<<

J. Behav. Ther. & Exp. Psychiat. 42 (2011) 298e308

Contents lists available at ScienceDirect Journal of Therapy and Experimental Psychiatry

journal homepage: www.elsevier.com/locate/jbtep

Modifying cognitive promotes cognitive well being: A new approach to modification

Kathryn J. Lester a,*,1, Andrew Mathews b, Phil S. Davison c, Jennifer L. Burgess d,2, Jenny Yiend a,**,3 a Department of Psychiatry, University of Oxford, UK b Department of , University of California, Davis, USA c Oxfordshire Mental Healthcare NHS Trust, Warneford Hospital, Oxford, UK d Department of , University of Oxford, UK article info abstract

Article history: Background: Cognitive Bias Modification (CBM) procedures have been used to train individuals to Received 14 July 2010 interpret ambiguous in a negative or benign direction and have provided evidence that Received in revised form negative causally contribute to emotional . 29 December 2010 Method: Here we present the development and validation of a new form of CBM designed to manipulate Accepted 3 January 2011 the cognitive errors known to characterize both depression and anxiety. Our manipulation was designed to modify the biased cognitions identified by Beck’s cognitive categories (e.g. arbitrary , Keywords: overgeneralisation) and typically targeted during therapy. Cognitive bias modification Cognitive errors Results: In a later test of spontaneous , unselected (Experiment 1) and vulnerable participants Anxiety (Experiment 2) who had generated positive alternatives rather than errors perceived novel hypothetical Depression events, their causes and outcomes in a non-distorted manner. These groups were also less vulnerable to Interpretation two different types of emotional stressor (video clips; and an imagined social situation). Furthermore participants’ interpretation of their own performance on a problem-solving task was improved by the manipulation, despite actual performance showing no significant change. Conclusions: These findings demonstrate that Cognitive Error Modification can promote positive infer- ences, reduce vulnerability to stress and improve self-perceptions of performance. Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction attributional style (e.g. Ahrens & Haaga, 1993; Cropley & MacLeod, 2003; Reardon & Williams, 2007), judge future negative events and Cognitive models propose that inferential biases in how people outcomes as more likely to occur than positive outcomes (e.g. interpret events, attribute the causes of events and predict future Butler & Mathews, 1983, 1987; MacLeod & Cropley, 1995; MacLeod, events are important for the aetiology and maintenance of Tata, Kentish, & Jacobsen, 1997) and systematically interpret psychological disorders (Beck, Rush, Shaw, & Emery, 1979). There is ambiguous cues as threatening (e.g. Eysenck, Mogg, May, Richards, compelling evidence that high trait anxious, dysphoric and clini- & Mathews, 1991; Lawson, MacLeod, & Hammond, 2002; Mogg, cally anxious and depressed individuals demonstrate a pessimistic Bradbury, & Bradley, 2006; Richards & French, 1992). Recently, experimental techniques known as cognitive bias modification for interpretation (CBM-I) have focused on this latter * Corresponding author. Social, Genetic and Developmental Psychiatry Centre, form of inferential bias, the interpretation of emotional ambiguity. PO80 Institute of Psychiatry, De Crespigny Park, London, SE5 8AF, UK þ44 207 848 CBM-I has been developed to simulate interpretation biases in the 5414. laboratory and has produced evidence suggesting biased interpre- ** Corresponding author. PO63 Institute of Psychiatry, De Crespigny Park, London tations are causally implicated in anxiety (Mackintosh, Mathews, SE5 8AF, UK. E-mail addresses: [email protected] (K.J. Lester), [email protected] (J. Yiend, Ridgeway, & Cook, 2006; Murphy, Hirsch, Mathews, Smith, Yiend). & Clark, 2007; Wilson, MacLeod, Mathews, & Rutherford, 2006) 1 Kathryn Lester is now at the Social, Genetic and Developmental Psychiatry and depressive vulnerability (Holmes, Lang, & Shah, 2009). Several ’ Centre, Institute of Psychiatry, part of King s College London. studies have now demonstrated that CBM-I leads to congruent and 2 Jennifer Burgess is now at the Department of Psychiatry, University of Oxford. 3 Jenny Yiend is now at the Institute of Psychiatry, part of King’s College London spontaneous biases in interpretation of novel material (e.g. and King’s Health Partners. Mathews & Mackintosh, 2000). Furthermore, modifying biases in

0005-7916/$ e see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.jbtep.2011.01.001 K.J. Lester et al. / J. Behav. Ther. & Exp. Psychiat. 42 (2011) 298e308 299 a positive direction has also led to improvements in trait and social personalization, minimization and overgeneralisation (see Table 1). anxiety (Beard & Amir, 2008; Mathews, Ridgeway, Cook, & Yiend, We suggest that previous CBM versions have omitted important 2007; Salemink, van den Hout, & Kindt, 2009) and reduced antic- types of cognitive error categories, which are nevertheless ubiq- ipated anxiety when imagining performing a stressful speech uitous in clinical settings. If CBM techniques are to make their way (Murphy et al., 2007). These modification techniques have focused effectively into clinical settings, then translational researchers must solely on manipulating the interpretation of emotional ambiguity start to address this shortfall. It cannot be assumed that as yet and have largely ignored the potential of the technique to manip- untested forms of inferential reasoning will be amenable to bias ulate a wider range of inferential biases of the sort outlined above. modification in the same manner as previously found only for the In this paper, we validate a new form of modification designed to interpretation of ambiguity. First, we demonstrate its efficacy in manipulate a broader range of inferential biases including, but modifying inferences in a positive and negative direction and extending beyond, the interpretation of emotional ambiguity and response to stress in a healthy sample. Second, we report further which are characterized by cognitive errors. refinement to maximize translational relevance and we provide A limitation of previous CBM work is that no studies have evidence of efficacy in changing cognition, response to stress and developed a procedure whose content is designed specifically for performance evaluation in a vulnerable sample. use in anxiety and depression as it presents clinically. It is widely recognized that anxiety and depression co-occur (Carter, Wittchen, 2. Experiment 1 Pfister, & Kessler, 2001; Judd et al., 1998). Das-Munshi et al. (2008) have argued that subsyndromal mixed anxiety and depression Prior to conducting Experiment 1 an extensive phase of mate- (MADD) is the most prevalent psychological disorder, with a 1- rials development was undertaken to assemble new modification month point prevalence of 8.8% and an impact on health-related and test materials designed to capture 7 categories of cognitive quality of life similar to that of pure cases of anxiety and depression. error. Table 1 gives the range of errors targeted and provides A transdiagnostic approach would argue that such high rates of co- examples of both source and modification materials. Further details occurrence exist because different disorders (such as anxiety and of item formation and selection, including validity ratings, are depression) share maintaining cognitive processes, including provided in the Supplementary material. systematic biases and distortions in inference (Harvey, Watkins, Mansell, & Shafran, 2004). Therefore, targeting the negative dis- 2.1. Method torted inferential biases common across disorders offers an alter- native approach to treatment, which bypasses traditional 2.1.1. Participants diagnostic labels and has potentially broad application. Sixty undergraduate students, 23 males and 37 females (mean In the present study we apply this approach to CBM. As yet, little age ¼ 20.83, SD ¼ 3.55, range 18e42) took part and were awarded research has explicitly focused on developing CBM procedures in £10 for participation. All were fluent English speakers and had no such a way as to ensure that they are relevant across clinical current or previous history of psychological disorder. Half were conditions. The content used to date has been limited to experi- counterbalanced by participant number to generate cognitive mentally derived items concerning social threat, physical threat or errors during bias modification (error condition) and half to test anxiety situations, which have disorder specific relevance. By generating responses, which were not cognitive errors (non-error using clinically derived content of relevance to depression and condition). anxiety and formulating this content to allow a wide range of inferential biases, we adopt a functional approach, which highlights 2.1.2. Cognitive error modification procedure the transdiagnostic and real world clinical potential of CBM. Modification items were designed to allow a cognitive error to According to clinical theory and practice one manifestation of the occur (or not occur) in how described events were perceived, in inferential biases described above are cognitive errors (Harvey et al., inferring their cause or in forming an expectancy for a future 2004). Cognitive errors are systematic distortions of involving outcome (see Supplementary materials for details of development the drawing of erroneous conclusions and biases (Beck, Emery, & and selection procedure). Participants read 101, 3-line emotionally Greenberg, 1985; Beck et al., 1979). Beck identified seven categories ambiguous descriptions ending with a word fragment similar to of cognitive error e arbitrary inference, selective abstraction, over- previous CBM-I procedures (Mathews & Mackintosh, 2000). generalisation, catastrophising, minimization, personalization and Completion of the fragment resolved the ambiguity of the dichotomous thinking, which he proposed were the product of description to give an error or non-error meaning depending upon a biased information processing system (see Table 1 for definitions). assigned condition. A subsequent question (yes/no response There is evidence that cognitive errors play a prominent role in required) reinforced this by using feedback (correct/incorrect a range of disorders, particularly anxiety and depression (Blackburn according to group assignment). Twenty items were fixed valence & Eunson, 1989; Henriques & Leitenberg, 2002; Johnson, Johnson, & probes (10 negative and 10 benign trials) that had an identical Petzel, 1992; Lefebvre, 1981; Leitenberg, Yost, & Carroll-Wilson, outcome irrespective of group assignment. These were included to 1986; Weems, Berman, Silverman, & Saavedra, 2001). Reducing the test for development of modification effects during the modifica- cognitive errors associated with anxious and depressed mood is an tion procedure itself (speed of solution for items that were important therapeutic target within CBT and is considered a primary congruent or incongruent with assigned modification condition). mechanism of change (Ilardi & Craighead, 1999). This is based on the Table 1 provides a clinical example (column 3), indicates the premise that reducing erroneous processing will ultimately modify underlying inferential processes (column 5) and provides an underlying depressogenic and anxiogenic schemas, leading to cor- example of how each clinical exemplar was reworked into a modi- responding reductions in exaggerated emotional reactions. fication item (column 4). Our new ‘Cognitive Error Modification’ does not exclude examples of the type of biased interpretation of ambiguity used in 2.1.3. Cognitive effects of modification: Similarity Rating test previous procedures. Instead it was designed to manipulate a range Details of the development and selection of test items are given of inferential biases using examples of cognitive errors typically in the Supplementary materials. Appendix A gives one example of targeted in cognitive therapies. In doing so, it adds a significant a test item. There were 14 test items, corresponding to 2 items for proportion of previously untapped inferential biases, for example each cognitive error category. All were novel, 3-line descriptions Table 1 300 Definitions and examples of cognitive errors and associated modification items.

Cognitive Error Definition a Clinical Example b Example Modification Item Putative Underlying Inferential Biases c Arbitrary Drawing a specific conclusion in the absence of evidence You have invited a friend for dinner and Your friend comes round for dinner and Interpretive processing Inference to support the conclusion or when the evidence is you spend the time chatting, eating and you cook a special meal. You chat for a Expectancy reasoning contrary to the conclusion drinking. At 9.30pm your friend says good while and at 9.30 they get up to they really should leave. Despite having leave. You think they decided to leave had an enjoyable evening, you think because they were. that your friend has decided to leave .bored/sleepy because they are bored. Is your friend exhausted? Selective Focusing on a detail taken out of context, while ignoring A recent graduate begins a new position You have started a new job and hope to Interpretive processing Abstraction other more salient features of the situation and and is eager to make friends with their be friends with your colleagues. At the Expectancy reasoning conceptualizing the whole experience on the basis colleagues. They ask their new end of your first day you ask whether of this fragment colleagues whether they would like to people would like to go for a drink and 2 join them for a drink after work and 2 people offer to come out with you. You people accept their offer. They focus on think this means you have probably . the fact that some people declined and been rejected/accepted Have you 298 (2011) 42 Psychiat. Exp. & Ther. Behav. J. / al. et Lester K.J. think this means they aren’t liked failed to make friends? rather than being pleased that some of their colleagues are keen to socialize. Minimization Errors in evaluating the significance or magnitude of an You have a positive experience, such as You have to take an exam before a new Interpretive processing event, transforming neutral or positive experiences passing an exam but you tell yourself company will employ you. You have Attributional reasoning into negative ones and rejecting positive experiences that it was a fluke or it doesn’t revised hard but aren’t sure how well as not good enough really count. you will do. When the results are given you are told you have passed and you think this is a(n). fluke/achievement Was passing the exam down to luck? Dichotomous Tendency to place all experiences in one of two You’ve been trying to diet but you’ve You have been on a really strict diet for Expectancy reasoning Thinking opposite categories, e.g. flawless or defective eaten a few sweets over the weekend. a few weeks and have totally cut out rather than viewing them as existing on You tell yourself that you can never sweet things. However you couldn’t a continuum. In describing oneself, control yourself and that all your resist a piece of cake on your friend’s the extreme negative categorization is selected dieting and jogging over the whole birthday. You think your attempts at week have gone down the drain. dieting have been. futile/disciplined Have you completely failed in your attemptsYou have to diet? spent ages plucking up the Attributional reasoning Overgeneralisation Drawing a general rule or conclusion on the A shy young man summons up the courage to ask the person you fancy Expectancy reasoning she politely declines due to a prior out on a date. When you finally ask basis of one or more isolated incidents and courage to ask a girl for a date. When e

“ ’ 308 applying the concept across the board to related engagement, he says to himself I m they politely turn you down because and unrelated situations never going to get a date. No girl they already have a prior engagement. would ever want a date with me. I’ll You think you will spend the rest of be lonely and miserable forever” your life being. unwanted/courageous Are you pleased you were brave enough to ask the person out on a date? Personalization Tendency to relate external events to oneself when A young lady finds out that a close You are dismayed to hear a close friend Interpretive processing there is no basis for making such a connection friend has been ill. She reflects on has been ill. When you met last week Attributional reasoning when they met last week when you had a slight disagreement and they had a small falling out and haven’t spoken since. You wonder if your the fact that they haven’t spoken friend’s illness could have been caused since. She begins to think that by . the arguing/germs the argument could be the cause Could you have caused your friend’s illness? of her friend’s illness K.J. Lester et al. / J. Behav. Ther. & Exp. Psychiat. 42 (2011) 298e308 301

with the potential to elicit cognitive errors. They were presented with an identifying title in a fixed random order. Unlike modifica-

c tion items, a final word fragment preserved the meaning of the description in such a way that it was possible to infer a cognitive error or not. Descriptions were followed by a neutral comprehen- sion question (yes/no response, with feedback) about the factual content of the passage. Biases were assessed for each description by Inferential Biases Interpretive processing Attributional reasoning Expectancy reasoning presenting two disambiguating sentences, one reflecting a cogni- tive error (error target) and the other a non-distorted (non-error target) meaning of the original description. Two foil sentences were also presented reflecting positive or negative statements relevant to the passage, but which could not be directly inferred from it. Sentences were presented individually alongside their identifying description title and in random order within each title. Participants . rated each sentence for similarity in meaning to the original description using a 1 (very different) to 4 (very similar) scale. Generalization of Cognitive Error Modification effects would be indicated by higher similarity ratings to possible meanings cation Item Putative Underlying

fi congruent with assigned modification group.

2.1.4. Filler task Participants read three, 250-word emotionally neutral para- graphs describing a domestic task over a 10 min period and answered Example Modi You are watching your localforecast. weather You notice there isbe due a to storm in yourmay area be and some there damage toYou property. imagine that the amountdamage of to your house willextensive/tiny be Are you expecting your house to be badly damaged? 2 questions about each. This task was administered to allow any emotional differences, which occurred across the modification phase to dissipate prior to completion of the Similarity Rating test.

2.1.5. Stress challenge task: video stressor Participants viewed 4 potentially stress provoking video clips of the side of

“ life-threatening accidents (duration 6 min 19 s) taken from a real- it will cost a life documentary. Each clip provided the opportunity for partici- . pants to make cognitive errors in how the clips were perceived, in b

Initially he imagines inferring the possible causes of the events and in forming an ”

x. expectancy for the outcome of the event. fi

2.1.6. Self report measurements of mood State and trait anxiety were measured using short-form versions Clinical Example fortune to the repair bill will bethousand several pounds but after theshock intial he realizes the damageand is the minor repair costs will be minimal the house is wrecked A man hears that hisdamaged house as has a been result ofsequence a of storm. thoughts His are of the Spielberger State-Trait Anxiety Inventory (STAI, Spielberger et al., 1979). The Beck Depression Inventory-II (BDI-II, Beck, Steer, & Brown, 1996) was administered as a measure of depressive symp- tomatology and positive and negative affect (using both a ‘right now’ and ‘general’ time frame) was measured using the Positive and Negative Affect Scale (PANAS, Watson, Clark, & Tellegen, 1988).

2.1.7. Procedure After giving written , participants completed pen and paper versions of the mood measures (Time 1). Participants

cance or magnitude of fi fi continued to complete their assigned Cognitive Error Modi cation phase, which was presented using E-Prime 1.1. State mood measures were repeated on completion of the modification phase (Time 2). A pen and paper version of the filler task was then and online sources. administered. State mood was assessed again immediately prior to the Similarity Rating test (Time 3), which was administered using a E-Prime 1.1. Participants then watched, alone, the series of accident videos, with state mood assessed pre and post viewing (Times 4 nition Burns (2000) fi and 5). Finally, participants were debriefed and paid, with the . Errors in evaluating the signi an event, anticipation of extremeand adverse considering outcomes the most unfavorableoutcomes of of all a possible situation and

. experiment lasting 1 h 20 min.

) 2.2. Results cation) fi

Beck et al. (1979) Beck (1963, 1976) Harvey et al. (2004) 2.2.1. Participants continued ( As shown in Table 2, modification groups were comparable at (Magni From From From baseline on all measures of trait and state mood, age and gender c Catastrophising Cognitive Error De a b

Table 1 ratios. 302 K.J. Lester et al. / J. Behav. Ther. & Exp. Psychiat. 42 (2011) 298e308

Table 2 Participant characteristics (and SE) for Experiments 1 and 2. All group comparisons were non-significant.

Experiment 1 Experiment 2

Non-Error Modification Error Modification Control Group Non-Error Modification (n ¼ 30) (n ¼ 30) (n ¼ 35) (n ¼ 35) Age 21.50 (0.85) 20.17 (0.31) 20.03 (0.27) 20.80 (0.73) Gender (f:m) 21:9 16:14 21:14 21:14 State Anxiety (STAI) 14.23 (0.65) 13.73 (0.61) 18.57 (0.82) 17.40 (0.74) Trait anxiety (STAI) Baseline 16.90 (0.87) 16.83 (0.83) 47.71 (1.53) 45.00 (1.32) BDI-II Baseline 6.67 (1.17) 5.50 (0.87) 12.63 (1.38) 11.51 (1.28) Negative Affect (PANAS) Baseline 16.17 (0.90) 14.90 (0.62) 19.40 (1.06) 18.06 (0.81) Positive Affect (PANAS) Baseline 34.43 (0.93) 34.20 (1.17) 30.09 (1.14) 30.26 (1.09) CCL-D Baseline 19.60 (1.23) 18.34 (1.47) CCL-A Baseline 17.29 (1.11) 15.57 (0.78) Trait anxiety (STAI) Time 2 48.03 (1.74) 43.80 (1.74) BDI-II Time 2 13.77 (1.55) 10.80 (1.19) Negative Affect Time 2 19.91 (0.98) 17.31 (1.00) Positive Affect Time 2 31.11 (1.23) 31.20 (1.02) CCL-D Time 2 15.17 (1.70) 12.14 (1.54) CCL-A Time 2 10.63 (1.09) 9.09 (0.85)

2.2.2. Fixed probe latencies 2.2.5. Stress challenge task: video stressor Trials in which participants were unable to identify or incor- A MANOVA on state mood change scores (post stressepre stress, rectly completed the word fragment (1% of probe trials) and see Table 3) revealed a significant effect of modification group, outliers defined as reaction times, which exceeded 2.5 standard V ¼ 0.15, F(3, 56) ¼ 3.20, p < .05. This was followed up with deviations above or below individual means (4% of probe trials) discriminant analysis and univariate ANOVAs. Correlations were excluded. Mean latencies to complete word fragments on between each dependent variable and the discriminant function probe trials with fixed error and non-error meanings were entered revealed change in negative affect loaded most highly (r ¼ 0.55), into a mixed ANOVA with Modification Group (error, non-error) as followed by change in positive affect (r ¼ 0.45), while change in the between participants factor and Probe Type (error, non-error) state anxiety had a low loading (r ¼ 0.06). This indicates that as the within participants factor. Training effects were demon- change in negative and positive affect (but not state anxiety) strated by a significant Modification Group Probe Type interac- contributed to discriminating between groups. Separate univariate tion, F(1, 58) ¼ 16.72, p < .001. Non-error probe fragments were ANOVAs on each dependent variable revealed non-significant identified significantly faster by the non-error than error modifi- effects of group on state anxiety and positive affect, change scores cation group (t(58) ¼2.45, p < .05, d ¼ 0.63, 1334.24 vs. (F(1, 58) < 1, n.s; F(1, 58) ¼ 1.98, n.s) and a near significant statistical 1590.70 ms). Groups did not differ in speed of identification of error trend in the hypothesized direction for negative affect, F(1, probe fragments, although the direction of means followed the 58) ¼ 3.00, p ¼ 0.09, d ¼ 0.45. Thus the pattern of data revealed predicted pattern, (t(58) < 1, n.s, 1712.08 vs. 1602.06). a significant difference between groups in response to the stress challenge, most clearly carried by a larger increase in negative 2.2.3. Cognitive effects of modification: Similarity Rating test affect following error than non-error modification (4.30 vs. 2.53). Mean recognition ratings for each condition were entered into a mixed ANOVA with Modification Group (error, non-error) as the 3. Experiment 2 between participants’ factor and Sentence Type (target, foil) and Sentence Valence (error/negative, non-error/positive) as the within Experiment 1 demonstrated that Cognitive Error Modificationwas participants factors. Consistent with the presence of modification effective in inducing group differences in inferential bias and effects there was a significant Group Valence Sentence Type emotional response to a stressor in a healthy sample who had been interaction, F(1, 58) ¼ 36.52, p < .001, d ¼ 1.58. Separate follow up trained in an error or non-error direction. Experiment 2 was designed analyses were conducted for targets and foils. Significant to allow further development and testing of the non-error form of the Group Valence interactions were present for targets, F(1, new materials to ensure maximum relevance to clinical and 58) ¼ 59.13, p < .001, d ¼ 2.02 and foils F(1, 58) ¼ 13.46, p < .001, subclinical populations. After this development, Cognitive Error d ¼ 0.96. Non-error targets were rated as significantly more similar to modification was piloted in a selected vulnerable sample. We the previous passage by the non-error modification group compared compared a non-error modification condition compared to a no- to the error modification group (t(58) ¼ 3.80, p < .001, d ¼ 1.00, 2.79 modification control condition and explored effects on cognition, vs. 2.46). Conversely, similarity ratings for error targets were signif- response to stress and performance and its evaluation. Details of the icantly higher for those who received error modification, compared further materials development are given in the Supplementary to non-error modification (t(58) ¼6.18, p < .001, d ¼ 1.62, 2.64 vs. material. In brief, it involved clinicians generating more exemplars 2.07). The two groups did not differ in their similarity ratings of from their therapeutic practice (allowing sufficient material for two positive (t(58) ¼ 1.39, n.s) or negative foils (t(58) ¼1.13, n.s)(Fig.1). modification sessions); clinician ratings of these materials and grading of items to enhance ‘believability’. ‘Believability’ of individual 2.2.4. State mood effects of modification item solutions has been highlighted as an important factor, especially A MANOVA on dependent measures of mood change across bias in vulnerable groups. Mathews et al. (2007) reported that initial modification (see Table 3) showed no significant effect of modifi- attempts at positive bias modification in high trait anxious individuals cation group, V ¼ 0.08, F(3, 56) ¼ 1.55, n.s. Neither did the groups were unsuccessful because some struggled to associate themselves differ on a composite of state mood scores prior to completion of with the very positive and, for them, unrealistic interpretations the Similarity Rating test, V ¼ 0.06, F(3, 56) ¼ 1.09, n.s. Mood state offered. We therefore adopted a similar strategy to Mathews et al. effects induced by modification were therefore unable to account (2007) by using graded items: session 1 required primarily non- for subsequent effects of the modification procedure. negative inferences, while session 2 progressed to overtly positive K.J. Lester et al. / J. Behav. Ther. & Exp. Psychiat. 42 (2011) 298e308 303

completed the Cognition Checklist (Steer, Beck, Clark, & Beck, 1994). Only those scoring at least half a standard deviation above the student norms for the depression or anxiety subscale were invited to participate. Invited participants were also screened using the Mini International Neuropsychiatric Interview (Sheehan et al., 1998)by a trained researcher. The following exclusion criteria were applied: moderate to high suicidality, current or past mania or hypomania, OCD, current substance dependence, current or past psychosis. Participants were randomly allocated to non-error modification (n ¼ 35) or a no-modification control condition (n ¼ 35).

3.1.2. Cognitive error modification To maximize clinical relevance, additional modification items were developed from exemplars of cognitive errors made by patients during therapy and reported to us by clinicians. These items Fig. 1. Mean recognition ratings of non-error and error targets and positive and were combined with those used in Experiment 1 and based on negative foil sentences by group in Experiment 1 (error bars 1 standard error). * indicates group comparison significant at p < .001. clinician ratings; the 100 most clinically relevant items were selected for use (see Supplementary materials). Bias modification followed the same format as that used in Experiment 1, with the inferences. In addition previous work has shown that effects of bias following exceptions. Graded modification was implemented: modification can be enhanced by encouraging active processing and completion of the word fragment provided a non-negative outcome imagery of modification descriptions (Holmes,Mathews,Dalgleish,& in session 1 and an overtly positive outcome in session 2. Active Mackintosh, 2006; Hoppitt, Mathews, Yiend, & Mackintosh, 2010a, generation and mental imagery were encouraged by asking partic- 2010b) and we therefore incorporated this in our new procedure. ipants to imagine themselves in each situation described. Partici- Several variations to the test measures were implemented. We pants were instructed to imagine and report on a positive sought convergent evidence for the effects of our manipulation on continuation of each situation, prompted bya continuation sentence response to stress using a different stressor task (imagined social stem. Specifically, instructions were “Imagine the situation.” fol- interaction, Murphy et al., 2007). We added self report measures of lowed by a stem related to the previous description e.g. “when you known trait vulnerability markers, while acknowledging that with reflect on how well you are doing at work”. After 3 s imagining, only two sessions of non-error modification we were unlikely to participants were asked to report aloud a single sentence describing elicit significant change (cf. Mathews et al., 2007) compared to the their imagined continuation and rated how positive it was on a 7 non-modification control group. Finally we sought to assess effects point scale (1 ¼ not at all, 7 ¼ extremely). To ensure participants on objective and subjective performance under controlled condi- understood this aspect of the procedure, practice was given in tions. We reasoned that if modification targets inferences relating imagining situations similar to those presented during modification to ambiguous events, then this might be expected to transfer to (cf. Holmes & Mathews, 2005). To minimize length and maximize participants’ inferences about their own performance, especially efficacy, probe trials were no longer included. when that performance is ambiguous. We therefore devised and piloted an anagram task to elicit a near 50% completion success 3.1.3. Similarity Rating test, filler task rate. We hypothesized that participants interpretation of their These tasks were identical to those administered in Experiment 1. performance on this task should be positively biased by the non- error modification procedure compared to the control condition, 3.1.4. Self report measurements of mood and cognition despite actual success rates remaining unchanged. Trait anxiety was measured by the STAI-Trait, (Spielberger, Gorsuch, Lushene, Vagg, & Jacobs, 1983), depression by the BDI-II 3.1. Method (Beck et al., 1996) and trait positive and negative affect by the ‘general’ version of the PANAS (Watson et al., 1988). These measures 3.1.1. Participants were administered at the beginning of session 1 and the end of Seventy university students, 28 males and 42 females (M session 2. Frequency of anxiety and depression-related negative age ¼ 20.41 yrs, SD ¼ 3.25, range 18e40) participated and were cognitive distortions was measured by the Cognition Checklist (Steer awarded £20. All were fluent English speakers and had not received et al., 1994). State anxiety was measured using the short-form STAI- psychological or pharmacological treatment for a psychiatric state (Spielberger et al., 1979) before (start of session 1), prior to the disorder within the past 6 months. Prospective participants Similarity Rating test and after (end of session 2) the experiment.

Table 3 Mean (and SE) state mood across the experimental session by modification group for Experiment 1.

Time 1 Time 2 Mean change Time 3 Time 4 Time 5 Mean change across across Stress Modification Task Non-error State anxiety 14.23 (0.65) 13.97 (0.65) 0.26 (0.68) 13.03 (0.55) 13.67 (0.66) 18.97 (0.94) 5.30 (0.87) Negative affect 10.93 (0.34) 10.43 (0.12) 0.50 (0.34) 10.30 (0.12) 10.47 (0.18) 13.00 (0.57) 2.53 (0.48) Positive affect 30.50 (1.38) 28.63 (1.63) 1.87 (0.63) 27.80 (1.79) 27.80 (1.73) 27.30 (1.52) 0.50 (0.81) Error State anxiety 13.73 (0.61) 15.73 (0.90) 2.00 (0.93) 13.87 (0.60) 14.47 (0.72) 20.00 (1.30) 5.53 (1.00) Negative affect 11.23 (0.42) 12.20 (0.57) 0.97 (0.64) 10.73 (0.23) 10.67 (0.23) 14.97 (0.97) 4.30 (0.90) Positive affect 28.53 (1.44) 25.83 (1.38) 2.70 (0.84) 26.03 (1.52) 25.70 (1.48) 26.80 (1.60) 1.10 (0.80)

Time 1 ¼ Pre modification, Time 2 ¼ Post modification, Time 3 ¼ Pre Similarity Rating test, Time 4 ¼ Pre Stress challenge, Time 5 ¼ Post stress challenge. 304 K.J. Lester et al. / J. Behav. Ther. & Exp. Psychiat. 42 (2011) 298e308

3.1.5. Stress challenge task: imagined social interaction (n ¼ 13) met criteria for past major depressive episode, 10% (n ¼ 7) This task was administered in session 2 only. In an adaptation of for GAD, 4.3% (n ¼ 3) for Panic Disorder Lifetime and 1.4% (n ¼ 1) for Murphy et al. (2007) participants spent 15 s imagining that they Specific Phobia.4 were leading a discussion group of fifteen strangers for 30 min on a topic which they knew only a little about. After imagining this 3.2.2. Cognitive effects of modification: Similarity Rating test scenario, rating scales ranging from 3 (e.g. extremely relaxed) to As in Experiment 1, mean recognition ratings were entered into þ3 (e.g. extremely anxious) were used to measure anticipated a mixed ANOVA. Targets were endorsed more highly than foils (F(1, feelings (anxiety and hopelessness, e.g. How anxious do you 68) ¼ 272.75, p < .001, 2.44 vs. 1.60) and non-error sentences anticipate feeling in this situation?), ability to cope with feelings endorsed as more similar than error sentences (F(1, 68) ¼ 49.21, (e.g. How well do you think you would cope with any feelings of p < .001, 2.19 vs. 1.85). These effects were qualified by significant hopelessness you felt in this situation?) and perceived impression Group Valence, F(1, 68) ¼ 51.31 p < .001 and (e.g. How well do you think you would come across in this situa- Group Valence Sentence Type interactions, F(1, 68) ¼ 30.67, tion?) (See Table 4). p < .001, d ¼ 1.34. Separate analyses for target and foil sentences revealed the Group Valence interaction remained significant for 3.1.6. Performance evaluation task: anagram completion targets, F(1, 68) ¼ 51.37, p < .001, d ¼ 1.74 and foils F(1, 68) ¼ 16.22, Two parallel anagram tasks were developed and piloted allow- p < .001, d ¼ 0.98. Non-error targets were rated as significantly ing counterbalanced repeat administration at baseline and test. more similar to the previous passage by the non-error modification Piloting ensured it was possible to solve approximately half of the group compared to the control group, t(68) ¼ 4.71, p < .001, anagrams in the time available, thus objective performance was d ¼ 1.14, 2.92 vs. 2.38. Similarity ratings for error targets were ambiguous (see also results). Each version consisted of 25 items significantly higher for those in the control group relative to non- varying in difficulty, determined by number of letters (4 letters to error modification, t(68) ¼5.01, p < .001, d ¼ 1.20, 2.50 vs. 1.96. 13 letters); word frequency (Kucera Francis norms: low frequency The two groups also differed in their similarity ratings of positive anagrams ¼ 10e20 e.g. thumb; high frequency anagrams ¼ 80e125 foils, t(68) ¼ 2.59, p < .05, d ¼ 0.63 but not negative foils, t(68) < 1, e.g. yesterday) and extent of rearrangement, (easyesmall number n.s. The groups were comparable in state anxiety prior to comple- of letters moved out of position e.g. indoww (window), hardelarge tion of the Similarity Rating test, t(68) ¼1.76, n.s. Therefore, state number of letters moved out of position e.g.rginolei ()). dependent mood effects were an unlikely explanation for the Difficulty level was matched across versions. In addition, differential effects on cognition observed. 5 anagrams in each version were impossible because one letter of the solution was substituted. Participants were instructed to solve 3.2.3. Stress challenge task: imagined social interaction as many anagrams as possible within 3 min and immediately A MANOVA with ratings of anticipated anxiety, ability to cope afterward completed a series of visual analog scales (0e100) asking with feelings of anxiety, hopelessness, ability to cope with feelings about their mood and their performance. Mood measures asked of hopelessness and performance (see Table 4) entered simulta- participants to rate how they felt right now (anxious, depressed, neously as dependent variables revealed a near significant statistical happy) while performance measures asked participants to rate how trend of modification group on ratings, V ¼ 0.14, F(5, 64) ¼ 2.08, they felt about their performance on the task (frustrated, success- p ¼ 0.08. The MANOVA was followed up with discriminant analysis ful, hopeless). Actual performance was measured by the number of and univariate ANOVAs. Correlations between each of the depen- correctly solved anagrams. dent variables and the discriminant function revealed that ratings of anticipated performance loaded most highly (r ¼ 0.93), followed by 3.1.7. Procedure ratings of being able to cope with feelings of anxiety (r ¼ 0.56). This Bias modification participants attended 2 sessions approxi- indicates that the variables of anticipated performance and coping mately 7 days apart (mean interval ¼ 7.09 days, range 5e9 days). with feelings of anxiety contributed to discriminating between the The first session comprised baseline measures of mood, cognition modification and control group to the greatest extent. This was and performance evaluation, followed by the first session of supported by separate univariate ANOVAs on the outcome variables. cognitive error bias modification. The second session comprised the There was a significant effect of modification group on anticipated second bias modification, followed by test assessments of the same performance, F(1, 68) ¼ 9.60, p < .01, d ¼ 0.75, with those in the measures and the imagined stress task. Control participants modification group reporting significantly better anticipated attended on a similar schedule to complete baseline and test performance (1.03 vs. 0.14). There was a near significant statistical assessments only. Upon completion, participants were debriefed. trend of modification group on coping with feelings of anxiety, F(1, 68) ¼ 3.49, p ¼ 0.07, d ¼ 0.45, with the modification group reporting 3.2. Results themselves as better able to cope with feelings of anxiety (1.17 vs. 0.60). Groups did not differ on other rated measures (all Fs < 2). 3.2.1. Participants As shown in Table 2, groups were comparable at baseline on all 3.2.4. Performance evaluation task: anagram completion measures of trait and state mood, age and gender ratios. This 3.2.4.1. Actual performance. The number of anagrams correctly sample with a mean trait anxiety score of 46.36 was approximately completed was analyzed using a mixed ANOVA with the variables one standard deviation above the STAI-Trait reported norms and of Time (baseline, test) and Group (non-error, control). There was equivalent to approximately the 78th percentile for college an overall non-significant statistical trend for actual performance students (Spielberger et al., 1983). Mean BDI-II scores of 12.07 (number of anagrams solved) to improve from baseline to test, F(1, indicated a mild level of depressive symptoms (Beck et al., 1996). 68) ¼ 3.22, p ¼ 0.08, (14.19 vs. 14.67) and no interaction with group Mean negative affect scores of 18.73 were in line with reported (F(1, 68) < 1, n.s, see Fig. 2). norms, however, the mean positive affect score of 30.17 was lower than the norm reported for an undergraduate sample and was more similar to that of a mixed clinical sample (Watson & Clark, 1994). ¼ MINI diagnoses revealed that 70% (n 49) of the sample did not 4 The sum of participants meeting diagnostic criteria exceeds the sample size of meet criteria for any current or past diagnostic category, 18.6% 70 because 3 individuals met criteria for two diagnoses. K.J. Lester et al. / J. Behav. Ther. & Exp. Psychiat. 42 (2011) 298e308 305

Table 4 Mean (and SE) ratings for the stressor tasks used in Experiment 2.

Imagined Social Interaction (e3toþ3; given at test only)

Control Group Non-Error Modification Group Anticipated anxiety 1.46 (0.24) 1.23 (0.20) Cope with anxiety 0.60 (0.26) 1.17 (0.16) Anticipated hopelessness 0.14 (0.22) 0.29 (0.22) Cope with hopelessness 0.66 (0.22) 0.91 (0.21) Anticipated performance 0.14 (0.25) 1.03 (0.14)

Anagram Task

Control Group Non-Error Modification Group

Baseline Test Baseline Test Actual performance (/25) Number of correct responses 14.20 (0.30) 14.66 (0.28) 14.20 (0.34) 14.68 (0.40) Performance evaluation (0e100) Successfulness 43.20 (2.84) 41.80 (2.50) 41.31 (2.27) 49.23 (2.73) Frustration 55.71 (3.88) 47.97 (2.81) 54.83 (3.20) 54.00 (3.28) Hopelessness 51.51 (3.72) 54.86 (3.65) 52.03 (3.54) 52.66 (3.81) Mood (0e100) Anxiety 46.46 (4.80) 44.77 (4.26) 46.57 (4.23) 47.71 (4.43) Depression 46.17 (4.64) 46.66 (3.49) 46.31 (4.98) 47.06 (4.93) Happy 52.00 (2.50) 52.20 (2.46) 54.17 (2.62) 58.31 (2.54)

3.2.4.2. Evaluation of performance. A MANOVA with change in (see Table 2) entered simultaneously as dependent variables ratings of performance (test-baseline: successfulness, frustration and revealed no significant effects of modification group on either hopelessness) entered simultaneously as dependent variables change in trait mood or change in cognitive distortions (V ¼ 0.06, F revealed a near significant statistical trend effect of modification (4, 65) ¼ 1.15, n.s and V ¼ 0.03, F(2, 67) ¼ 1.06, n.s respectively). group on change in performance ratings, V ¼ 0.10, F(3, 66) ¼ 2.52, Direct comparisons showed a statistical trend for the modification p ¼ 0.07. The MANOVA was followed up with discriminant analysis group to report less trait anxiety (t(63.97) ¼1.92, p ¼ 0.06, 43.80 and univariate ANOVAs. Correlations between each of the dependent vs. 48.03) and less negative affect (t(68) ¼1.85, p ¼ 0.07, 17.31 vs. variables and the discriminant function revealed that change in 19.91) than controls after cognitive error modification. The two success ratings loaded most highly (r ¼ 0.83). This indicates that the groups did not differ at baseline (see Table 2). variables of change in ratings of successfulness contributed to discriminating between the non-error modification and control 4. General discussion groups to the greatest extent. This was supported by separate univariate ANOVAs on the outcome variables which revealed This paper presents the development and validation of a new a significant effect of modification group on change in success ratings, form of Cognitive Bias Modification (CBM) designed to manipulate F(1, 68) ¼ 5.33, p < .05, d ¼ 0.56. Those in the modification group the cognitive errors known to characterize depression and anxiety. showed significantly increased ratings of success (t(34) ¼3.23, Cognitive Error Modification was capable of inducing systematic p < .01, 41.31 vs. 49.23), whereas there was no change in ratings for group differences in how hypothetical events were perceived in those in the control condition (t(34) < 1, n.s, 43.20 vs. 41.86, see Fig. 2). both a healthy and vulnerable sample. In Experiment 1, participants We also performed a correlation analysis to explore whether increase who generated cognitive errors during the modification phase in success ratings correlated with positive change on a composite perceived novel hypothetical events, their causes and outcomes in cognitive bias score calculated from participant’s ratings to the a more distorted, error-prone manner compared to participants Similarity Rating test.5 An increase in ratings of success was signifi- who had not generated errors. In Experiment 2, participants who cantly correlated with a more positive cognitive bias score, r ¼ 0.29, began the experiment with a propensity toward negative distorted p < .01. Consistent with the previous univariate analysis, this asso- thinking but who were assigned to generate non-errors during the ciation was significant in the non-error modification group, r ¼ 0.31, modification phase, perceived novel events in a more benign, error- p < .05 but not in the control group, r ¼ 0.02, ns. free way compared to the naturalistic responses of no-modification A MANOVA with change in ratings of anxiety, depression and control participants. happiness entered simultaneously as dependent variables revealed In addition there is preliminary support that the procedure is no significant effects of modification group on change in mood capable of changing cognitive, and to a lesser extent, emotional ratings, V ¼ 0.01, F(3, 66) < 1, n.s. Data for the anagram task are responses to real and imagined stress. Experiment 1 showed that shown in Table 4. according to self-reported mood ratings, those receiving error modification were significantly more susceptible to the negative 3.2.5. Trait measures of mood and cognition effects of stressful video clips than those receiving non-error modi- Separate MANOVAs with change in trait anxiety, BDI-II, negative fication and in particular, this effect was driven by a (near-significant) and positive affect scores and change in CCL-D and CCL-A scores statistical trend for a greater increase in negative affect for those receiving error modification. Experiment 2 provided convergent evidence for this finding using a different stressor task and a vulnerable sample to compare individuals who had and had not 5 A composite cognitive bias score was calculated using similarity ratings in the received the manipulation designed to reduce errorful thinking. following way e Cognitive bias score ¼ (Non-error target e Error target) þ (Positive fi foil e Negative foil). A positive score reflects a tendency toward less distorted, less When imagining a stressful social situation, non-error modi cation errorsome thinking. participants reported expecting to perform significantly better and 306 K.J. Lester et al. / J. Behav. Ther. & Exp. Psychiat. 42 (2011) 298e308

inferencing about personal performance, without performance itself being correspondingly affected. In line with this, the actual number of anagrams completed by each group was not significantly changed by the modification procedure. In contrast participants’ perception of their success on the task was altered, in that those receiving non-error modification considered their performance to be significantly improved compared to controls. Furthermore, within the non-error modification group, those participants who, after undertaking non-error modification, showed the least dis- torted thinking had the greatest increase in their perceptions of success on this task. Overall this result may reflect the induced tendency toward positive inferencing being deployed to a real-life ambiguous experience. As such this result may demonstrate one potentially important mechanism by which the beneficial effects of induced biases are operationalized at an individual, self-referent Fig. 2. Change in actual and rated performance on anagram task by group in Experi- ment 2 (error bars 1 standard error). level. In contrast to these effects on anagram performance, no differential effects on mood ratings were observed for the same task. Further analyses revealed that changes in mood ratings from reported at a statistical trend level that they would cope with their baseline to time 2 across the entire sample did not differ from 0 (all anxiety better, compared to controls. Overall emotional effects of t values < 1) suggesting that this task was not optimal to detect modification were mixed and notably smaller than the cognitive differences in emotional response across modification groups. effects. Although this is in line with previous studies using similar The data presented here on cognitive effects of error modification modification procedures, it highlights the need to further develop the add to those from several studies across different laboratories which efficacy and generalization of these procedures. Nevertheless our have demonstrated the efficacy of related procedures in effecting findings provide early promise that the kind of inferential biases change in cognition in selected high anxious samples (Beard & Amir, targeted by Cognitive Error Modification may causally contribute to 2008; Mathews et al., 2007; Murphy et al., 2007; Salemink et al., emotional vulnerability in response to potentially stressful events. 2009). Our effect sizes for cognitive change were large (Experiment This is consistent with models of emotional disorders that consider 1 targets: non-error d ¼ 0.97; error d ¼ 1.61) and broadly comparable inferential biases to be an important contributing factor for increased to those obtained for positive (d ¼ 0.55) and negative targets vulnerability to experiencing emotional disorders (D.M. Clark & (d ¼ 1.96) using related procedures (Mathews & Mackintosh, 2000). Wells, 1995; Mathews & Mackintosh, 1998; J.M.G. Williams, Watts, In our vulnerable sample (Experiment 2) effect sizes were even MacLeod, & Mathews, 1997). stronger than in the healthy sample (non-error targets d ¼ 1.14, error No significant differential effects of modification condition were targets d ¼ 1.20), thus ameliorating concerns that those with observed on any trait mood measures. Indeed trait mood was not a propensity to engage in cognitive distortions would be more expected to be significantly affected with only two sessions of active resistant to cognitive change as effected by the modification proce- modification (Experiment 2). However, it is indicative that direct dure. As in previous work, state mood was comparable between contrasts revealed near significant trends for reduced trait anxiety modification groups prior to completing the cognitive test rendering and negative affect in the modification group compared to controls it improbable that cognitive effects were epiphenomena of state and inspection of other mean scores (BDI-II, CCL) revealed that most mood changes induced by modification procedures. moved in a consistent direction. This is consistent with the pattern However, one caveat is that dysfunctional cognitions as of data on stressor tasks, in which broadly the same mood measures measured by the CCL, decreased significantly and comparably in were affected by modification, albeit in a more transient and latent both conditions. Both the CCL and Similarity Rating test were manner. We suggest that a longer intervention comprising addi- intended to be measures of distorted cognitions and thus, we might tional sessions of non-error modification could allow sufficient have anticipated observing significant modification effects on both opportunity for the acquisition of non-error inferencing to translate measures. The Similarity Rating test was more closely matched in into enduring reductions in trait vulnerability. This would be terms of the content and required response relative to the modifi- consistent with the pattern seen in related laboratory based modi- cation procedure, compared to the CCL, and this match could have fication work (Beard & Amir, 2008; Mathews et al., 2007; Salemink served to enhance training effects on this task. However, we note et al., 2009) in which reductions in trait vulnerability have been that reduction in CCL-Depression scores from baseline to test was observed following extended delivery schedules of bias modifica- correlated significantly with the composite cognitive bias score on tion. It is noteworthy that Salemink, van den Hout, and Kindt (2007) the Similarity Rating test (r ¼0.38, p < .01) and when modification observed that a single positive modification session consisting of 72 groups were examined independently, this effect remained signifi- modification trials was capable of bringing about a significant cant in the non-error modification group only (r ¼0.51, p < .01). decrease in trait anxiety. However, Salemink et al. (2007) suggest This work has a number of limitations beyond those already that this effect could simply be an artifact of their experimental mentioned. Most obviously, the present results cannot yet be procedure. In their experimental up state anxiety was consis- generalized to clinical samples, despite the ultimate goal of applying tently assessed prior to trait anxiety, meaning that responses to the the procedure for therapeutic gain. Nevertheless research using trait anxiety questionnaire may have been primed by responding to related methods has been effective in modifying biases and reducing the state anxiety questionnaire. Clearly, the potential of bias modi- anxiety vulnerability in clinically anxious individuals (Beard & Amir, fication interventions to bring about enduring changes in trait 2008). In ongoing work we are examining the efficacy of Cognitive is worthy of further investigation. Error Modification in clinically disordered samples. Several of the Results from Experiment 2 also revealed a predicted dissocia- important reported effects, (e.g. change in negative affect to the tion in the effects of the manipulation on actual and evaluated video stressor in Experiment 1, effects on the imagined social performance. We had reasoned that reducing the propensity to interaction test in Experiment 2, effect on performance measures on make cognitive errors should be reflected in more positive the anagram completion task) did not quite reach statistical K.J. Lester et al. / J. Behav. Ther. & Exp. Psychiat. 42 (2011) 298e308 307 significance but instead were near significant statistical trends all impact upon efficacy. The total number of modification trials in (p < .10). With a larger sample size and increased power, these Experiment 2 did not exceed the standard 100 trials often used in effects may have attained statistical significance. Post-hoc power previous work. It is very likely that stronger effects, for example analyses identified achieved b’s ranging from 0.53 to 0.70 for the changes in trait mood (Mathews et al., 2007) will require a larger aforementioned analyses. Sample size calculations based on the number of modification trials over extended delivery schedules. For (small to medium) effect sizes obtained, revealed that increased example, Salemink et al. (2009) observed significant reductions in sample sizes ranging from 43 participants per group (social inter- trait anxiety and general psychopathology after 8 sessions of action test analysis), to 63 participants per group (video stressor positive interpretation bias modification on consecutive days analysis) would be required to attain 80% power with an a of .05. (amounting to 512 modification trials). This supports the notion In additionwe did not include baseline measures of cognitive bias that aspects related to the delivery of the modification procedure meaning that group differences attributed to the modification may play an important role in its cognitive and emotional conse- procedure may simply have reflected pre-existing chance variation quences and will be important to systematically investigate. in bias. However we suggest this is an unlikelyexplanation, given the In summary, we attempted to improve ecological and clinical evidence of successful randomization of participants to groups (see validity by developing modification materials collected from clin- Table 2). Our data is also limited in terms of the conclusions that can ical sources and in consultation with practicing clinicians. In doing be drawn about the underlying cognitive mechanisms of change. this we hoped to begin to bring together experimental and clinical Although our technique was intended to tap a wider range of perspectives on biased information processing. The present find- inferential biases involving not only interpretation of ambiguity, but ings demonstrate that Cognitive Error Modification, which specif- also attributions and expectancies, there is no direct evidence vali- ically targets the type of maladaptive cognitions reported during dating this assumption. However our purpose here, in the first cognitive therapy, can promote positive inferences, reduce instance, was to establish the feasibility and efficacy of to stress and improve self-perceptions of performance Error Modification. Experiment 2 only included a ‘wait’ and not in both healthy individuals and those vulnerable to disorder such as a ‘no-contingency’ control group. In principle then, results could be anxiety and depression. This approach is particularly promising accounted for simply by exposure to an interactive testing session because patterns of thinking involving cognitive errors are known (analogous to non-specific therapeutic effects). Similarly, without to be relevant across a range of clinical conditions. Given the a no-contingency control, it remains possible that mere exposure to positive results reported here, targeting maladaptive cognition the experimental modification materials, as opposed to the direction using Cognitive Error Modification has the potential to offer a broad and type of cognitive operations performed on them, could account spectrum of application across anxiety and depression. The thera- for the effects observed. However we consider this an unlikely peutic potential of this new adaptation of CBM now requires explanation because work using no-contingency control groups has empirical testing in samples recruited from clinical services. already shown this not to be the case (e.g. Salemink et al., 2009). A related question which we did not seek to answer, is whether Acknowledgments Cognitive Error Modification delivers a better outcome for those receiving it compared to previous laboratory derived modification This research was funded by an MRC Postgraduate Research procedures. This question relates to content specificity: do bias Studentship awarded to Kathryn Lester and student support funds modification procedures tailored to specific disorder relevant content from the Department of Experimental Psychology, University of yield stronger beneficial effects than those less specifically targeted? Oxford. The authors wish to acknowledge the clinicians in Oxford To our knowledge this has not yet been experimentally tested and is and Buckinghamshire Mental Healthcare Trust who contributed to probably a pressing question to address, given the increasing variety the generation and ratings of the materials reported here. of cognitive modification techniques now under development (e.g. fi disorder speci c versions). What can be said at present based on Appendix A. Example Similarity Rating test item: Arbitrary fi comparison of effect sizes is that Cognitive Error Modi cation is at inference least as good in modifying cognition for those prone to make cognitive distortions as CBM-I procedures have been for those reporting high The Local Restaurant levels of trait anxiety. While the effects of Cognitive Error Modifica- tion on cognition were large (d ¼ 0.97e1.61), mood and stress reac- “You invite your parents for a meal at your local restaurant. tivity effects were mixed and noticeably smaller (for significant tests When you arrive at the restaurant it is very busy and you are only, d ¼ 0.45e0.75). Previous CBM-I interventions also report smaller glad you reserved a table. You sit down but after a few minutes effect sizes for emotional effects (e.g. d ¼ 0.79) in response to the you are still waiting to be.” s-rv-d (served) accident videos in Mackintosh et al. (2006) compared to cognitive effects. Wilson and colleagues (Wilson et al., 2006) observed that Did you take your sister for dinner? (No) CBM-I modified anxiety and depressive responses to the same video As you sit, you think the restaurant is very popular and you are stressor as that used in Experiment 1. While participants assigned to lucky to have a reservation (Non-error target) error modification were significantly more susceptible to the negative As you sit, you think the waiter is purposely not waiting on your effects of stressful video clips than those receiving non-error modi- table (Error target) fication, it remains to be established whether Cognitive Error modi- As you sit, you think the people at the table next to you seem fication is better capable of modifying anxiety and depressed mood polite (Positive foil) compared to previous CBM-I procedures. As you sit, you think the people at the table next to you seem There are other obvious possibilities, yet to be empirically disruptive (Negative foil). examined, for how best to maximize efficacy of bias modification procedures. The extent of active engagement with item content (Hoppitt et al., 2010a, 2010b) and whether verbal processing or Appendix. Supplementary material imagery is used (Holmes et al., 2009, 2006) are already being investigated. In addition factors such as number of modification Supplementary data related to this article can be found online at sessions, spacing of sessions and number of items per session may doi:10.1016/j.jbtep.2011.01.001. 308 K.J. Lester et al. / J. Behav. Ther. & Exp. Psychiat. 42 (2011) 298e308

References Lawson, C., MacLeod, C., & Hammond, G. (2002). Interpretation revealed in the blink of an eye: depressive bias in the resolution of ambiguity. Journal of , 111(2), 321e328. Ahrens, A. H., & Haaga, D. A. (1993). The specificity of attributional style and Lefebvre, M. F. (1981). and cognitive errors in depressed expectations to positive and negative affectivity, depression, and anxiety. psychiatric and low back pain patients. Journal of Consulting and Clinical Cognitive Therapy and Research, 17(1), 83e98. Psychology, 49(4), 517e525. Beard, C., & Amir, N. (2008). A multi-session interpretation modification program: Leitenberg, H., Yost, L. W., & Carroll-Wilson, M. (1986). Negative cognitive errors in changes in interpretation and social anxiety symptoms. Behaviour Research and children: questionnaire development, normative data, and comparisons Therapy, 46(10), 1135e1141. doi:10.1016/j.brat.2008.05.012. between children with and without self-reported symptoms of depression, low Beck, A. T. (1963). Thinking and depression:1. Idiosyncratic content and cognitive self-esteem, and evaluation anxiety. Journal of Consulting and Clinical distortions. Archives of General Psychiatry, 9, 324e333. Psychology, 54(4), 528e536. Beck, A. T. (1976). Cognitive Therapy and the Emotional Disorders. New York: Penguin Mackintosh, B., Mathews, A., Yiend, J., Ridgeway, V., & Cook, E. (2006). Induced Group. biases in emotional interpretation influence stress vulnerability and endure Beck, A. T., Emery, G., & Greenberg, R. L. (1985). Anxiety disorders and phobias: A despite changes in context. Behavior Therapy, 37(3), 209e222. doi:10.1016/ cognitive perspective. New York: Basic Books. j.beth.2006.03.001. Beck, A. T., Rush, A. J., Shaw, B. F., & Emery, G. (1979). Cognitive therapy of depression. MacLeod, A. K., & Cropley, M. (1995). Depressive future-thinking: the role of valence New York: The Guildford Press. and specificity. Cognitive Therapy and Research, 19,35e50. Beck, A. T., Steer, R., & Brown, G. (1996). Beck depression inventory - second edition MacLeod, A. K., Tata, P., Kentish, J., & Jacobsen, H. (1997). Retrospective and prospective manual. San Antonio, Texas: The Psychological Corporation. cognitions in anxiety and depression. Cognition and , 11(4), 467e479. Blackburn, I. M., & Eunson, K. M. (1989). A content analysis of thoughts and Mathews, A., & Mackintosh, B. (1998). A cognitive model of selective processing in elicited from depressed patients during cognitive therapy. British anxiety. Cognitive Therapy and Research, 22(6), 539e560. doi:10.1023/ Journal of , 62(Pt 1), 23e33. A:1018738019346. Burns, D. D. (2000). Feeling good: The new mood therapy. New York: William Morrow & Co. Mathews, A., & Mackintosh, B. (2000). Induced emotional interpretation bias and Butler, G., & Mathews, A. (1983). Cognitive processes in anxiety. Advances in anxiety. Journal of Abnormal Psychology, 109(4), 602e615. doi:10.1037/0021- Behaviour Research and Therapy, 5,51e62. 843X.109.4.602. Butler, G., & Mathews, A. (1987). Anticipatory anxiety and perception. Cognitive Mathews, A., Ridgeway, V., Cook, E., & Yiend, J. (2007). Inducing a benign inter- Therapy and Research, 11(5), 551e565. pretational bias reduces trait anxiety. Journal of Behavior Therapy and Experi- Carter, R. M., Wittchen, H. U., Pfister, H., & Kessler, R. C. (2001). One-year prevalence mental Psychiatry, 38(2), 225e236. doi:10.1016/j.jbtep.2006.10.011. of subthreshold and threshold DSM-IV generalized in Mogg, K., Bradbury, K. E., & Bradley, B. P. (2006). Interpretation of ambiguous a nationally representative sample. Depression and Anxiety, 13(2), 78e88. information in clinical depression. Behaviour Research and Therapy, 44(10), Clark, D. M., & Wells, A. (1995). A cognitive model of social phobia. In 1411e1419. doi:10.1016/j.brat.2005.10.008. R. G. Heimberg, M. Liebowitz, D. Hope, & F. Scheier (Eds.), Social phobia: Diag- Murphy, R., Hirsch, C. R., Mathews, A., Smith, K., & Clark, D. M. (2007). Facilitating nosis, assessment, and treatment (pp. 69e93). New York: The Guildford Press. a benign interpretation bias in a high socially anxious population. Behaviour Cropley, M., & MacLeod, A. K. (2003). Dysphoria, attributional reasoning and future Research and Therapy, 45(7), 1517e1529. doi:10.1016/j.brat.2007.01.007. event . and Psychotherapy, 10, 220e227. Reardon, J. M., & Williams, N. L. (2007). The specificity of cognitive vulnerabilities to Das-Munshi, J., Goldberg, D., Bebbington, P. E., Bhugra, D. K., Brugha, T. S., emotional disorders: anxiety sensitivity, looming vulnerability and explanatory Dewey, M. E., & Prince, M. (2008). Public health significance of mixed anxiety style. Journal of Anxiety Disorders, 21(5), 625e643. and depression: beyond current classification. British Journal of Psychiatry, 192 Richards, A., & French, C. C. (1992). An anxiety-related bias in semantic activation (3), 171e177. doi:10.1192/bjp.bp.107.036707. when processing threat/neutral homographs. Quarterly Journal of Experimental Eysenck, M. W., Mogg, K., May, J., Richards, A., & Mathews, A. (1991). Bias in Psychology, 45A(3), 503e525. interpretation of ambiguous sentences related to threat in anxiety. Journal of Salemink, E., van den Hout, M., & Kindt, M. (2007). Trained : val- Abnormal Psychology, 100(2), 144e150. idity and effects on anxiety. Journal of Behavior Therapy and Experimental Harvey, A. G., Watkins, E., Mansell, W., & Shafran, R. (2004). Cognitive behavioural Psychiatry, 38(2), 212e224. doi:10.1016/j.jbtep.2006.10.010. processes across psychological disorders: A transdiagnostic approach to research Salemink, E., van den Hout, M., & Kindt, M. (2009). Effects of positive interpretive and treatment. Oxford: Oxford University Press. bias modification in highly anxious individuals. Journal of Anxiety Disorders, 23 Henriques, G., & Leitenberg, H. (2002). An experimental analysis of the role of (5), 676e683. doi:10.1016/j.janxdis.2009.02.006. cognitive errors in the development of depressed mood following negative Sheehan, D. V., Lecrubier, Y., Harnett-Sheehan, K., Amorim, P., Janavs, J., Weiller, E., social feedback. Cognitive Therapy and Research, 26(2), 245e260. et al. (1998). The Mini International Neuropsychiatric Interview (M.I.N.I.): the Holmes, E. A., Lang, T. J., & Shah, D. M. (2009). Developing interpretation bias development and validation of a structured diagnostic psychiatric interview. modification as a ‘cognitive vaccine’ for depressed mood - imagining positive Journal of Clinical Psychiatry, 59,22e33. events makes you feel better than thinking about them verbally. Journal of Spielberger, C. D., Barker, L. R., Russell, S. F., Crane, R. S., Westberry, L. G., Knight, J., & Abnormal Psychology, 118,76e88. Marks, E. (1979). The preliminary manual for the state-trait personality inventory. Holmes, E. A., & Mathews, A. (2005). Mental imagery and emotion: a special rela- Tampa: University of South Florida. tionship? Emotion, 5(4), 489e497. Spielberger, C. D., Gorsuch, R. L., Lushene, R., Vagg, P. R., & Jacobs, G. A. (1983). Holmes, E. A., Mathews, A., Dalgleish, T., & Mackintosh, B. (2006). Positive inter- Manual for the state-trait anxiety inventory. Palo Alto, California: Consulting pretation training: effects of mental imagery versus verbal training on positive Psychologists Press. mood. Behavior Therapy, 37(3), 237e247. Steer, R. A., Beck, A. T., Clark, D. A., & Beck, J. S. (1994). Psychometric properties of Hoppitt, L., Mathews, A., Yiend, J., & Mackintosh, B. (2010a). Cognitive bias modi- the cognition checklist with psychiatric outpatients and university students. fication: the critical role of active training in modifying emotional responses. Psychological Assessment, 6(1), 67e70. Behavior Therapy, 41(1), 73e81. doi:10.1016/j.beth.2009.01.002. Watson, D., & Clark, L. A. (1994). The PANAS-X: Manual for the positive and negative Hoppitt, L., Mathews, A., Yiend, J., & Mackintosh, B. (2010b). Cognitive mechanisms affect schedule - expanded form. The University of Iowa. underlying the emotional effects of bias modification. Applied Cognitive Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief Psychology, 24(3), 312e325. doi:10.1002/acp.1678. measures of positive and negative affect: the PANAS scales. Journal of Personality Ilardi, S. S., & Craighead, W. E. (1999). Rapid early response, cognitive modification and , 54(6), 1063e1070. and nonspecific factors in cognitive behavior therapy for depression: a reply to Weems, C. F., Berman, S. L., Silverman, W. K., & Saavedra, L. M. (2001). Cognitive Tang and DeRubeis. Clinical Psychology: Science and Practice, 6, 295e299. errors in youth with anxiety disorders: the linkages between negative cognitive Johnson, K. A., Johnson, J. E., & Petzel, T. P. (1992). Social anxiety, depression and errors and anxious symptoms. Cognitive Therapy and Research, 25(5), 559e575. distorted cognitions in college students. Journal of Social and Clinical Psychology, Williams, J. M. G., Watts, F. N., MacLeod, C., & Mathews, A. (1997). Cognitive 11(2), 181e195. psychology and emotional disorders (2nd ed.). Chichester: John Wiley & Sons. Judd, L. L., Kessler, R. C., Paulus, M. P., Zeller, P. V., Wittchen, H. U., & Kunovac, J. L. Wilson, E. J., MacLeod, C., Mathews, A., & Rutherford, E. M. (2006). The causal role of (1998). Comorbidity as a fundamental feature of generalized anxiety disorders: interpretive bias in anxiety reactivity. Journal of Abnormal Psychology, 115(1), results from the National Comorbidity Study (NCS). Acta Psychiatrica Scandi- 103e111. doi:10.1037/0021-843X.115.1.103. navica, 98,6e11.