNIVERSITY OF NEW SOUTH WALES PARPTICIPANT INFORMATION STATEMENT The generalisation of emotional learning in humans
Participant Selection and Purpose of Study You are invited to participate in a study on the physiological responses produced by emotional stimuli. We hope to learn more about the cognitive processes involved in learning and anxiety. You were selected as a possible participant in this study because your responded to an advertisement for research on learning and emotion.
Description of Study and Risks The emotional stimulus will be electric shock, presented through electrodes attached to one finger. The physiological measure recorded will be skin conductance, a sensitive measure of emotional state. If you decide to participate, we will ask you to sample the electric shock and select a level of shock which you personally judge to be definitely uncomfortable but not painful. During the experiment, you will be presented with a series of visual stimuli, and asked to make responses. You may receive up to 10 electric shocks over a testing period of approximately 20 minutes. There will be rest periods between the test trials, during which no shock is given. After the experiment we will also ask you to report your thoughts about the stimuli. If you decide to participate but change your mind during the experiment, please notify us straight away and we will terminate the experiment.
Confidentiality and Disclosure of Information Any information that is obtained in connection with this study and that can be identified with you will remain confidential and will be disclosed only with your permission or except as required by law. If you give us your permission by signing this document, we plan to publish the results in scientific journals. In any publication, information will be provided in such a way that you cannot be identified, for example by presenting only group data.
Recompense to participants You will receive $10 to compensate for your time.
Your consent Your decision whether or not to participate will not prejudice your future relations with The University of New South Wales. If you decide to participate, you are free to withdraw your consent and to discontinue participation at any time, without disadvantage. For example, if you become excessively anxious at any stage, we ask you to tell us so that we can terminate the experiment immediately. You will still receive your course credit. You are reminded that your involvement in the School of Psychology research participation scheme is voluntary, and that you may obtain the same course marks through an alternative form of assessment.
Inquiries If you have any questions or concerns following your participantion, Alex Wong (0404118987; [email protected]) and Professor Peter Lovibond (9385 3830; [email protected]) will be happy to address them.
Complaints may be directed to the Ethics Secretariat, The University of New South Wales, SYDNEY 2052 AUSTRALIA (phone 9385 4234, fax 9385 6648, email [email protected]).
You may keep this information sheet and a copy of the Participant Consent Form. The investigator will keep the other signed copy. Both copies should be signed by you and the investigator.
375
THE UNIVERSITY OF NEW SOUTH WALES Approval No HC14003 PARTICIPANT CONSENT FORM
The generalisation of emotional learning in humans
You are making a decision whether or not to participate. Before that, we ask you to answer the following question concerning any potential heart problem you might have.
DO YOU HAVE A HEART PROBLEM? No Yes
Your signature below indicates that: a) you have read the Participant Information Statement, b) you have answered “No” to the above question about heart problem, and c) you have decided to participate.
…………………………………………………… ……………………………………………… Signature of Research Participant Please PRINT name
…………………………………………………… Date
…………………………………………………… …………………………………………………. Signature of Investigator Please PRINT Name
......
REVOCATION OF CONSENT
The generalisation of emotional learning in humans
I hereby wish to WITHDRAW my consent to participate in the research proposal described above and understand that such withdrawal WILL NOT jeopardise my relationship with the researchers, the School of Psychology or the University of New South Wales.
…………………………………………………… ………………………………………………… Signature of Research Participant Please PRINT name
…………………………………………………… Date
The section for Revocation of Consent should be forwarded to Alex Wong ([email protected])
376
Post-experimental questionnaire
How many different colour(s) was/were presented in the experiment? Name the colour(s) you saw.
______
______
______
______
______
______
______
377
Experiment 4
Acquisition phase (Expectancy data)
B1 = Between anxiety groups B1W1 = Interaction between B1 and W1 B2 = Between conditions B2W1 = Interaction between B2 and W1 B3 = Interaction between groups and conditions B3W1 = 3-way interaction between B3 and W1 W1 = Linear trend
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 169.325 1 169.325 0.157 Condition B2 435.837 1 435.837 0.404 Interaction B3 62.871 1 62.871 0.058 Error 103437.547 96 1077.474 ------Within ------Linear W1 30842.275 1 30842.275 53.353 B1W1 0.021 1 0.021 0.000 B2W1 3533.028 1 3533.028 6.112 B3W1 6906.729 1 6906.729 11.948 Error 55495.635 96 578.080 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 -0.921 2.323 -5.532 3.690 Condition B2 -1.477 2.323 -6.088 3.134 Interaction B3 0.561 2.323 -4.050 5.172 Linear W1 13.614 1.864 9.915 17.314 B1W1 -0.023 3.728 -7.422 7.377 B2W1 -9.216 3.728 -16.615 -1.816 B3W1 -12.885 3.728 -20.285 -5.486 ------
378
Acquisition phase (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.000 1 0.000 0.413 Condition B2 0.000 1 0.000 0.925 Interaction B3 4.20929e-7 1 4.20929e-7 0.001 Error 0.036 96 0.000 ------Within ------Linear W1 0.003 1 0.003 18.030 B1W1 0.000 1 0.000 0.846 B2W1 0.000 1 0.000 0.325 B3W1 0.000 1 0.000 1.913 Error 0.015 96 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 -0.001 0.001 -0.004 0.002 Condition B2 -0.001 0.001 -0.004 0.001 Interaction B3 0.000 0.001 -0.003 0.003 Linear W1 -0.004 0.001 -0.006 -0.002 B1W1 0.002 0.002 -0.002 0.006 B2W1 -0.001 0.002 -0.005 0.003 B3W1 -0.003 0.002 -0.006 0.001 ------
379
Extinction phase (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 39965.721 1 39965.721 10.128 Condition B2 110089.460 1 110089.460 27.899 Interaction B3 5942.724 1 5942.724 1.506 Error 378809.899 96 3945.936 ------Within ------Linear W1 108427.776 1 108427.776 184.246 B1W1 3107.182 1 3107.182 5.280 B2W1 602.290 1 602.290 1.023 B3W1 2051.458 1 2051.458 3.486 Error 56495.524 96 588.495 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 14.147 4.445 5.323 22.971 Condition B2 -23.480 4.445 -32.304 -14.656 Interaction B3 5.455 4.445 -3.369 14.279 Linear W1 -25.527 1.881 -29.260 -21.794 B1W1 8.642 3.761 1.177 16.108 B2W1 3.805 3.761 -3.661 11.271 B3W1 -7.022 3.761 -14.488 0.444 ------
380
Extinction phase (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 7.33484e-6 1 7.33484e-6 0.024 Condition B2 0.000 1 0.000 0.117 Interaction B3 0.000 1 0.000 0.363 Error 0.030 96 0.000 ------Within ------Linear W1 0.000 1 0.000 1.374 B1W1 0.000 1 0.000 0.824 B2W1 0.000 1 0.000 2.642 B3W1 3.50470e-7 1 3.50470e-7 0.005 Error 0.007 96 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.000 0.001 -0.002 0.003 Condition B2 0.000 0.001 -0.003 0.002 Interaction B3 0.001 0.001 -0.002 0.003 Linear W1 0.001 0.001 -0.001 0.002 B1W1 -0.001 0.001 -0.004 0.001 B2W1 0.002 0.001 0.000 0.005 B3W1 0.000 0.001 -0.003 0.003 ------
381
Comparison of last acquisition trial to first extinction trial (Expectancy data)
W1 = Comparison between last acquisition trial and first extinction trial
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 50.781 1 50.781 0.115 Condition B2 7140.352 1 7140.352 16.142 Interaction B3 12.567 1 12.567 0.028 Error 42466.363 96 442.358 ------Within ------ACQ vs EXT W1 4885.312 1 4885.312 13.551 B1W1 242.887 1 242.887 0.674 B2W1 1812.925 1 1812.925 5.029 B3W1 681.812 1 681.812 1.891 Error 34609.476 96 360.515 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 1.009 2.977 -4.900 6.917 Condition B2 -11.960 2.977 -17.869 -6.051 Interaction B3 0.502 2.977 -5.407 6.411 ACQ vs EXT W1 9.893 2.687 4.558 15.227 B1W1 -4.412 5.375 -15.080 6.257 B2W1 12.053 5.375 1.384 22.721 B3W1 -7.391 5.375 -18.060 3.277 ------
382
Comparison of last acquisition trial to first extinction trial (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.000 1 0.000 0.388 Condition B2 1.50450e-7 1 1.50450e-7 0.001 Interaction B3 0.000 1 0.000 0.118 Error 0.013 96 0.000 ------Within ------ACQ vs EXT W1 0.000 1 0.000 0.215 B1W1 0.000 1 0.000 0.611 B2W1 0.000 1 0.000 0.548 B3W1 0.000 1 0.000 4.158 Error 0.006 96 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 -0.001 0.002 -0.004 0.002 Condition B2 0.000 0.002 -0.003 0.003 Interaction B3 0.001 0.002 -0.003 0.004 ACQ vs EXT W1 -0.001 0.001 -0.003 0.002 B1W1 0.002 0.002 -0.003 0.006 B2W1 -0.002 0.002 -0.006 0.003 B3W1 -0.005 0.002 -0.009 0.000 ------
383
Test phase (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 27978.236 1 27978.236 13.512 Condition B2 14145.939 1 14145.939 6.832 Interaction B3 10.289 1 10.289 0.005 Error 198777.397 96 2070.598 ------Within ------Linear W1 11865.389 1 11865.389 52.363 B1W1 150.510 1 150.510 0.664 B2W1 2593.809 1 2593.809 11.447 B3W1 0.002 1 0.002 7.35272e-6 Error 21753.503 96 226.599 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 19.330 5.259 8.892 29.768 Condition B2 13.745 5.259 3.307 24.183 Interaction B3 0.371 5.259 -10.067 10.809 Linear W1 -15.417 2.131 -19.646 -11.188 B1W1 3.473 4.261 -4.985 11.931 B2W1 -14.417 4.261 -22.875 -5.958 B3W1 -0.012 4.261 -8.470 8.447 ------
384
Test phase (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.000 1 0.000 0.246 Condition B2 0.001 1 0.001 3.756 Interaction B3 0.000 1 0.000 0.407 Error 0.021 96 0.000 ------Within ------Linear W1 0.001 1 0.001 5.317 B1W1 0.000 1 0.000 0.260 B2W1 0.002 1 0.002 13.685 B3W1 0.000 1 0.000 0.122 Error 0.015 96 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.001 0.002 -0.003 0.004 Condition B2 0.003 0.002 0.000 0.007 Interaction B3 -0.001 0.002 -0.004 0.002 Linear W1 -0.004 0.002 -0.008 -0.001 B1W1 -0.002 0.004 -0.009 0.005 B2W1 -0.013 0.004 -0.020 -0.006 B3W1 -0.001 0.004 -0.008 0.006 ------
385
Comparison of last extinction trial to first test trial (Expectancy data)
W1 = Comparison between the last extinction trial and the first test trial
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 13471.678 1 13471.678 13.716 Condition B2 373.097 1 373.097 0.380 Interaction B3 0.372 1 0.372 0.000 Error 94288.770 96 982.175 ------Within ------EXT vs TEST W1 17006.969 1 17006.969 30.611 B1W1 11.215 1 11.215 0.020 B2W1 19740.737 1 19740.737 35.531 B3W1 37.127 1 37.127 0.067 Error 53336.776 96 555.591 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 16.428 4.436 7.623 25.232 Condition B2 2.734 4.436 -6.071 11.539 Interaction B3 0.086 4.436 -8.718 8.891 EXT vs TEST W1 -18.458 3.336 -25.080 -11.836 B1W1 -0.948 6.672 -14.192 12.296 B2W1 -39.772 6.672 -53.016 -26.527 B3W1 1.725 6.672 -11.519 14.969 ------
386
Comparison of last extinction trial to first test trial (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 1.52383e-6 1 1.52383e-6 0.008 Condition B2 0.002 1 0.002 8.076 Interaction B3 0.000 1 0.000 0.526 Error 0.019 96 0.000 ------Within ------EXT vs TEST W1 0.002 1 0.002 12.306 B1W1 0.000 1 0.000 1.036 B2W1 0.001 1 0.001 8.228 B3W1 0.000 1 0.000 0.690 Error 0.014 96 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.000 0.002 -0.004 0.004 Condition B2 0.006 0.002 0.002 0.010 Interaction B3 0.001 0.002 -0.003 0.005 EXT vs TEST W1 -0.006 0.002 -0.009 -0.003 B1W1 -0.003 0.003 -0.010 0.003 B2W1 -0.010 0.003 -0.016 -0.003 B3W1 0.003 0.003 -0.004 0.009 ------
387
Exploratory analyses
Comparison of last acquisition trial to first test trial (Expectancy data)
ABA condition only
B1 = Between anxiety groups
W1 = Comparison between the last acquisition trial and the first test trial
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 859.222 1 859.222 1.098 Error 37557.470 48 782.447 ------Within ------ACQ vs TEST W1 4913.184 1 4913.184 8.271 B1W1 2626.632 1 2626.632 4.422 Error 28512.885 48 594.018 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 5.867 5.599 -5.390 17.125 ACQ vs TEST W1 14.030 4.878 4.221 23.839 B1W1 -20.517 9.757 -40.134 -0.899 ------
388
Comparison of last acquisition trial to first test trial (Skin conductance data)
ABA condition only
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 4.53717e-7 1 4.53717e-7 0.002 Error 0.014 48 0.000 ------Within ------ACQ vs TEST W1 0.004 1 0.004 21.511 B1W1 0.000 1 0.000 0.443 Error 0.008 48 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.000 0.003 -0.007 0.007 ACQ vs TEST W1 -0.012 0.003 -0.017 -0.007 B1W1 -0.003 0.005 -0.014 0.007 ------
389
Experiment 5
Acquisition phase (Expectancy data)
B1 = Between anxiety groups B1W1 = Interaction between B1 and W1 B2 = Between conditions B2W1 = Interaction between B2 and W1 B3 = Interaction between groups and conditions B3W1 = 3-way interaction between B3 and W1 W1 = Linear trend
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 741.358 1 741.358 0.676 Condition B2 285.579 1 285.579 0.261 Interaction B3 133.881 1 133.881 0.122 Error 106327.406 97 1096.159 ------Within ------Linear W1 35041.746 1 35041.746 61.204 B1W1 1.064 1 1.064 0.002 B2W1 672.970 1 672.970 1.175 B3W1 87.449 1 87.449 0.153 Error 55536.366 97 572.540 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 1.920 2.335 -2.714 6.555 Condition B2 1.192 2.335 -3.443 5.827 Interaction B3 0.816 2.335 -3.819 5.451 Linear W1 14.463 1.849 10.794 18.133 B1W1 0.159 3.698 -7.179 7.498 B2W1 -4.009 3.698 -11.347 3.330 B3W1 1.445 3.698 -5.893 8.784 ------
390
Acquisition phase (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.001 1 0.001 2.500 Condition B2 0.000 1 0.000 0.590 Interaction B3 0.001 1 0.001 1.150 Error 0.056 97 0.001 ------Within ------Linear W1 0.003 1 0.003 10.048 B1W1 0.000 1 0.000 0.282 B2W1 0.000 1 0.000 0.608 B3W1 0.001 1 0.001 2.391 Error 0.026 97 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.003 0.002 -0.001 0.006 Condition B2 0.001 0.002 -0.002 0.005 Interaction B3 0.002 0.002 -0.002 0.005 Linear W1 -0.004 0.001 -0.006 -0.001 B1W1 0.001 0.003 -0.004 0.006 B2W1 -0.002 0.003 -0.007 0.003 B3W1 -0.004 0.003 -0.009 0.001 ------
391
Extinction phase (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 18735.647 1 18735.647 5.288 Condition B2 221842.583 1 221842.583 62.616 Interaction B3 4456.824 1 4456.824 1.258 Error 343661.818 97 3542.905 ------Within ------Linear W1 69389.264 1 69389.264 86.151 B1W1 1229.105 1 1229.105 1.526 B2W1 3986.889 1 3986.889 4.950 B3W1 4161.712 1 4161.712 5.167 Error 78127.406 97 805.437 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 9.654 4.198 1.322 17.987 Condition B2 -33.221 4.198 -41.553 -24.888 Interaction B3 4.709 4.198 -3.624 13.041 Linear W1 -20.353 2.193 -24.705 -16.001 B1W1 5.418 4.386 -3.287 14.122 B2W1 9.757 4.386 1.053 18.461 B3W1 -9.969 4.386 -18.673 -1.265 ------
392
Extinction phase (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.001 1 0.001 2.799 Condition B2 0.000 1 0.000 0.210 Interaction B3 0.000 1 0.000 0.572 Error 0.046 97 0.000 ------Within ------Linear W1 0.000 1 0.000 0.650 B1W1 0.000 1 0.000 0.198 B2W1 0.000 1 0.000 2.101 B3W1 0.000 1 0.000 0.101 Error 0.012 97 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.003 0.002 0.000 0.006 Condition B2 0.001 0.002 -0.002 0.004 Interaction B3 0.001 0.002 -0.002 0.004 Linear W1 0.001 0.001 -0.001 0.002 B1W1 -0.001 0.002 -0.004 0.003 B2W1 0.002 0.002 -0.001 0.006 B3W1 0.001 0.002 -0.003 0.004 ------
393
Comparison of last acquisition trial to first extinction trial (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 1199.271 1 1199.271 1.874 Condition B2 21293.556 1 21293.556 33.277 Interaction B3 1366.633 1 1366.633 2.136 Error 62069.588 97 639.893 ------Within ------ACQ vs EXT W1 15374.351 1 15374.351 44.680 B1W1 581.701 1 581.701 1.691 B2W1 12118.595 1 12118.595 35.219 B3W1 1046.622 1 1046.622 3.042 Error 33377.440 97 344.097 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 4.885 3.568 -2.197 11.967 Condition B2 -20.585 3.568 -27.667 -13.502 Interaction B3 5.215 3.568 -1.867 12.297 ACQ vs EXT W1 17.491 2.617 12.298 22.684 B1W1 -6.805 5.233 -17.191 3.582 B2W1 31.058 5.233 20.671 41.445 B3W1 -9.127 5.233 -19.514 1.260 ------
394
Comparison of last acquisition trial to first extinction trial (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.000 1 0.000 2.065 Condition B2 0.000 1 0.000 0.064 Interaction B3 8.10399e-6 1 8.10399e-6 0.050 Error 0.016 97 0.000 ------Within ------ACQ vs EXT W1 5.73734e-6 1 5.73734e-6 0.089 B1W1 1.49128e-6 1 1.49128e-6 0.023 B2W1 2.73449e-6 1 2.73449e-6 0.042 B3W1 8.19771e-6 1 8.19771e-6 0.127 Error 0.006 97 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.003 0.002 -0.001 0.006 Condition B2 0.000 0.002 -0.004 0.003 Interaction B3 0.000 0.002 -0.003 0.004 ACQ vs EXT W1 0.000 0.001 -0.002 0.003 B1W1 0.000 0.002 -0.004 0.005 B2W1 0.000 0.002 -0.005 0.004 B3W1 0.001 0.002 -0.004 0.005 ------
395
Test phase (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 15535.951 1 15535.951 9.521 Condition B2 6808.754 1 6808.754 4.173 Interaction B3 136.963 1 136.963 0.084 Error 158282.510 97 1631.778 ------Within ------Linear W1 23544.967 1 23544.967 73.281 B1W1 211.836 1 211.836 0.659 B2W1 3198.061 1 3198.061 9.954 B3W1 372.198 1 372.198 1.158 Error 31165.841 97 321.297 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 14.356 4.653 5.122 23.590 Condition B2 9.504 4.653 0.270 18.738 Interaction B3 1.348 4.653 -7.886 10.582 Linear W1 -21.645 2.529 -26.664 -16.627 B1W1 4.106 5.057 -5.931 14.143 B2W1 -15.955 5.057 -25.992 -5.918 B3W1 5.443 5.057 -4.594 15.480 ------
396
Test phase (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.000 1 0.000 0.427 Condition B2 3.15047e-6 1 3.15047e-6 0.016 Interaction B3 0.000 1 0.000 0.326 Error 0.020 97 0.000 ------Within ------Linear W1 0.002 1 0.002 11.854 B1W1 0.000 1 0.000 0.166 B2W1 0.000 1 0.000 2.881 B3W1 0.000 1 0.000 0.424 Error 0.013 97 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.001 0.002 -0.002 0.004 Condition B2 0.000 0.002 -0.003 0.003 Interaction B3 0.001 0.002 -0.002 0.004 Linear W1 -0.006 0.002 -0.009 -0.002 B1W1 0.001 0.003 -0.005 0.008 B2W1 -0.006 0.003 -0.012 0.001 B3W1 0.002 0.003 -0.004 0.009 ------
397
Comparison of last extinction trial to first test trial (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 7144.475 1 7144.475 8.739 Condition B2 35.760 1 35.760 0.044 Interaction B3 213.704 1 213.704 0.261 Error 79303.367 97 817.560 ------Within ------EXT vs TEST W1 27629.214 1 27629.214 47.225 B1W1 179.077 1 179.077 0.306 B2W1 17521.276 1 17521.276 29.948 B3W1 6.698 1 6.698 0.011 Error 56750.350 97 585.055 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 11.923 4.033 3.918 19.929 Condition B2 -0.844 4.033 -8.849 7.162 Interaction B3 -2.062 4.033 -10.067 5.943 EXT vs TEST W1 -23.448 3.412 -30.220 -16.676 B1W1 3.775 6.824 -9.769 17.319 B2W1 -37.345 6.824 -50.889 -23.801 B3W1 -0.730 6.824 -14.274 12.814 ------
398
Comparison of last extinction trial to first test trial (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.000 1 0.000 0.040 Condition B2 0.001 1 0.001 2.237 Interaction B3 0.000 1 0.000 0.486 Error 0.028 97 0.000 ------Within ------EXT vs TEST W1 0.001 1 0.001 3.552 B1W1 0.000 1 0.000 0.190 B2W1 2.49422e-6 1 2.49422e-6 0.011 B3W1 5.42805e-6 1 5.42805e-6 0.023 Error 0.023 97 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.000 0.002 -0.004 0.005 Condition B2 0.004 0.002 -0.001 0.008 Interaction B3 0.002 0.002 -0.003 0.006 EXT vs TEST W1 -0.004 0.002 -0.008 0.000 B1W1 0.002 0.004 -0.007 0.010 B2W1 0.000 0.004 -0.008 0.009 B3W1 0.001 0.004 -0.008 0.009 ------
399
Exploratory analyses
Comparison of last acquisition trial to first test trial (Expectancy data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 1666.886 1 1666.886 3.538 Condition B2 2049.796 1 2049.796 4.351 Interaction B3 13.742 1 13.742 0.029 Error 45697.586 97 471.109 ------Within ------ACQ vs TEST W1 23365.562 1 23365.562 46.485 B1W1 919.030 1 919.030 1.828 B2W1 6579.367 1 6579.367 13.089 B3W1 69.281 1 69.281 0.138 Error 48756.743 97 502.647 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 5.759 3.062 -0.318 11.836 Condition B2 6.387 3.062 0.310 12.463 Interaction B3 -0.523 3.062 -6.600 5.554 EXT vs TEST W1 21.563 3.163 15.286 27.840 B1W1 -8.553 6.325 -21.107 4.001 B2W1 -22.884 6.325 -35.438 -10.330 B3W1 2.348 6.325 -10.206 14.902 ------
400
Follow-up of exploratory analyses (Expectancy data)
AAC condition
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 996.721 1 996.721 1.998 Error 24438.103 49 498.737 ------Within ------ACQ vs TEST W1 27510.818 1 27510.818 51.310 B1W1 750.293 1 750.293 1.399 Error 26272.291 49 536.169 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 6.282 4.444 -2.648 15.213 ACQ vs TEST W1 33.005 4.608 23.746 42.264 B1W1 -10.901 9.215 -29.420 7.618 ------
401
Follow-up of exploratory analyses (Expectancy data)
ABC condition
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 685.487 1 685.487 1.548 Error 21259.483 48 442.906 ------Within ------ACQ vs TEST W1 2560.662 1 2560.662 5.467 B1W1 240.603 1 240.603 0.514 Error 22484.452 48 468.426 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 5.236 4.209 -3.227 13.699 ACQ vs TEST W1 10.121 4.329 1.417 18.824 B1W1 -6.205 8.657 -23.611 11.202 ------
402
Exploratory analyses
Comparison of last acquisition trial to first test trial (Skin conductance data)
Analysis of Variance Summary Table
Source SS df MS F ------Between ------Anxiety B1 0.000 1 0.000 0.269 Condition B2 0.000 1 0.000 0.366 Interaction B3 0.000 1 0.000 0.237 Error 0.024 97 0.000 ------Within ------ACQ vs TEST W1 0.002 1 0.002 20.660 B1W1 0.000 1 0.000 1.150 B2W1 0.000 1 0.000 1.811 B3W1 3.54911e-6 1 3.54911e-6 0.031 Error 0.011 97 0.000 ------
Confidence intervals ------Contrast Value SE ..CI limits.. Lower Upper ------Anxiety B1 0.002 0.004 -0.008 0.013 Condition B2 0.003 0.004 -0.008 0.013 Interaction B3 0.002 0.004 -0.009 0.013 ACQ vs TEST W1 -0.007 0.001 -0.010 -0.004 B1W1 0.006 0.006 -0.008 0.021 B2W1 -0.008 0.006 -0.023 0.007 B3W1 -0.001 0.006 -0.016 0.014 ------