The Role of Presentation Media in the Moral Domain
A Doctoral Thesis
Caitlin McCurrie
Supervised By: Dr Simon Laham
A thesis submitted for the degree of Doctor of Philosophy at The University of Melbourne in November, 2018
Melbourne School of Psychological Science
https://orcid.org/0000-0002-7444-8284
Abstract As the prevalence of text-based and computer-mediated communication has increased, a number of researchers in psychology have identified ways in which medium affects perception and behaviour. Yet, there is very limited research attention by moral psychology researchers on the topic of medium effects. This thesis presents a systematic exploration of the ways in which medium might impact moral outcomes. The first stream of this thesis examined the role of medium in moral outcomes in a specific example of text- based communication, computer-mediated-communication (CMC). I utilised a theoretical framework from the CMC literature, the cues-filtered-out approach, as a means of understanding moral behaviour in the context of CMC. Overall, there was limited support for this theoretical approach; however, both studies in this stream identify ways in which medium affects positive and negative moral behaviours and moral perception. Study 1 demonstrated that listening to the voice of someone who disagrees with your political opinion can lead to more aggression and paradoxically, humanisation, compared to reading an equivalent text passage. Study 2 shows that, under certain circumstances, reading a dating profile can lead to more self-disclosure compared to watching an equivalent video. Study 2 also found that in the text condition some aspect of person perception was exaggerated.
In the second stream of this thesis, I took a more basic approach to exploring the effect of presentation medium on moral outcomes. First, in Study 3, I conducted a systematic review of the moral psychology literature that revealed that more than 92% of research on moral judgements relied only on text stimuli. Given that text stimuli lack the rich variety of morally-relevant social and contextual cues available in everyday interactions, I argue that a consequence of this pervasive ecological invalidity may be that moral psychological theories are mischaracterized by an overreliance on cue-impoverished moral stimuli. I address this limitation by developing an ecologically valid and cue-rich, moral video stimulus set with
matching text (Study 4). Next, I use this stimulus set to explore the role of medium on moral judgment, perception, and affect (Study 5). I found that judging moral transgressions in a cue-poor medium (i.e., text) results in a pattern of moral judgement and affective responses
(i.e., wrongness, moral foundation categorisations, arousal, and humanness) that did not generalise to a more ecologically valid medium (i.e., video). Overall, there is evidence that presentation medium affects a number of morally relevant outcomes (i.e., wrongness, arousal, moral foundation categorisation and humanness), although the size of these effects was smaller than expected (in most cases) and the directionality of some effects was contrary to expectations. Specifically, when content was presented via text (compared to video) judgements were harsher, but (contrary to expectations) the content was less arousing, and there was greater consensus on why a moral transgression was wrong (i.e., moral foundation categorisation). Contrary to expectations, when content was presented via text (compared to video) both perpetrators and victims were also judged as more human. There was no evidence to suggest that the presentation medium of the moral content affected perceptions of causality, blame, nor evoked empathy (cognitive or emotional).
Across five studies, I find evidence that the medium by which we communicate or present a stimulus shapes morally relevant outcomes. However, the exact mechanism by which medium affects moral outcomes remains unclear, that is, there was limited support for
CMC theories in accounting for medium effects of a positive or negative moral behaviour.
Importantly, contrary to dominant rhetoric in the CMC literature and general population, text as a presentation or communication medium does not necessarily lead to more negative outcomes. Instead, I find evidence that text can have positive implications for moral behaviour: facilitating more moral judgment, less antisocial behaviour (e.g., aggression), and more positive moral behaviours (e.g., self-disclosure). Overall, this thesis has identified the
importance of presentation medium in moral psychology and identified a fertile ground for future research.
Declaration by author
This thesis is composed of my original work, and contains no material previously published or written by another person except where due reference has been made in the text.
I have clearly stated the contribution by others to jointly authored works that I have included in my thesis. I have clearly stated the contribution of others to my thesis as a whole, including statistical assistance, survey design, data analysis, significant technical procedures, professional editorial advice, and any other original research work used or reported in my thesis. The content of my thesis is the result of work I have carried out since the commencement of my research higher degree candidature and does not include a substantial part of work that has been submitted to qualify for the award of any other degree or diploma in any university or other tertiary institution. I have clearly stated which parts of my thesis, if any, have been submitted to qualify for another award. I acknowledge that an electronic copy of my thesis must be lodged with the University Library and, subject to the policy and procedures of The University of Melbourne, the thesis be made available for research and study in accordance with the Copyright Act 1968 unless a period of embargo has been approved by the Dean of the Graduate School. I acknowledge that copyright of all material contained in my thesis resides with the copyright holder(s) of that material. Where appropriate I have obtained copyright permission from the copyright holder to reproduce material in this thesis. This thesis is fewer than the maximum word limit in length, 100,000 words, exclusive of tables, maps, bibliographies and appendices.
Publications during candidature
Falzon, L., McCurrie, C., & Dunn, J. (2017, July). Representation and Analysis of Twitter
Activity: A Dynamic Network Perspective. In Proceedings of the 2017 IEEE/ACM
International Conference on Advances in Social Networks Analysis and Mining 2017
(pp. 1183-1190). ACM
McCurrie, C., Crone, D., Bigelow, F. J., & Laham, S. (2018). Moral and Affective Film Set
(MAAFS): A Normed Moral Video Database. PLoS ONE 13(11).
Simpson, A., McCurrie C., Rios, K., (under revision). Perceived Morality and Anti-Atheist
Prejudice: A Replication and Extension. International Journal for Philosophy of
Religion.
Publications included in this thesis
McCurrie, C., Crone, D., Bigelow, F. J., & Laham, S. (2018). Moral and Affective Film Set
(MAAFS): A Normed Moral Video Database. PLoS ONE.
An adapted version of this publication appears as Chapter 6 of this thesis.
Contributor Statement of contribution
McCurrie, C. Designed experiments (70%) Created study materials (100%) Collected data (100%) Analysed data (80%) Wrote manuscript (80%)
Crone, D. Analysed data (10%) Provided comments on manuscript (50%) Wrote manuscript (10%)
Bigelow, F. J Analysed data (10%)
Laham, S. Designed experiments (30%) Provided comments on manuscript (50%) Wrote manuscript (10%)
Contributions by others to the thesis My advisor, Dr Simon Laham, provided guidance on theory and study design, and commented on the thesis drafts and the manuscript included as a chapter.
Statement of parts of the thesis submitted to qualify for the award of another degree None
Acknowledgements
This thesis has been a life changing journey. By far, the most challenging venture I have ever attempted: my thesis has taken me to live overseas, extended me far beyond my comfort zone, and given me a passion for topics I would never expected. While this has not always been an easy journey, in fact, it was rarely easy, I’m thankful to my 2014-self for choosing this path. I’m proud of the person I have become.
None of this would have been possible without the love, support, and patience of those that have joined me along the journey and those that have been with me throughout my life. Now, to begin a long list of those that I need to pay thanks to - how lucky I am to have all these wonderful people in my life!
First, I would like to thank my supervisor, Simon, for his guidance and feedback.
While sometimes challenging, his feedback has improved my skills as a researcher exponentially and I am thankful to him for that. Next, I would like to give thanks to the very many people in the academic world who have provided comments on drafts, given feedback, or listened to me complain. (In no particular order) thank you Shaheed, Ravini, Fabricio,
Damien, Ain, Melissa, Hanne, and Kanishka.
I would like to acknowledge a great mentor and friend, Lucia. The opportunity to do work outside of my PhD reminded me of my capabilities and helped me find some balance in the chaotic, PhD-world. Thank you for your support and friendship.
To my great friends Michael and Sean, you have supported me both mentally and through reading tedious drafts.
Michael: I am so appreciative of your support and encouragement. Thank you for your patience in listening to me complain, encouraging me to persevere, and being a shoulder to cry on. I look forward to a long friendship of trading (mostly friendly) insults.
Sean: You are both a friend and a role model. Thank you for taking the time to mentor me, listen to me vent, and providing hugs. I so admire your abilities and look forward to watching your career blossom. I hope that I will be a part of that.
One of the few people that has been along this whole journey with me, Nick. We made it. Who would have thought 6 years ago that we would be here? Not too shabby.
Thank you to all my friends outside of the academic world, while you didn’t have the delight of providing feedback on my work, you helped my sanity. Especially those in the
PTC and TSF communities (Emma, Matti, Tom, J.P.). Importantly, thank you to my best friend Louis for his emotional support.
To Tristan, I wouldn’t have made it through this without you. You came into my life when I was struggling the most and helped me find my way through it. For our whole relationship, I have been on this PhD journey. You have heard me complain (endlessly!), cry, celebrate. You have put up with me working long hours and becoming consumed in my work. You even made a cameo in a stimulus set. I know that this hasn’t been easy for you.
You have travelled this road as much as I have. Thank you for all that you have done. I hope that I have made you proud. One day, I will give you that same support to chase your passion.
Lastly, I owe thanks to my parents. Quite obviously, I would never have been able to attempt this thesis without them. How lucky I am to have these parents. Thank you for managing the duress I put you under when I chose to move overseas. Thank you for all that you do. I can’t list all those things because the list would be as long as my thesis. Know that I never take it for granted. All that I do, I do to make you both proud.
“And they all lived happily ever after. The end.”
I dedicate this thesis to my parents, Jackie Batten and Shane McCurrie.
The most wonderful parents that a 12-year old could ask for.
Table of Contents
Chapter 1 : General Introduction ...... 1 The Importance of Medium in Moral Psychology ...... 2 Non-Verbal Social and Contextual Cues ...... 3 Stream 1. The Effects of Media on Moral Behaviours ...... 7 Chapter 2: Cues Filtered Out Theory ...... 7 Cues-Filtered-Out Theories ...... 8 Chapter 3: Computer Mediated Communication and Aggression ...... 10 Social and Human Presence ...... 13 Dehumanising of the Self Online ...... 17 Study Overview ...... 19 Stimuli Development...... 19 Voice Pilot Test ...... 21 Main Study ...... 23 Methods ...... 23 Participants ...... 23 Materials and Procedue ...... 24 Results ...... 31 Pre-Processing ...... 31 Analysis Overview ...... 33 Exploratory Correlation ...... 33 Communication Medium Comparisons ...... 34 Structural Equation Modelling ...... 35 Discussion ...... 39 Moral Disengagement Underlies Flaming ...... 39 Voice Conveys Human Uniquenes ...... 41 The Paradoxical Effects of Communication Medium ...... 43 Practical Implications ...... 44 Study Limitations and Future Directions...... 44 Conclusion ...... 46 Chapter 4: Self-Disclosure in Computer Mediated Communication ...... 47 Self-Disclosure and Computer-Mediated-Communication ...... 49 Hyperhumanisation and Hyperpersonalism ...... 50
More Human in Text: Greater Moral Status ...... 51 Alternative Theoretical Accounts of Self-Disclosure ...... 53 Possible Moderators, Mediators, and Control Variables ...... 56 Study Overview ...... 62 Stimuli Development...... 64 Pilot Study One ...... 67 Pilot Study Two ...... 69 Video Stimuli Development ...... Error! Bookmark not defined. Main Study ...... 72 Methods ...... 72 Participants ...... 72 Study Design...... 73 Procedure and Measures ...... 73 Results ...... 79 Data Analysis Strategy ...... 79 Correlations and Descriptive Statistics ...... 80 Manipulation Check and Medium Effects on Humanness ...... 84 Hyperhumanisation ...... 87 Reduced Cues Theory ...... 92 Social Identity and Deindividuation Effects ...... 93 Exploratory Analysis ...... 95 Discussion ...... 99 Expectations...... 106 Conclusion ...... 107 Chapter 5: Discussion of Stream 1...... 109 Medium Effects on Moral Behaviour ...... 110 Medium Effects on Humanness ...... 111 The Role of Expectations for Flaming ...... 112 The Role of Expectations for Self-Disclosure ...... 116 Stream 2: The Effects of Medium on Moral Judgement...... 117 Research Agenda for Stream 2 ...... 117 Chapter 5: A Systematic Literature Review of Presentation Medium in Moral Psychology 119 Overview of the Systematic Review ...... 124
Methods ...... 125 Review Scope and Theoretical Definitions ...... Error! Bookmark not defined. Results and Discussion ...... 133 Presentation Medium Over Time ...... 134 Implications of Over-Reliance on Text Stimuli ...... 135 Medium Effects on Moral Character ...... 138 Conclusion ...... 138 Chapter 6: Developing a Moral and Affective Film Set ...... 140 Overview of Stimulus Set Development and Validation ...... 141 Method: Video Collection ...... 142 Participants ...... 142 Procedure and Materials ...... 142 Results and Discussion: Video Collection ...... 149 Methods: Video Validation ...... 149 Participants ...... 149 Procedure and Materials ...... 151 Results and Discussion: Video Validation ...... 153 Inter-Rater Reliability ...... 158 Breadth and Representativeness of Moral Content ...... 160 Moral and Affective Features ...... 165 Previous Exposure to the Clips ...... 171 Commonness ...... 172 Possible Applications for the MAAFS ...... 173 Conclusion ...... 174 Chapter 7: Empirically Testing the Effect of Presentation Medium on Moral Judgement ... 176 Moral Foundation Categorisations ...... 176 Arousal and Empathy ...... 177 Wrongness Judgments ...... 179 Person Perception ...... 180 Individual Difference Moderators ...... 181 Transcription of Moral Videos to Moral Texts ...... 184 Method ...... 185 Participants ...... 185
Procedure ...... 185 Results ...... 185 Inter-Rater Reliability ...... 185 Transcription Representativeness ...... 186 Testing the Effect of Presentation Medium...... 191 Method ...... 191 Design ...... 191 Participants ...... 192 Procedure and Materials ...... 192 Results ...... 197 Analysis Strategy ...... 197 Correlations and Descriptive Statistics ...... 197 Moral Foundation Categorisation (H1) ...... 201 Arousal and Empathy (H2 and H3) ...... 204 Wrongness Judgements (H4) ...... 204 Humanness Perception (H6) ...... 205 Blameworthiness and Intention (H7 and H8) ...... 207 Individual Differences ...... 207 Discussion ...... 209 Medium Effects on Moral Emotions ...... 212 Medium Effects on Person Perception ...... 214 Limitations and Future Directions ...... 218 Conclusion ...... 219 Chapter 8: General Discussion...... 221 Supplementary Materials...... 242 Appendix A...... 242 Appendix B ...... 249 Glossary ...... 252
Table of Figures and Tables
Figure 1. Hypothesized mechanism for the effect of communication medium on flaming. ... 17
Figure 2. One of the stimuli used in the text condition ...... 20
Figure 3. The procedure of allocating participants to each stimulus ...... 25
Figure 4. Initial, hypothesised model with standardised estimates...... 36
Figure 5. Refined model...... 37
Figure 6. The distribution of HU and HN scores for each social category ...... 68
Figure 7. Least squares (LS) means for HN ratings for each of the experimental stimuli...... 70
Figure 8. Least squares (LS) means for HU ratings for each of the experimental stimuli...... 71
Figure 10. Simple slopes for HN ratings for high and low levels of the HN and HU manipulations ...... 87
Figure 11. Simple slopes for breadth.log ratings for high and low levels of the HU manipulation and medium...... 91
Figure 12. An overview of publication selection and exclusion for the systematic literature review...... 129
Figure 13. Proportion of presentation media in the reviewed studies...... 134
Figure 14. Frequency of Text, Non-Text Stimuli, and Total Number of Studies From 2001 to
2017...... 135
Figure 15. An overview of the development of the MAAFS including video collection and video validation phases...... 142
Figure 16. Box-plots of averages for each video in the MAAFs for moral judgements, arousal...... 158
Figure 21. Proportions of variance attributable to video, rater, and residual error, sorted in order of variance attributable to video ...... 160
Figure 17. Box-plots of uniqueness scores for videos categorised into each moral foundation
...... 162
Figure 18. Confusion matrix where the moral videos (N = 69) are categorised according to the most frequently selected moral foundation and the proportion with which each alternative foundation was selected is shown...... 165
Figure 19. Distributions of averages for each video in the MAAFs for discrete emotions ... 166
Figure 20. A comparison between affective film sets’ and the MAAFS’ capacity to induce discrete emotions...... 170
Figure 22. Correlational matrix of dependent measures and presentation medium...... 199
Figure 23. Confusion matrix where moral videos are categorised according to the most frequently selected moral foundation and the proportion with which each alternative foundation was selected is shown...... 203
Figure 24. Confusion matrix where moral texts are categorised according to the most frequently selected moral foundation and the proportion with which each alternative foundation was selected is shown...... 203
Table 1. Flaming Response Scales for Participants for or Against Same-Sex Marriage 27
Table 2. Testing for Moderation Effects of Attitude Strength and Attitude Orientation ...... 32
Table 3. Exploratory Pearson Bivariate Correlations Between Measured Variables ...... 34
Table 4. Moderated Mediation Analyses ...... 39
Table 5. Mixed-Effects Models for Assessing the Effectiveness of the Humanness
Manipulation ...... 71
Table 6. Marginal Means and Pairwise Comparisons for the Manipulation Check ...... 71
Table 7. Factor Loadings and Communalities ...... 78
Table 8. Descriptive Statistics for the Two Online Expectations Factors ...... 78
Table 9. Correlational Matrix for Dependent Measures ...... 82
Table 10. Descriptive Statistics for Measured Variables ...... 83
Table 11. Polychoric Correlations Between Medium and Dependent Measures ...... 84
Table 12. Mixed-Effects Model for Assessing the Effectiveness of the Humanness
Manipulation ...... 85
Table 13. Means and SD for HN Rating at Different Levels of the Manipulations ...... 85
Table 14. Means and SD for HU Rating at Different Levels of the Manipulations ...... 85
Table 15. Mixed Effects Models That Assessed Hyperpersonal/Hyperhumanness Hypotheses
...... 90
Table 16. Mixed Effects Model fitted to assess H7 and H8 ...... 93
Table 17. Estimated Mediation Effects of Self-Consciousness on Medium and Self-
Disclosure ...... 93
Table 18. Mixed Effects Model fitted to assess H7 and H8 ...... 94
Table 19. Estimated Mediation Effects of Assumed-Similarity on Medium and Self-
Disclosure ...... 94
Table 20. Estimated Mediation Effects of Social Presence on Medium and Self-Disclosure . 95
Table 21. Mixed Effects Model fitted to Assess Trust as a Moderator ...... 96
Table 22. Estimated Mediation Effects of Attraction on Medium and Self-Disclosure ...... 97
Table 23. Mixed Effects Models for Expectations of Interactions and Self-Disclosure ...... 98
Table 24. The Derivation of Search Terms from Prominent Theories and Studies ...... 127
Table 25. Theoretical Definitions and Scope ...... 126
Table 26. Effect Codes, Definitions, and Coding Options...... 131
Table 27. Inter-rater reliabilities for the coding of the included papers ...... 133
Table 28. A complete list of vignettes used in the development of the Moral Video Set ..... 143
Table 29. Moral Foundation Definitions Used as Search Prompts in the Development of the
Moral Video Set ...... 148
Table 30. A Comparison Between the Rating Frequency and Sample Size of the MAAFs and
Comparable Affective Film or Moral Stimulus Sets...... 150
Table 31. Summary of the Measures Used to Norm and Validate the Moral Videos ...... 152
Table 32. Summary Descriptions of the MAAFS ...... 154
Table 33. Features of the Stimulus Set: Descriptive and Distributional Measures for each
Variable ...... 157
Table 34. Bivariate Correlations Between the Affective and Moral Ratings ...... 171
Table 35. Waves of Modification and Representativeness Ratings for Vignettes With < 4.0
Representativeness Ratings ...... 187
Table 37. Summary of the Measures Used to Rate the Video and Text Stimuli ...... 193
Table 38. Descriptive Statistics for Dependent Measures ...... 197
Table 39. Results for the Mixed-Effects Models ...... 200
Table 40. Estimated Mediation Effects of Arousal on Medium and Wrongness ...... 205
Table 41. Estimated Mediation Effects of Humanness on Medium and Wrongness ...... 206
Table 42. Studies with Non-Text Moral Stimuli ...... 242
Chapter 1: General Introduction 1
Chapter 1 : General Introduction We live in a world saturated with computer technology: from singles ‘Tindering’ for a soul mate to politicians ‘Tweeting’ their rhetoric. Although the technological era has had a profound impact on countless industries and institutions, it has caused a particularly significant change to the way we communicate: 74% of American adults use social media and communication services (e.g. chat programs), and over 90% own a mobile phone (World
Internet Project, 2014). Consequently, much communication is now fundamentally different than in the past, when interactions were (almost) exclusively face-to-face (FtF). For example, social interactions may now be anonymous, may occur across long periods of time or distances, and, importantly, are often text-based. More than 20 decades of research have identified a myriad of ways in which technology, specifically text-based communication, has impacted how people behave and interact. For example, text-based communication, relative to FtF, has been associated with ‘flaming’ or hostile, aggressive and anti-social behaviour
(Chui, 2014; Douglas, 2008; Joinson, 2007; Kiesler, Zubrow, Moses, & Geller, 1985; Suler,
2004) and field studies have found that in online interactions levels of interpersonal intimacy and self-disclosure can exceed that of comparative FtF interactions (Jiang, Bazarova, &
Hancock, 2011; Joinson, Reips, Buchanan, & Schofield, 2010; Walther, 1996; Walther, &
Parks, 2002; Walther, Anderson, & Park, 1994). Yet, the implications for these technological changes in the moral domain remain largely unaddressed. I argue that the medium by which we communicate has implications for various kinds of moral phenomena, from moral perception to judgement and behaviour. Communication medium refers to the representation format used to disseminate information. Examples include computer-mediated- communication, voice, radio, film, books (Kress, 2009). Across two streams, this thesis examines the way in which moral psychology is impacted by (1) the medium used to
Chapter 1: General Introduction 2
communicate (e.g., during an interaction) and (2) the medium used to present a stimulus (e.g., during an experiment).
The Importance of Medium in Moral Psychology As the prevalence of text-based and computer-mediated communication has increased, a growing number of researchers in various fields of psychology have identified ways in with medium affects perception and behaviour. For example, research in organisational psychology has shown that the differences in the social and contextual cues available in different media influence: the performance of teams on problem-solving and negotiation tasks (Fletcher & Major, 2006), the frequency of teamwork behaviours (Fletcher
& Major, 2006), negotiation outcomes (Swaab, Galinsky, Medvec, & Diermeier, 2012a), and the persuasiveness of a message (Chaiken & Eagly, 1976). Cognitive psychology research has found a ‘modality effect’, an idea closely linked to medium, by which some memory and learning tasks are improved when the content is presented audibly (i.e., using sound) rather than visually (i.e., using imagery) (Herz 1998, Gottfried, Smith et al., 2004; Ginns 2005).
Research in social psychology has shown that people rely more heavily on stereotypes (Epley
& Kruger, 2005), their expectations of others (Walther & Tong, 2014), and are more likely to exaggerate the importance of the available information (Hancock & Dunham, 2001) in text compared to non-text communication (e.g., voice, FtF). The nature of interactions that occur in text-only environments (e.g. email) are also different from those that occur in richer media: text-mediated interactions can reduce the salience of social hierarchies (Dubrovsky, Kiesler,
& Sethna, 1991), increase the influence of social identity on behaviour (Postmes, Spears, &
Lea, 1998), and change both what people choose to communicate (Berger & Milkman, 2012) and how something is communicated (Walther, 1996).
Yet, there is very limited research attention by moral psychology researchers on the topic of medium effects. This oversight has a number of important implications: first, many
Chapter 1: General Introduction 3
of the behavioural phenomena discovered from studying the effects of communication medium on behaviour and perception are morally relevant. For example, some researchers have found that there are increased levels of aggression in text versus non-text (e.g., FtF) interactions (Abrams, 2003; Castellá et al., 2000; Moor et al., 2010). By drawing on aspects of moral-psychological theories, we may be able to better understand how and why text leads to increased aggression.
Second, if content presented via text affects morally relevant perception and behaviour, then the use of text to present experimental stimuli in moral psychology research may lead to a different pattern of results than if a non-text media was used. As a result, the use of text-based stimuli could lead to biased results that do not generalise to real-world moral responses.
Non-Verbal Social and Contextual Cues Despite decades of evidence that medium affects perception and behaviour, there is disagreement over precisely how medium influences behaviour. Some researchers suggest that text-based interaction causes task-orientated, impersonal conversations (Constant,
Sproull et al., 1996; Haythornthwaite, 2002; Kemp and Rutter 1982; Kiesler, Siegel et al.
1984; Short, Williams et al., 1976; Siegel, Dubrovsky et al., 1986; Sproull & Kiesler, 1986) and promotes deindividuation and anti-social behaviour (Douglas 2008; Hiltz, Joinson 2007;
Luzón, 2011; Postmes, Spears et al. 1998; Turoff et al., 1989). Yet, others demonstrate relational intimacy in text interactions, positively skewed impression formation, and self- disclosure that surpasses what is seen in equivalent FtF relationships (Collins Tidwell &
Walther 2002; Walther 1996, 2007). However, most researchers agree that the absence of non-verbal social and contextual cues in text relative to cue-rich stimuli (e.g., FtF, image, or voice) is central to understanding medium effects.
Chapter 1: General Introduction 4
Non-verbal cues have an important role in communication and so their absence can impact how (or how accurately) an actor interprets a message or stimulus. Non-verbal cues refer to all elements of message communication that occur without sending or receiving language (Brehm 1999; Sundaram and Webster 2000; Cassell, Nakano et al., 2001; Payrató
2009). These may include visual cues (such as facial expression, posture), paralingual factors
(such as the use of voice), touch, distance (proxemics/chronemics), and environmental conditions (e.g., the context) (Cassell, Nakano et al., 2001; Payrató, 2009). Contemporary perspectives assert that non-verbal cues are a fundamental component of the functional competence of an individual’s communicative ability, independent of verbal cues (Payrató,
2009). Consistent with this, estimates have shown that as little as 7% of any message is conveyed through verbal means, with one study demonstrating 55% of a message conveyed through body language and 38% through paralingual cues (vocal tone) (Mehrabian & Ferris,
1967, Mehrabian & Wiener, 1967). Although other estimates have varied from these proportions (Argyle, Salter et al., 1970, Hsee, Hatfield et al., 1992), even lower estimates (for the contribution of non-verbal cues) reflect the importance of non-verbal cues in communication. As such, the restriction of these cues in text compared to richer-media can hinder or obstruct message transmission, such as when conveying mixed-emotions that might be typically conveyed by facial expression (Knapp, 2013).
The restriction of non-verbal cues could have a range of consequences for moral phenomena in text-based stimuli or communication versus cue-rich stimuli. For example, the absence of visual cues in text, relative to cue-rich media, can create anonymity or reduce identifiability of actors (Christopherson, 2007). Anonymity can lead to actors to become deindividuated and disinhibited, allowing people to dissociate their real identities from their online personas (Morio & Buchholz 2009) and (potentially) disengage from their moral values (Bandura, 1999; Diener, 1977). Likewise, person perception is changed as a function
Chapter 1: General Introduction 5
of the absence of non-verbal cues in text (Hancock & Dunham, 2001; Walther, 1993). The filtering out of non-verbal cues in text means that people have less information to form an impression and so compensate for the absence of non-verbal cues by exaggerating available verbal cues (Hancock & Dunham, 2001; Walther, 1993). Given that the morality dimension of person perception is central to impression formation (Wojciszke, 2005), medium effects on person perception may also lead to exaggerated perceptions of moral traits (e.g., trust, honesty, friendliness, kindness).
Empathy is one further morally relevant construct that could be impacted by medium and the presence or absence of non-verbal cues. Evolutionary accounts of empathy posit that non-verbal cues, such as facial expressions, have a special capacity to communicate emotion
(Decety & Jackson, 2004). According to embodied cognition theorists, viewing emotional facial expressions can lead to spontaneous facial mimicry, recreating corresponding feelings in the perceiver that aid in emotion recognition and generate feelings of empathy (Adolphs,
2002; Niedenthal, 2007; Niedenthal, Barsalou, Winkielman, Krauth-Gruber, & Ric, 2005).
Thus, emotional experiences conveyed by text may obstruct emotional recognition and the experience of empathy, compared to emotional experiences that are conveyed in a media that includes visual cues. Indeed, Pfeil and Zaphiris (2007) illustrated a number of clear distinctions between FtF and online (a primarily text medium) in emotional communication.
Considering these examples of how medium and non-verbal cues could impact morally relevant variables, together with the many examples of medium effects from other domains of psychology, it’s reasonable to expect that medium of communication or presentation of a stimulus will have effects in the moral domain. This thesis will explore this topic in two streams: first, I use theories of computer-mediated-communication (CMC) to examine the role of medium in the context of communication. Second, I use theories of the
Chapter 1: General Introduction 6
moral domain to examine the role of medium in the context of presenting a stimulus in an empirical study.
Chapter 2: Cues Filtered Out Theory 7
Stream 1. The Effects of Media on Moral Behaviour
Chapter 2: Cues Filtered Out Theory
To begin to examine how medium impacts moral outcomes, I consider one specific example of (a typically) text-restricted media: computer-mediated-communication (CMC).
CMC has a large body of literature to draw from as a framework for examining medium effects on morally relevant behavioural phenomena. CMC is also an example of a text- restricted medium that is widely used in real-world situations, thus an increased understanding of CMC effects will have important implications for online interactions.
Finally, there is already evidence that the use of CMC compared to non-text media gives rise to a number of moral behaviours. On the one hand, research shows that the use of CMC is associated with ‘flaming’ or hostile, aggressive and anti-social behaviour (Chui, 2014;
Douglas, 2008; Joinson, 2007; Kiesler, Zubrow, Moses, & Geller, 1985; Suler, 2004); whereas on the other hand, field studies of online interactions have found levels of interpersonal intimacy and self-disclosure can exceed that of comparative FtF interactions
(Jiang, Bazarova, & Hancock, 2011; Joinson, Reips, Buchanan, & Schofield, 2010; Walther,
1996; Walther, & Parks, 2002; Walther, Anderson, & Park, 1994). Taken together, these qualities make CMC a useful context to begin exploring medium effects on moral psychology.
CMC refers to any interaction in which two or more electronic devices are used as an intermediary (Walther, 1992). CMC is typically restricted to text communication, although video conferencing tools (such as Skype) also fall within this definition. For the purposes of this thesis, CMC refers to text-based communication that utilises computer(s) to mediate the interaction. Although introduced to the mainstream in the 90s, CMC has risen in popularity in the last decade and has now become a dominant and pervasive means for communication.
For example, between 2012 and 2013 74.8% of all US households have access to the internet
Chapter 2: Cues Filtered Out Theory 8
in their home and 63% of the population of the European Union are accessing daily (United
States Census Bureau, 2012; Eurostat, 2013). For many researchers, and general population alike, the main cause for concern in this shift from face-to-face (FtF) communication to CMC is that it removes or diminishes most non-verbal aspects of communication, thereby, limiting users to written text and providing anonymity (Derks, Fischer et al., 2008; Nguyen, Bin et al.,
2012; Tanis & Postmes, 2003; Walther, 1996, 2002). Indeed, it is no longer necessary to physically interact with our conversational partners. In fact, many may never actually meet those they engage with online. As the frequency of CMC has increased over the last decade, so has research attention: a number of theories have attempted to account different behavioural phenomena that result from CMC compared to other media for communication.
Cues-Filtered-Out Theories Most theories in the CMC literature have developed with a focus on the filtering out of social and contextual cues when communicating via text compared to FtF (or other media).
Collectively, these theoretical approaches have been described as the cues filtered out (CFO) perspective. The unifying theme central to these approaches is that there is a reduction of non-verbal social and relational cues in CMC that produces a depersonalised form of communication, decreased awareness of others, and inhibited interpersonal relations
(compared to FtF, see Walther, 1996 for a review). Each theory under this perspective has subtle differences in their proposed mechanisms for how the absence of cues impacts behaviour. For example, the reduced cues theory predicts that the absence of cues obstructs the communication of behavioural norms and social feedback, leading to less fear of negative evaluation and disinhibition (Culnan & Markus, 1987). The social identity and deindividuation effects model argues that the absence of visual cues in CMC emphasizes people’s social identity and minimises their personal identity (Lea, Spears, & de Groot,
2001). As the shared social identity is salient in cue-poor contexts (i.e., CMC), the actor is
Chapter 2: Cues Filtered Out Theory 9
more likely to behave in ways consistent with the group’s norms, compared to cue-rich context (e.g., FtF).
However, these theoretical approaches are largely born of early studies that demonstrated an increased tendency for people to aggress in CMC (specifically, email) compared to FtF interaction. As a result, these theories struggle to account for the more recently identified, positive medium effects on behaviour. That is, while some CMC interactions may ultimately lead to aggression, there are also countless highly intimate, warm and positive interactions that occur online (Derks et al., 2008; Walther, 1996; Whitty & Carr,
2006). CMC may paradoxically result in ‘hyperpersonal’ interactions that match or surpass
FtF in feelings of intimacy and self-disclosure, (elaborated on in Chapter 4) (Hammick &
Lee, 2014; Jiang et al., 2011; Joinson, 2001; Nguyen, Bin, & Campbell, 2012; Stritzke,
Nguyen, & Durkin, 2004; Walther, 1996, 2007).
In the first stream of this thesis, I attempt to provide a unified account of how, and when, these various moral phenomena occur in CMC. Specifically, I draw together a set of claims from the CFO approach, along with theorising about the role of humanness, as a way of accounting for both positive and negative effects of medium on moral behaviour. I suggest that communicating online results in changes to humanness perception for both the target and the self (i.e., changes to how human we see others and ourselves). Further, I hypothesize that these changes underpin both positive and negative moral behaviours online.
Chapter 3: Computer-Mediated Communication and Aggression 10
Chapter 3: Computer-Mediated Communication and Aggression The notion that computer technology promotes uninhibited, anti-social behaviour
(referred to as ‘flaming’) is widespread and enduring in both academic and public spheres
(Jane, 2015; Joinson, 2007; Stein, 2017; Suler, 2004). Although definitions and terms vary, flaming (also referred to as cyber-incivility, e-bile, cyber-bullying or toxic-disinhibition) is generally defined as hostile, aggressive verbal behaviour that occurs online (Jane, 2015; Lea,
O'Shea, Fung, & Spears, 1992; O’Sullivan, & Flanagin, 2003; Suler, 2004). Victims of flaming can experience serious physical and emotional distress that can impact online behaviour (e.g. developing a fear of online interactions), create feelings of injustice, and negatively affect their offline relationships and behaviour (Moor, Heuvelman, & Verleur,
2010; Park, Fritz, & Jex, 2015; Yuin, 2006). For example, when flaming occurs in the context of work, victims are more likely to quit their jobs (Lim & Teo, 2009). Despite the size and scope of the problem, psychologists have yet to fully articulate the psychological mechanisms responsible for flaming (or more precisely, why aggression is allegedly more pronounced online).
The literature is contradictory and inconsistent about the nature of the exact effect of communication medium on aggressive behaviour. While some researchers have argued that
CMC promotes aggressive behaviour (e.g., flaming) due to the absence of vital social and contextual cues (Abrams, 2003; Moor et al., 2010), others claim that aggression occurs at a similar frequency online and in FtF interaction (Lea, O'Shea, Fung, & Spears, 1992; Reid &
Reid, 2005). Many of these studies do not compare CMC to other media in facilitating aggression, limiting conclusions that can be inferred about the role of medium and cues in facilitating aggression (Abrams, 2003; Aiken & Waller, 2000; Alonzo & Aiken, 2004;
Lapidot-Lefler & Barak, 2012; Molaei, 2014; Mungeam, 2011; Ng & Detenber, 2005). Chapter 3: Computer-Mediated Communication and Aggression 11
Importantly, without comparing CMC to other media, it is not clear whether CMC leads to more aggression than richer media.
Few studies have experimentally compared the frequency of aggression in CMC to other media. Three very early studies outlined in Kiesler, Siegel, & McGuire (1984) compared levels of hostile behaviour across four conditions: FtF, anonymous computer conferencing, non-anonymous computer conferencing, and email. In each experiment, the
CMC conditions had more hostility than the FtF condition., consistent with the CFO perspective. However, more recent work has been critical of these early studies’ methodology, for example, Lea, O'Shea, Fung, and Spears (1992) critique the reliability of subjective estimations of uninhibited behaviour reported by users in some studies and
Walther (1996) is critical of the ecological validity of these laboratory studies. To my knowledge, only one recent study has tested the effect of communication medium on the frequency of aggression by comparing CMC to another, non-text medium. Castellá et al.
(2000) required participants to complete, in groups, a ‘moon landing’ ordering task. This task requires participants to imagine that they have crash landed on the moon and (as a group) rank the importance of a list of survival items. Groups were randomly assigned to communicate either via CMC, video-conference, or FtF. Castellá et al. (2000) found that there was a significant decrease in the prevalence of aggressive behaviour with increased richness of the communication channel (CMC, Video, FtF). This result is consistent with the
CFO perspective, that is, there is a higher number of occurrences of aggressive behaviour in
CMC contexts compared to other media and that this effect is, in part, due to the absence of certain non-verbal cues in CMC.
Also relevant, Lapidot-Lefler and Barak (2012) did not directly compare CMC to other media for communication but instead assessed how certain social and contextual cues typically absent in CMC affect aggression. The authors instructed dyads to debate whom of Chapter 3: Computer-Mediated Communication and Aggression 12 the two should receive a lifesaving drug. All dyads communicated by CMC, however, the authors manipulated identifiability (via the presence or absence of their name in the chat window), visibility (via the presence or absence of a webcam to the participant’s side), or eye-contact (via the presence or absence of a webcam mounted at eye level). Participants were assigned to one of these eight experimental conditions. The authors found that the absence of eye contact was the largest contributor to the presence of flaming. The results of this study suggest that non-verbal cues (i.e., eye contact - a visual cue) absent in CMC but present in some richer media facilitates aggression. However, inferences about how presentation medium impact aggression are limited because the study did not compare CMC to another communication medium.
These more contemporary studies, however, did not examine the mechanism by which the absence of such cues leads to an increase in flaming in CMC compared to the other communication media. Further, both studies were conducted in a very specific context: group problem solving. Given that group interaction is associated with a different set of psychological processes, compared to dyadic interaction (Abrams & Hogg, 2006), it’s not clear that the results of this study would generalise to other CMC contexts.
Given the lack of clarity in the literature regarding medium effects on aggression and, in particular, the limited number of studies that have compared aggression in CMC to a non- text medium, the current study will provide an additional test of the contested, basic effect of medium on aggression. Building on limitations of previous tests of the effect of medium on aggression, this study will utilise an experimental paradigm that has high ecological validity for contemporary CMC use (detailed in methods). Further, I present and test a possible mechanism that explains how the reduced social and contextual cues in CMC increase the tendency to flame by extending and adapting the Cues-Filtered-Out Model (Culnan &
Markus, 1987; Sproull & Kiesler, 1986). Chapter 3: Computer-Mediated Communication and Aggression 13
Social and Human Presence Despite disagreement about the causes of behavioural change in CMC, it is widely accepted that there are fewer social and contextual cues in CMC than in FtF and most researchers agree that this factor is pivotal in explaining online behaviour (Gunawardena,
1995; Lapidot-Lefler & Barak, 2012; Sproull & Kiesler, 1986; Tanis & Postmes, 2003).
Central to this line of reasoning, and discussed most prominently by the CFO model, is the idea that the absence of (or decrease in) non-verbal cues (a) decreases perceptions of social presence and (b) increases perceptions of psychological distance. Perceived social presence is the awareness that the other person is both present and ‘real’, while perceived psychological distance is the experience of things as a part of immediate reality (as opposed to a distal one) (Gunawardena & Zittle, 1997; Kehrwald, 2008; Tu, 2002; Tu &
McIsaac, 2002). The cues that signal social presence and close psychological distance (such as facial expression, body language and voice tone) are often absent in CMC. As a result, the communicative partner is less salient during CMC compared to other communication media
(Burgoon et al., 2002; Tu, 2002). It has further been argued that this lowered salience disinhibits behaviour, leading to both anti-social, flaming behaviour (Joinson, 2001; Lapidot-
Lefler & Barak, 2012) and prosocial behaviours, such as excessive self-disclosure (Joinson,
2001; Lee & Wagner, 2002).
I explore the possibility that decreased salience of others licenses anti-social behaviour because it impacts humanness perceptions. Several researchers have noted that the salience of others is important for perceptions of humanness and attributions of mind. For example, Opotow (1990) argued that psychological distance is a form of moral exclusion that permits dehumanisation, while Trope and Liberman (2003) proposed that increased psychological distance (or reduced salience) promotes simpler, ‘colder’ and more abstract perceptions of the other person, perceptions that are comparative to mechanistic dehumanisation (Haslam, 2006; see below). Chapter 3: Computer-Mediated Communication and Aggression 14
This work suggests that the absence of non-verbal cues may not only decrease social presence but human presence (the quality of being a human moral agent or patient). In other words, in CMC, one is less aware of interacting with a human moral agent. During FtF communication there are certain non-verbal cues, such as voice, movement or facial expressions, which have been shown to demonstrate personhood and moral traits (DiSalvo,
Gemperle, Forlizzi, & Kiesler, 2002; Eyssel, Kuchenbrandt, & Bobinger, 2011; Nowak,
2004; Nowak & Biocca, 2003; Nowak & Rauh, 2005; Thompson, Trafton, & McKnight,
2011; Wexelblat, 1998; Zhang, Liu, Jin, & Liu, 2010). As CMC is largely restricted to text, many of these non-verbal cues are absent. Therefore, the absence of cues (such as in CMC) may facilitate dehumanisation. Indeed, some studies have found that experimentally introducing non-verbal cues (e.g. face or expression, voice intonation) to online interactions with computers results in inflated perceptions of humanness and ratings of morally relevant traits, including trustworthiness, likeability and attractiveness, as compared to text only conditions (Koda & Maes, 1996; Nowak, 2004; Nowak & Biocca, 2003; Nowak & Rauh,
2005; Wexelblat, 1998). Most convincingly, Schroeder and Epley (2016) demonstrated consistently (across eight studies) that when participants listened to a passage that was read aloud, rather than simply reading a text rendering, they rated the author as more thoughtful, mindful, emotional and fundamentally human. Participants were asked to either listen to a speaker or read a transcript of the speech (text only) and rate the target on a range of human perception (e.g., extent to which they are competent, thoughtful, likeable, emotional) and mind perception measures (e.g., capacity for self-control, remembering, planning, capacity, and experience pleasure/pain). Observers who listened to speakers talk about emotional experiences or decisions rated the speaker as more human (specifically, more able to think and feel) than those participants who only read the speech transcript. In a later study, the effect of audio on humanness ratings was fully mediated by pitch variance or intonation. Chapter 3: Computer-Mediated Communication and Aggression 15
However, Schroeder and Epley (2016) did not explore downstream effects of medium- induced changes to humanness.
What remains unclear from the findings Schroeder and Epley (2016) is whether dehumanisation in text interactions is associated with changes to morally relevant judgments and behaviours. Evidence from Bastian et al. (2011) illustrates that lowered perceptions of humanness are associated with lower moral standing. In this work, measurements of humanness were drawn from Haslam’s (2006) two sense of humanness: 1) human nature
(HN) which describes characteristics that differentiate humans from robots, including the capacity to feel, warmth, or emotionality, and 2) human uniqueness (HU) which describes characteristics that differentiate humans from animals, such as the capacity to think, rationality, and culture. In study one, participants were asked to imagine members from a range of social groups (e.g. Muslim, lawyer) engaging in a number of pre-determined, morally relevant behaviours. Participants then made judgments of HU and HN and ascribed correctional strategies for each of the social groups. In support of the authors’ main hypothesis, the extent to which each social category was seen to possess HU traits was positively correlated with judgments of blameworthiness and moral responsibility. While perceptions of HN traits for each of the social categories was positively correlated with moral praise and patiency (capacity to be a recipient of moral behaviour). Relative to CMC, if individuals are generally perceived as less human when online, then according to Bastian et al. (2011), they are likely to be allocated lower moral status.
The lowering of moral status allows for moral disengagement and the potential for disinhibited behaviours. As suggested by Bandura (2002), the denial of moral agency or patiency allows people to disengage from moral self-sanctions. If targets are perceived to have less HN, with less associated moral patiency, then they can be excluded from normative moral treatment. Although the process of moral disengagement is relatively unstudied in Chapter 3: Computer-Mediated Communication and Aggression 16
CMC, there is some evidence that disengagement may occur in similar ways in both online and offline contexts. In very specific online contexts (not CMC), dehumanisation and moral disengagement have been associated with aggressive behaviour (Bastian, Jetten, & Radke,
2012; Greitemeyer & McLatchie, 2011). Specifically, playing violent video games leads to dehumanisation of the self and the victim, moral disengagement, and, in turn, evokes violent behaviour. Hence, we may expect a similar effect of dehumanisation and moral disengagement for CMC specific disinhibited behaviours, including flaming. We may further predict that moral disengagement and aggression may be primarily associated with dehumanisation of HN (rather than HU). According to Bastian et al., (2010, 2011) targets that are perceived to be low in HN are also perceived to have less moral patiency (less capacity to be harmed) thereby permitting anti-social behaviour.
There is some evidence that qualities related to humanness are reduced in CMC and lead to less cooperative behaviours. The use of CMC when completing a prisoner’s dilemma has been linked to less cooperative behaviour, reduced ratings in intelligence (a quality related to HU) and increased levels of deception (Jensen, Farnham, Drucker, & Kollock,
2000; Kiesler, Sproull, & Waters, 1996). These results suggest that CMC may facilitate dehumanisation (although this was not directly measured) leading to less cooperative behaviour. Relatedly, Parise, Kiesler, Sproull, and Waters (1999) show that adding human- like qualities to a computer interface (specifically, a humanoid avatar) facilitates cooperation with the computer agent, compared to a computer agent with fewer human qualities
(specifically, a dog avatar). Thus, illustrating that the presence of humanness cues in text interactions is relevant to cooperative behaviour. However, this study only addressed how experimentally included humanness cues in CMC affects perceptions of/behaviour toward non-human agents; what remains unclear if changes to humanness perception in text interactions affects behaviour toward human-agents. Chapter 3: Computer-Mediated Communication and Aggression 17
Taking together the findings of Schroeder and Epley (2016), Bastian and colleagues
(2010, 2011, 2012, 2013), Jensen, Farnham, Drucker, & Kollock, (2000), and Kiesler,
Sproull, & Waters (1996), it is reasonable to expect that CMC interactions may lead to
dehumanisation of a human target, moral disengagement, and increased anti-social behaviour.
Therefore, using the cues-filtered-out account as a theoretical framework, I hypothesize that:
in a CMC condition compared to a voice condition, there will be: (H1) dehumanisation (or
the denial of uniquely human or human nature qualities), (H2) which will lead to moral
disengagement (or a reduction in the perceived moral patiency or agency of others), and (H3)
an increased tendency to flame. This hypothesized mechanism is summarised below, in
Figure 1:
Figure 1. Hypothesized mechanism for the effect of communication medium on flaming.
Dehumanising of the Self Online CMC might not only facilitate aggression via modifying perceptions of others, but
may also trigger a range of self-focused processes that also increase moral disengagement.
More specifically, the filtering out of cues in CMC may not only alter perceptions of the
humanness of the other but may also alter perceptions of one’s own HU and HN. The absence
of cues has been shown by a number of researchers to alter both private and public self-
awareness and consequently change self-perception (Joinson, 2001; Sassenberg, Boos, & Chapter 3: Computer-Mediated Communication and Aggression 18
Rabung, 2005; Vasalou, Joinson, & Pitt, 2007; Yao & Flanagin, 2006). Most prominently, the social identity deindividuation effect (SIDE) model argues that the absence of cues in
CMC depersonalizes the self, resulting in a shift in self-perception such that one’s group identity becomes more salient than one’s individual identity (Lea, Spears, & de Groot, 2001).
Other research groups have found that use of an avatar is another way in which CMC can alter self-representation (Fox, Bailenson, & Tricase, 2013; Peña, Hancock, & Merola, 2009;
Yee & Bailenson, 2007; Yee, Bailenson, & Ducheneaut, 2009). Taken together, these examples suggest that CMC has the potential to change self-perception.
What has not yet been considered within this literature is whether a) these established changes to the perception of the self during CMC include changes to the perception of one’s own humanness, and b) if such changes contribute to anti-social online behaviours.
Perceptions about our own humanness and changes to these perceptions, are linked to morally relevant behaviours. Research has established that when people engage in immoral behaviour, they not only dehumanise others but also themselves. Bastian et al. (2012) first demonstrated that engaging in immoral behaviour in the form of video game violence results in dehumanisation of the self, a finding that is particularly applicable to this study as video games are comparable to online communication in their (limited) number of social cues. In a follow-up series of studies, Bastian et al. (2013) extended this finding to confirm that self- dehumanisation by the perpetrator occurs in response to their own interpersonal, moral transgressions. Drawing from self-perception theory (Bem, 1972), Bastian et al. (2013) argue that individuals infer a lack of humanity from observations of their own, unjustified immoral behaviour. However, as the causal direction is not definitively established by this study, one further possibility that warrants study is whether attributing less HU to oneself diminishes the associated responsibility to behave as a moral agent, consequently facilitating immoral behaviour. Therefore, the current study will assess the hypothesis (H4) that CMC leads to a Chapter 3: Computer-Mediated Communication and Aggression 19 reduction in self-HU, which in turn leads to moral disengagement and increased tendency to flame.
Study Overview This was a between-subjects study in which communication medium was manipulated
(text or voice) and aggression was the primary dependent measure. Participants either read text or listened to a pre-recorded audio track that described an opposing political opinion.
Participants were led to believe (falsely) that the author had been an earlier participant in the study who had shared his or her opinion on the study’s discussion forum. The participant was then asked to respond to the author via the “study’s discussion forum” (the flaming measure).
Finally, the participant rated both the author and themselves on a range person perception measures (including humanness). While both conditions are variations of CMC, this study will focus explicitly on the role of restriction to text.
This study involved two parts: first, the development and piloting of the voice and text stimuli, second, the main study.
Stimuli Development The voice and text stimuli were carefully developed both to maximise ecological validity and to ensure consistency across text and voice to minimise confounds. The stimuli contained political opinions on the topic of same-sex marriage, a contentious social issue at the time of data collection (2015).
Six real opinions (3 for same sex marriage, 3 against same sex marriage) were first collected from either politically conservative (e.g. The Blaze, TFP Student Action) or politically liberal (civilliberty.com, policy.mic, Facebook) discussion forums. During presentation to participants in the pilot, all six opinions were presented on the same, nondescript forum page. The author for each opinion was also held constant and was Chapter 3: Computer-Mediated Communication and Aggression 20 ambiguously named “TFP student”, with a default forum avatar (see Figure 2 for an example).
Figure 2. One of the stimuli used in the text condition Next, voice versions for each of the six opinions were developed. The voice stimuli were crowd-sourced using Amazon’s Mechanical Turk (AMT). Eight participants were required to record themselves voicing all 6 opinions; they were instructed to:
Imagine that you are the person who wrote the essay. We want you to imbue your words with all of the thoughts, emotions, and substance that the writer him/herself felt. We want you to read it as if you were actually coming up with the lines naturally off the top of Chapter 3: Computer-Mediated Communication and Aggression 21 your head rather than reading from an essay. We want you to speak as naturally as you would if you were in the midst of a real conversation.
Eight (4 female, 4 male) complete sets of voices were collected (each set contains 6 recordings resulting in a total of 48 stimuli).
I then collected humanness ratings for these voice recordings (detailed below). The male and female voice with the highest humanness rating (averaged across their set of 6 recordings) was used in the main study. Voices that received high humanness ratings were because, presumably, those actors spoke most naturally and most successfully imbued emotion in their speaking. Therefore, voices with high humanness ratings will give the study the best opportunity to test if voice (compared to text) has a humanising effect. By intentionally selecting voices that make the manipulation as strong as possible, then we can have reasonable grounds to assume that no effect exists, if no effect emerges.
Voice Pilot Test The humanness of each voice was assessed by an independent sample of 86 (37 female, age: M=31.44, SD=8.03) participants from Mechanical Turk. The data from this pilot also assessed the claim by Schroeder and Epley (2016) that voice tone is the most important paralinguistic cue for perceptions of humanness.
Each participant rated three random voice clips from the total of 48 potential voice stimuli. This resulted in 5 ratings, on average, per clip. Participants were instructed to listen to each clip and rate each speaker’s humanness. Participants rated each speaker on humanness (1 = much less than the average person, 7 = much more than the average person), their agreement or disagreement with the stated political opinion (1 = strongly disagree, 7 = strongly agree), as well as their general beliefs on gay marriage and demographics (political orientation, life satisfaction, income). Chapter 3: Computer-Mediated Communication and Aggression 22
The pitch variance for each voice sample was extracted using Praat software; technical problems with the recording of one clip meant that pitch variance could not be extracted and so this clip was omitted from the analyses (Boersma & Weenink, 2014). An average humanness rating was then calculated for each voice clip (M=4.38, SD=0.63) and for each speaker overall.
The hypothesis that voice tone predicts humanness ratings was assessed at the level of the voice clip (N = 47). A bivariate correlation between the humanness ratings and pitch variance revealed that pitch variance was significantly, moderately correlated with humanness ratings (r(46) = .33, p = .024). This result is consistent with the claim that pitch
(often) variance conveys humanness (Schroeder & Epley, 2016).
Next, I selected the voice recordings for use in the main study. This analysis was conducted at the level of speaker (N = 8). The male and female speaker with the highest humanness ratings, averaged across their 6 recordings, was then used in the main study. One male and one female were selected to ensure results were generalizable to both genders.
Studies suggest that people can make stereotyped judgements about the masculinity or femininity of the speaker based on voice (Ko, Judd, & Blair, 2016). Given that gender stereotypes are associated with qualities relevant to humanness and morality, such as women and warmth (Fiske, Cuddy, Glick, & Xu, 2002), it is reasonable to expect differences in humanness and morality judgements across genders.
Consequently, the pool of stimuli used in the main study included: six text stimuli
(three opinions for same-sex marriage, three opinions against same-sex marriage) and 12 voice stimuli (one male and one female audio version of each of the six text transcripts). Chapter 3: Computer-Mediated Communication and Aggression 23
Main Study Methods
Participants Participants were 395 Amazon’s Mechanical Turk workers, who participated in exchange for a small monetary reward. According to power analyses of an independent, two- tailed t-test, a sample size of 350 provides adequate power (80%) to detect a small-medium effect (d = 0.3; derived from the most conservative effect in Schroeder & Epley, 2016) with
α = .05, two-tailed. I over-recruited to account for the exclusion of participants due to inattentive responding.
Twenty-one participants were removed for unengaged responding: one for minimal variation in responses (< 1 SD across responses), five for submitting the page containing the stimuli in <30 seconds, 12 for failing attention checks. This left a final sample of 179 participants in the voice condition and 198 in the text condition.
The final sample was predominantly liberal, 49% of participants identified as (in general) liberal, 25% as conservative, and 26% as neither conservative or liberal. The sample was 63 % socially liberal (identified as either slightly, moderately, or strongly liberal on social issues), 21% of the sample identified as (slightly, moderately, or strongly) conservative, and 16% were neither conservative or liberal. Economic political orientation was evenly distributed: 42% identified as (slightly, moderately, or strongly) liberal on economic issues, 41% identified as (slightly, moderately, or strongly) conservative on economic issues, and 17% identified as neither conservative nor liberal. Their ages ranged from 13-69 (M=35.90, SD=10.91) and the sample consisted of 51% females. The average household income was $35,000 to $50,00 and ranged from under $15,000 to more than
$150,000. The average life satisfaction was the scale centre, “neither satisfied nor dissatisfied”. There were no significant differences in overall or economic political Chapter 3: Computer-Mediated Communication and Aggression 24 orientation between the conditions (2s < 4.71, df=2, ps>0.09), although there were slightly more social liberals and moderates in the text condition (2 < 7.19, df=2, p = 0.03).
Materials and Procedue
Pre-Screening of Political Opinions Participants were first asked their opinions on a number of U.S. political issues.
Participants were asked if they oppose or support: (1) affirmative action, (2) same sex marriage (3) restrictions on gun ownership. Responses were given on a six-point bipolar
Likert scale, ranging from 1 (strongly oppose) to 6 (strongly support), (M = 4.76, SD = 1.83).
Responses to the issue of same sex marriage determined whether the participant was presented with the stimulus that was against or for same sex marriage (see Stimulus
Presentation, below). This ensured that the participant was presented with a message that opposed their views.
Attitude Strength For each political issue, participants then answered six items assessing a potential moderator of aggression, attitude strength. The six items assess a range of factors related to attitude strength (Howe & Krosnick, 2017), including, attitude importance (e.g., how important is this issue to you personally?), certainty (e.g., how sure are you that your opinion on this issue is right?), knowledge volume (e.g., how knowledgeable are you on gay marriage in the United States?), moral conviction (e.g., how representative of your values is your attitude toward gay marriage in the United States?”). Responses to each item were measured with fully-labelled 5-point Likert scales.
The six items relating to participants’ same-sex marriage opinions were then averaged to form a measure of attitude strength for same-sex marriage (ɑ = .67; M = 2.76, SD
= 0.65). Chapter 3: Computer-Mediated Communication and Aggression 25
Stimulus Presentation The participant’s attitude toward same-sex marriage determined whether they were presented with a ‘pro’ or ‘against’ same sex marriage opinion. If the participant was opposed to gay marriage (denoted by responses: 1 [strongly oppose], 2 [somewhat oppose], 3 [slightly oppose]) then they were randomly presented with one of three possible ‘pro’ same sex marriage opinions. Conversely, if the participant was supportive of same sex marriage
(denoted by the responses: 4 [slightly support], 5 [somewhat support], 6 [strongly support]) then they were randomly presented with one of three possible ‘against’ same sex marriage opinions. Participants were also randomly assigned to either a text-only or a voice condition.
Thus, this opinion was either presented in a text-only or spoken format. The process of allocating people to each stimulus (i.e., opinion) is summarised in Figure 3.
Figure 3. The procedure of allocating participants to each stimulus and number of participants per stimulus. Pro refers to those against same-sex marriage (in their opinion) and con refers to those that were supporting same-sex marriage.
Note. Participants that are against same-sex marriage would be presented with the ‘con’ stimulus and vice versa. Each opinion is numbered (1 to 6), there are three different opinions for each of the pro or con positions on same-sex marriage. Chapter 3: Computer-Mediated Communication and Aggression 26
On completing the attitude items, participants were told:
In an earlier phase of this study we had participants, similar to yourself,
elaborate on the social and political opinions that they had disclosed to us. With their
permission, these opinions have been posted to our study's online discussion forum so
that we may explore the public opinion on these issues in more depth. Clicking to the
next page will randomly generate one of these opinions. Think critically about the
opinion of this individual as you will be asked to make a response (anonymously) to
add to our study's discussion forum.
Participants were then required to either read or listen to a passage describing another
‘participant’s’ opposing opinion on the topic of same sex marriage.
After reading or listening to the opinion, participants were first asked an attention check question: ‘what was the topic of the recording you just listened to?’ (1) Affirmative
Action, (2) Gun Control, (3) Gay Marriage, (4) Immigration, (5) Taxation. Any participant that did not select (3) was excluded from all further analyses (N =12).
Flaming Measure Next, each participant completed a flaming measure by selecting a response, from a list of six pre-determined options, which they would like to post as a comment on the article.
Specifically, they were instructed: “You now have the opportunity to post a reply anonymously; select which of the options you would most like to post as a response.” This measurement of flaming was derived from an intention to flame measure developed by
Hutchens, Cicchirillo, and Hmielowski, (2014). Flaming behaviours as defined by Hutchens,
Cicchirillo, and Hmielowski, (2014) (e.g., responding in all caps, swearing), were ordered from neutral to strongly negative (flaming) to form a response scale. Example statements applicable to the context of a debate on same-sex marriage that included these flaming behaviours were created and included in the response scale. Example statements were derived Chapter 3: Computer-Mediated Communication and Aggression 27 from real responses to the opinions on discussion forums that were used to develop the stimuli. Scale items were then coded from 1 (less flaming) to 6 (more flaming) to generate a continuous measure of flaming (M = 2.45, SD = 1.51).
Participants received different response sets according to whether they were exposed to messages for or against same sex marriage. Response sets were conceptually equivalent apart from whether statements were in support of or are opposed to gay marriage action, see
Table 1 for the response sets.
Table 1. Flaming Response Scales for Participants for or Against Same-Sex Marriage
Flaming Politically liberal response scale Politically conservative Intensity response scale
1 You would prefer to ignore the message and not Equivalent
make any response at all
2 A polite statement that you disagree, e.g., I Equivalent
understand that you have a different view, but I
disagree…
3 A more emphasized, passionate disagreement Equivalent
e.g., I disagree VERY strongly with this...
4 An argumentative response e.g., You are wrong An argumentative response
because marriage isn’t a purely religious e.g., You are wrong because
institution and others should not impose their this violates the holy union
religious beliefs upon it… of marriage between a man
and a woman.... Chapter 3: Computer-Mediated Communication and Aggression 28
5 An angry response e.g., this is the opinion of a An angry response e.g.,
narrow minded, bigoted individual… Anyone who supports gay
marriage supports sin and so
is a sinner themselves....
6 A very angered and passionate response e.g., A very angered and
you are a f*king idiot if you believe a child passionate response: e.g.,
cannot be raised equally happily by a same-sex you are a f*king idiot if you
couple... don't think this is law is a
moral violation that has
serious implications for us
all....
Participants were then asked to rate the author/speaker of the opinion and themselves on a series of measures assessing aspects of person and self-perception: humanness (Haslam et al., 2005) and moral status (Bastian et al., 2011).
Humanness Perception Perceptions of humanness were measured with two subscales: Human Uniqueness
(HU, six items α = .83), Human Nature (HN, eight items, α = .80) (Haslam et al., 2005). All items were measured on a 7-point Likert scale from 1 (much less than the average person) to
7 (much more than the average person). Items relevant to HU included whether the author/speaker was: (1) refined and cultured; (2) rational and logical; (3) lacks self-restraint
(reverse-scored); (4) unsophisticated (reverse-scored); (5) like an adult, not a child; (6) seemed less than human, like an animal (reverse-scored). Items measuring HN were whether the author/speaker is: (1) open-minded; (2) emotional; (3) responsive; (4) warm; (5) Chapter 3: Computer-Mediated Communication and Aggression 29 superficial; (6) lacking depth (reverse-scored); (7) mechanical and cold, like a robot (reverse- scored); (8) like an object, not a human (reverse-scored). Although the subscales were moderately correlated (r = .68), I treated these scales as distinct for these analyses, according to theoretical accounts of humanness (Bastian et al., 2011; Haslam et al., 2005). Item scores were then averaged for each of HN (M = 3.90, SD = 0.93) and HU (M = 3.88, SD = 1.13) for each participant.
Moral Status Next, participants completed measures of moral status of the author/speaker. Moral status was measured with two separate scales; Moral Agency (MA, eight items, α=.80) and
Moral Patiency (MP, four items, α=.79). MA was further distinguished into prescriptive MA
(expected moral behaviour) and prohibitive MA (expected inhibition of immoral behaviour)
(Bastian et al., 2011; Loughnan, Haslam, & Bastian, 2010). For the items relating to prohibitive MA, participants were asked to imagine the target acting in certain immoral ways.
The participants were required to judge the extent to which they would reprimand the target for this hypothetical action. Specifically, to what extent would they ‘hold the target responsible, reprimand, or blame them’. These items include: ‘Making a promise and not keeping it’; ‘Cheating on a significant other and never telling them’; ‘Refusing aid to a parent when they are in need’; ‘Pushing someone out of the way so [they] can be first’. Four moral behaviours captured perceptions of prescriptive MA. Participants imagined the target acting in moral ways and judged how responsible they are for their actions. Specifically, they were asked, ‘to what extent would they praise, respect, or give credit to the target’. Items include:
‘Returning a lost wallet or purse with the money intact’; ‘Supporting parents when they are in need’; ‘Being nice to your co-workers even under stress’; ‘Not cheating on a test even if you have the answers in front of you’. Finally, three maltreatments related to perceptions of the target’s MP. Participants judged how sorry they would feel for the target given certain Chapter 3: Computer-Mediated Communication and Aggression 30 maltreatments occurred. Specifically, they were asked how likely they are to intervene or feel indignation for ‘the target’. Items included: ‘You heard someone bad-mouthing the person behind their back’; ‘You saw someone refusing aid to the person when they really need it’;
‘The person goes out to dinner with a friend who refuses to pitch in for dinner because it’s not their favourite restaurant’. Responses were made for all items on 7-point Likert scales that range from 1 (very unlikely) to 7 (very likely). MA and MP subscales were weakly correlated (r =.34) and were treated separately in the main analyses. Average scores were calculated for each participant for each of the MA (M =3.94, SD = 0.65) and MP (M =5.58,
SD = 1.54) sub-scales.
Self-Humanness Participants then rated their own humanness using the same scale but with the item stems changed to be relevant to the self, e.g. ‘“I felt like I was mechanical and cold, like a robot”. Perceptions of the observer’s own humanness were, again, measured with two scales: self-HU (six items, α=.70, M = 3.91, SD = 0.49), and self-HN (eight items, α=.81, M =3.56,
SD = 0.53). Average scores were calculated for each HN and HU subscale.
Demographics Finally, short demographics were assessed. Participants nominated their household income using a categorical scale from 1 (under $15,000) to 8 (over $150,000), (M = 4.01, SD
= 1.73). Next, I measured life satisfaction on a fully labelled 7-point Likert scale where 1
(completely dissatisfied) to (completely satisfied); (M = 4.93, SD =1.49). Religiosity was measured in two ways: first, participants were asked whether they identified with any religion or denomination with a binary scale (yes or no). Next, participants rated their religiosity.
Participants were asked “to what extent do you identify yourself as a religious person?” and responded with a 5-point Likert scale, 1 (not at all) to 5 (very much so), (M = 2.14, SD =
1.32). Nationality, age, gender, ethnicity, and country of birth were described using open- Chapter 3: Computer-Mediated Communication and Aggression 31 ended responses. Finally, participants identified their political orientation using three items.
Participants were asked “to what extent would you describe yourself on each of the following items… liberal issues/economic issues/in general” and responded using a fully labelled 7- point Likert scale, 1 (strongly liberal) to 7 (strongly conservative).
Results
Pre-Processing There was a larger number of participants for same-sex marriage (N =290) than against same sex marriage (N = 87). This is consistent with liberal biases identified by those who have studied the demographic qualities of the Mechanical Turk population (N > 15,000,
34% democratic, 22% republican, and 26% independent; Huff & Tingley, 2015).
I tested whether (1) viewing the pro same-sex marriage stimuli or the against same- sex marriage stimuli, and (2) the strength of the participant’s attitude toward same-sex marriage impacted the dependent measures. Linear models were fit predicting each of flaming, HN, HU, MA, MP, from attitude strength and orientation (for same-sex marriage =
0, against same-sex marriage = 1). Results are shown in Table 2. There was a significant effect of attitude strength on flaming, HN, and HU, but no significant effects of orientation, nor any significant interactions between orientation and attitude strength. Therefore, the analyses collapse across pro same-sex marriage and against same-sex marriage conditions, but account for the effect of attitude strength.
Chapter 3: Computer-Mediated Communication and Aggression 32
Table 2. Testing for Moderation Effects of Attitude Strength and Attitude Orientation
Dependent Predictor B t-value p-value Variable
Flaming Attitude Strength -.22 -2.67 .007
Orientation .82 1.86 .06
Strength*Orientat .03 0.211 .83 ion
F (3,373) = 6.97, p <.01, R2 = .05
HN Attitude Strength -.22 -2.67 .007
Orientation .82 1.862 .06
Strength*Orientat .03 0.211 .83 ion
F (3,373) = 27.86, p <.01, R2 = .18
HU Attitude Strength -.33 -3.30 .001
Orientation .44 0.84 .40
Strength*Orientat .12 0.64 .52 ion
F (3,373) = 16.87, p <.01, R2 = .11
MA Attitude Strength .04 0.61 .54
Orientation .007 0.02 .98
Strength*Orientat .06 0.56 .57 ion
F (3,373) = 2.13, p = .09, R2 = .008
MP Attitude Strength -.10 -0.73 .47
Orientation .25 0.33 .74
Strength*Orientat .09 0.33 .74 ion
F (3,373) = 2.61, p = .05, R2 = .01
Chapter 3: Computer-Mediated Communication and Aggression 33
To test for whether there was an effect of the speaker’s gender on any of the main dependent measures, t-tests for independent samples were calculated for each of dehumanisation, moral status and flaming. There was no significant difference between the speaker's gender in the level of flaming (t(177) =1.09; p=.28), perceptions of moral status
(averaged across agency and patiency) (t(177)=-1.13; p =.26), or of humanness (averaged across HN/HU) (t(177 ) = -0.92; p =0.36). This indicates that the results of this study are generalizable across genders and consequently, the main analyses collapsed across speaker- gender.
Analysis Overview First, I conducted bivariate correlations to explore the zero-order relationships between variables. I then conducted planned comparisons between the conditions to assess whether the variables vary as a function of condition, as hypothesised. Finally, structural equation modelling tested the hypothesized causal path: lowered perceptions of HN in CMC leads to a reduction in moral status and an increase in tendency to flame, as compared to voice communication.
Exploratory Correlation First, two-tailed Pearson bivariate correlations were run in order to explore the basic association between variables. These results are presented in a correlation matrix in Table 3.
Correlation analyses suggest that flaming is negatively associated with HN, HU, MP (but not
MA), consistent with hypotheses H1 and partially consistent with H2. Inconsistent with H4, correlations also revealed that there was no relationship between self-humanness ratings and flaming.
Chapter 3: Computer-Mediated Communication and Aggression 34
Table 3. Exploratory Pearson Bivariate Correlations Between Measured Variables
HN HU MP MA AS Self-HN Self- HU
Flaming -.35** -.18** -.26** -.04 .20** .08 -.003
HN - .72** .46** .24** -.17** .04 -.29
HU - .40** .23** -.19** .01 -.10
MP - .41** -.04 .12* -.04
MA - .05 -.01 -.01
AS - -.03 .01
Note. *p<.05; **p<.01; (Self-)HN= (self-rated) human nature, Self-HU = (self-rated) human uniqueness, (self-rated), MA = moral agency, MP = moral patiency, AS = attitude strength, condition = communication medium
Communication Medium Comparisons I conducted planned comparisons between two conditions (communication medium:
voice and text), in order to test the hypotheses that there would be more dehumanisation of
the other and the self (for both HN and HU) in the text-only condition than in the voice
condition, and whether there were greater levels of flaming in the text only condition than in
the voice condition.
Flaming. Contrary to H1, there was significantly more flaming when participants
listened to an opinion (M=2.68, SD =1.52) than when they only read the same opinion
(M=2.27, SD=1.48); t(371)=1.61, p=.009, d=.27.
Other-Humanness. Consistent with H2, observers perceived authors to be more
uniquely human when they listened to their opinion (M = 4.1, SD = 1.02) as compared to
reading the opinion (M = 3.64, SD = 1.19), t(371) = 3.77, p < .001, d =.42. However, there
were no significant differences in HN ratings when observers listened to the opinion (M = Chapter 3: Computer-Mediated Communication and Aggression 35
3.64, SD = 0.98) as compared to when they only read the opinion (M = 3.66, SD = 0.97), t(371) = -.089, p= .93, d = .01.
Moral Status. Neither perceptions of MA (Mvoice = 3.95, SDvoice = 0.67; Mtext = 3.92,
SDtext = 0.67), t(371) = -0.963, p =.34, d = .05, nor MP (Mvoice=5.52, SDvoice=1.55; Mtext=5.64,
SDtext = 1.52) differed as a function of the communication medium; t(371) = 0.49, p=.63, d = 0.08.
Self-Humanness. Contrary to H4, there were no significant differences in the observer’s perceptions of their own humanness, both HU (Mvoice = 4.92, SDvoice = 0.80, Mtext =
4.97, SDtext = 0.81) and HN (Mvoice = 5.01, SDvoice = 0.84, Mtext = 5.03, SDtext = 0.81), across communication medium (ts(376) < -0.28, ps >.57).
Structural Equation Modelling
Given the total effect of condition on flaming in the opposite direction to what was expected, I sought to unpack this in relation to the hypothesized indirect effects. Structural equation modelling was then used to test explore the possible role of HU in accounting for
CMC effects on flaming. Specifically, structural equation modelling explored whether lowered perception of humanness leads to a reduction in moral status, moral disengagement and finally an increase in tendency to flame, as compared to voice communication. Attitude strength was included as a control in this model, due to its associations with endogenous variables. The error terms between HN and HU, and between MA and MP were allowed to co-vary. This is an acceptable practice when there is theoretical justification for this co- variation (Jöreskog & Sorbom, 1993), namely the sub-forms of humanness and moral status have been shown to correlate (e.g., Bastian et al., 2011). All models were estimated with
2,000 bootstraps and 95% confidence intervals. Chapter 3: Computer-Mediated Communication and Aggression 36
Path coefficients for the initial model are summarised in the Figure 4. The model fit for this model was outside of recommended fit metrics. This suggests that the hypothesized model is not a good representation of the data. In terms of absolute fit indices: the χ2 was significant (χ2(7) = 45.8, p < .001) the RMSEA is slightly above the recommended threshold of 0.07 (RMSEA = 0.073) (Steiger, 2007), the AGFI is below the recommended value of
0.095 (AGFI = 0.867) (Tabachnick & Fidell, 2007). Finally, for the incremental fit indices the CFI and TLI (NNFI) are both slightly below acceptable range of 0.95 (CFI = 0.930, TLI =
0.789) (Tabachnick & Fidell, 2007).
.440** [.345, .525]
Human Nature Moral Patiency
-.270** [-.379 , -.155] -.001 [-.104, .105]
Communication .124 * [.026, .219] Flaming Medium
.200** [.105, .299] Human Moral Agent .064 [-.037, .168] Uniqueness
.211** [.123, .300] ** p<.01; *p<.05;
Figure 4. Initial, hypothesised model with standardised estimates (95% confidence interval),
2000 iterations of bootstrapping, controlling for the effect of attitude strength on: flaming,
HN, HU, MA, MP. Note. Text is dummy-coded as 0, voice is coded as 1.
A revised model was developed that increased model fit and removed the moral status variables from the model: modification indices revealed that HN is a better (direct) predictor of flaming than either MA or MP (Figure 5). Although both moral status items were associated with humanness items and moral patiency to flaming (see Table 3). Moral status Chapter 3: Computer-Mediated Communication and Aggression 37 was not related either medium nor flaming (see Table 3), and so was removed from this mediation model. HN was retained in the model despite the non-significant relationship with communication medium because the bivariate correlation suggests that HN is a strong predictor of flaming (Table 3).
Human Nature -.001 [-.104, .105] -.408** [-.537, -.273]
.113* [.014, .207] Communication Flaming Medium
.200** [.105, .299] Human -.117 [-.039, .263] Uniqueness
Figure 5. Refined model: HN and HU errors were co-varied, controls for the effect of attitude strength on flaming, HU and HN, all values are estimated with 2000 bootstrapped samples and 95% confidence intervals.
The refined model had excellent fit indices, suggesting that the path model is representative of the data; χ2(1) = 0.522, p = .47, = RMSEA <0.001; AGFI = 0.992, CFI =
1.00, TLI (NFI) = 1.01. The total effect of communication medium on flaming is 0.136 (95%
CI = .039 -.234, p =.01). There is a significant direct effect of communication medium on flaming, such that there is more flaming in the voice condition (direct effect = .113; 95% CI
= .014 - .207; p = .022). There is also a small but significant negative indirect effect of communication medium on flaming through HU (indirect effect = -.038 [-.071, -.014], p =
.002). However, there is no indirect effect of communication medium through HN (indirect effect <0.001 [ -.033, .035], p = .98). According to this model, communicating via text leads to animalistic dehumanisation (or the denial of uniquely human traits), which in turns leads to more flaming. Paradoxically, there is also a direct effect of communication medium on flaming that opposes the indirect effect, such that there was more flaming in the voice Chapter 3: Computer-Mediated Communication and Aggression 38 condition. This direct effect is larger than the indirect effect through HU, manifesting in more flaming in the voice condition than in the text (as shown by the t-tests). According to this model, communication medium does not affect HN, however mechanistic dehumanisation
(the denial of HN) is associated with an increased tendency to flame (independent of medium).
Moderated Mediation Next, I tested whether any of the mediation paths described in the two structural equation models were moderated by communication medium, using the Lavaan package for R. Path coefficients are estimated in a similar to Hayes (2012). There was only one significant interaction: there was a conditional mediation effect of HU on flaming via MA. Specifically,
MA significantly mediated the effect of HU on flaming in the text condition (conditional indirect effect for the text condition = .047 [.004, .109]) but not the voice condition
(conditional indirect effect for the voice condition = -.03 [-.081, .003]). Results for all models are shown in Table 4.
Chapter 3: Computer-Mediated Communication and Aggression 39
Table 4. Moderated Mediation Analyses
Predictor Dependent Interaction Effect Interaction Index of Measure Tested Coefficient [95% moderated CI] mediation
Medium Flaming HU*Medium .066 [-.224, .366] -
Medium Flaming HN*Medium -.082 [-.378, .239] -
HN Flaming MP*Medium .160 [-.041, .367] -
HU Flaming MA*Medium .607 [.155, 1.051] .078 [.020, .162]
HN MP HN*Medium .184 [-.097, .478] -
HU MA HU*Medium .039 [-.066, .148] -
Discussion There was mixed support for the hypothesized mechanism. Consistent with H1, CMC
(compared to voice) lead to more dehumanisation (for HU only). Consistent with H2, dehumanisation resulted in moral disengagement, although there was no downstream effect of disengagement on flaming. There was mixed support for H3, there was a small, indirect effect of medium on flaming, via dehumanisation (of HU), such that text lead to more dehumanisation (compared to voice), and there was, in turn, increased flaming. However, there was a larger direct effect of medium on flaming that was larger but in the opposing directionality, such that there was more flaming in the voice condition than in the text condition (the opposite directionality to the hypothesis). Finally, there was no support for H4,
CMC (compared to voice) did not lead to a reduction in self-humanness (HU or HN).
Moral Disengagement Underlies Flaming Moral status did not explain the effect of communication medium on flaming, although moral status was relevant to flaming independent of communication medium.
Taking together the results of the t-tests (of HU/flaming across different media) and the path Chapter 3: Computer-Mediated Communication and Aggression 40 model, there is no evidence to suggest that moral status can explain the effect of communication medium on flaming (neither directly nor indirectly). However, moral status does predict flaming independent of communication medium. Specifically, when the target was denied HN, the target was also denied moral patiency, and this predicted more flaming.
HN relates to the perception that someone is emotional, and, by extension, capable of being harmed. Thus, mechanistic dehumanisation (denial of HN) denies someone the capacity to be harmed, thereby excluding them from normative moral treatment. Thereby allowing the participant to aggress toward the target (i.e., flame) without violating their moral values.
In addition, there was a condition indirect effect of HU on flaming, via moral agency.
Specifically, in the text condition only, animalistic dehumanisation lead to the denial of the target’s moral agency, which increased flaming. There was no significant moderation by communication medium of any other pathways. This suggests that the mechanism by which flaming occurs differs depending on whether the target is presented by voice or text. While, moral disengagement via mechanistic dehumanisation may be a mechanism shared by any instance of aggressive behaviour, both CMC or when interacting via rich media (e.g., FtF, voice). Instead, moral disengagement via animalistic dehumanisation may be a mechanism that contributes to flaming only in CMC.
These findings for moral status make several important contributions to the literature.
First, to my knowledge, moral disengagement has not been documented in the context of
CMC. These results demonstrate that dehumanisation and moral disengagement are one of the mechanisms that contribute to aggression in CMC contexts.
Second, these results show that not all psychological mechanisms are consistent across communication media. One implication of the variation in social and contextual cues available across different presentation media is that psychological mechanisms may function differently in CMC compared to FtF or voice. For example, in the absence of non-verbal Chapter 3: Computer-Mediated Communication and Aggression 41 cues such as facial expressions or body language, the way in which impressions are formed in a CMC context is distinct from the way impressions form in a FtF context (Walther, 1993,
2007; Walther & Tong, 2014). These results show that, at least for moral disengagement (of
MP) via animalistic dehumanisation, the part of the psychological mechanism underlying aggressive behaviour is specific to CMC. This finding highlights the importance that researchers do not assume that psychological mechanisms are equivalent across different communication media.
Voice Conveys Human Uniqueness Unlike perceptions of HN, communication medium did impact ratings of HU. This result suggests that there are social and contextual cues conveyed by listening to a voice, but filtered-out by CMC, that convey uniquely human characteristics. Importantly, this might imply that there is a tendency to animalistically dehumanise others when perceiving someone in a social and contextually restricted medium (such as CMC), relative to voice or FtF contexts. These results are consistent with those identified by Jensen, et al., (2000), who found that voice interaction lead to greater perceptions of intelligence (a HU quality) and are partially consistent with those found by Schroeder and Epley (2016). While I find that voice conveys HU and CMC (or text) conceals HU, unlike Schroeder and Epley (2016), I fail to find any effect of communication medium on HN. However, the authors note that voice
(versus text) had a significantly larger impact on HU than HN ratings and conclude that voice more clearly conveys cues related to HU qualities than HN qualities. Therefore, the effect of voice on HU may be a larger and more reliable effect, that is easier to consistently identify.
Methodological differences between the current study and Schroeder and Epley
(2016) may explain why this study failed to find an effect of medium on HN. Specifically,
Schroeder and Epley (2016), recorded people speaking naturally about experiences or opinions, then transcribed these recordings to text. In the current study, I created the text Chapter 3: Computer-Mediated Communication and Aggression 42 stimuli, then asked people to voice these scripts. As people were voicing opinions that were not their own in my study, people may not have conveyed the same degree of emotional experience and interpersonal warmth (or HN qualities)1 as the voice stimuli used in
Schroeder and Epley (2016).
This interpretation also suggests that HU is easier than HN to convey in an authentic manner. One possibility is that people are more sensitive to authenticity of emotionality (i.e., a quality central to HN perceptions) than authenticity of qualities related to HU. Indeed, there is fMRI evidence to suggest that people are sensitive to the authenticity of emotion in voice
(Drolet, Schubotz, & Fischer, 2012). Specifically, fMRI imaging show that people respond in a systematically different way to real versus acted voices (Drolet et al., 2012). Therefore, the authenticity of the voice may have limited perceptions of HN but not of HU. Although additional research is required to examine this possibility.
These results also show that animalistic dehumanisation in CMC has a downstream effect on the likelihood of flaming. Dehumanisation had a small, indirect effect on flaming through communication medium. Specifically, there was greater dehumanisation in the text condition than in the voice condition, and this dehumanisation increased the tendency to aggress. However, this effect may not be practically meaningful as the effect is very small (β
= -.038) relative to the direct effect of communication medium on flaming (β =.113) and the effect of human nature on flaming (β = -.408).
Finally, the effect of medium on dehumanisation may be driven by positivity effects.
Specifically, the humanness measure does not de-confound general positivity and humanness.
However, if the effect of medium on mechanistic dehumanisation was an effect of only general positivity, then we would also expect that this would be the case for animalistic
1 It should be noted that participants were asked to convey emotional qualities, in an attempt to manage this potential limitation. Chapter 3: Computer-Mediated Communication and Aggression 43 dehumanisation. Medium only impacted HU and not HN, implying that the result was not simply driven by positivity.
The Paradoxical Effects of Communication Medium The path model produced rather unexpected results: while dehumanisation increases flaming and there was less dehumanisation in the voice condition, paradoxically there was more flaming in the voice condition. This seems to suggest that there are multiple, independent factors underlying the relationship between communication medium and flaming. On the one hand, voice confers more HU, which decreases likelihood of aggressive response – this is indicated by the indirect effect in the model displayed in Figure 5.
However, on the other hand, a direct effect of medium on flaming response, unmediated by humanness, acts in the opposite direction. This effect is stronger and explains why the overall effect emerges in the direction that it does, namely, more aggression in voice.
This result is inconsistent with the CFO model. The CFO model predicts that the reduced number of social-contextual cues in text-restricted online communication
(particularly anonymity) is a central factor in enabling flaming (Chui, 2014; Lapidot-Lefler &
Barak, 2012; Mungeam, 2011; Reinig & Mejias, 2004). The CFO model cannot account for why a lean medium might result in less aggression, compared to a richer medium. Similarly, the two experiments that have compared the prevalence of flaming across communication mediums (Castellá, Abad, Alonso, & Silla, 2000; Lapidot-Lefler & Barak, 2012) are inconsistent with our findings in that they found an increased tendency to flame in text- restricted contexts compared to richer mediums. These studies instead support the prediction of the CFO Model, that the absence of social-contextual cues should promote flaming.
In contrast, the results of this study imply that the presence of certain cues in the cue- rich condition (i.e., voice condition) drives flaming. It is unclear why the cues in the voice condition facilitated flaming in this study, but not in previous studies. One possibility is that Chapter 3: Computer-Mediated Communication and Aggression 44 there is an interaction between the medium and a quality of the actor, expectations. This interpretation is discussed in detail in Chapter 4.
Practical Implications There is an interesting practical implication of these findings. The direct effect of communication medium on flaming suggests that those with competing political orientations may benefit from interactions via lean media, such as social media. This is particularly relevant to recent concerns about increasing political polarization and the contribution to this polarization by social media ‘echo-chambers’ (Bright, 2016; Gentzkow, 2016). Some have claimed that social media may be increasing divisiveness by failing to create dialogues between those of opposing political views (Bright, 2016; Gentzkow, 2016). Instead, our results suggest that social media (such as the forum we used in the context of the study) could be utilized to benefit political discussion amongst those with competing values by reducing hostility.
Relatedly, there is some evidence that long term contact between outgroups of different cultural backgrounds (and to some degree, different political backgrounds) can reduce prejudice, in the context of CMC (Walther, Hoter, Ganayem, & Shonfeld, 2015).
Although there was no comparison to another communication medium (Walther et al., 2015).
This research expands upon this area of research, highlighting the possibility that CMC could be used strategically to help outgroups communicate with less hostility. Future research could explore prejudice reduction methods could benefit from incorporating CMC, compared to more typical means of communicating in prejudice reduction studies (e.g., FtF).
Study Limitations and Future Directions One methodological limitation that may have contributed to the paradoxical results is the small number of voice stimuli. Only one male and one female voice was used in this study, which is sufficient to suggest that results are not driven by idiosyncratic qualities of Chapter 3: Computer-Mediated Communication and Aggression 45 one voice. This was statistically reiterated when there were no significant differences in main dependent variables between the voice stimuli. However, using only two voices may limit the generalisability of results, especially if those voices happen to be skewed in some way. This possibility seems less likely given that there are no significant differences between the stimuli. Future research should replicate this effect with a larger number of voice stimuli to confirm the validity of the medium effect.
A number of studies of online behaviour have highlighted the importance of time in understanding medium effects on perception and behaviour (Tong & Walther, 2015; Walther,
1995, 1996). This study only explored how initial encounters between people with opposing political beliefs differ across different communication media. It is possible that a bidirectional, extended interaction may generate a different pattern of medium effects.
Communication is less efficient in CMC as a function of the fewer socio-contextual cues, retarding many interpersonal processes, including impression development (Walther, 1995).
Therefore, the initial humanness perceptions of the target might change with extended interaction. The Social Information Processing model would predict that the difference in animalistic dehumanisation in the CMC and voice condition would disappear with extended interaction. According to this model, extended interaction allows actors to compensate for the absence of non-verbal cues by conveying the information that would be typically conveyed in non-verbal cues with verbal cues (Walther, Loh, & Granka, 2005). For example, facial expression and voice tone are non-verbal cues heavily utilised when identifying emotion in a
FtF interaction. When interacting via CMC, the Social Information Processing Model claims that people articulate their emotions using verbal information to compensate for the absence of facial expression and paralinguistic cues (Derks, Fischer, & Bos, 2008). Although beyond the scope of this thesis, future research should assess whether verbal cues are able to replace the role of voice in conveying humanness during extended interactions. Chapter 3: Computer-Mediated Communication and Aggression 46
Extending the interaction may also diminish the direct effect of communication medium on flaming. Lean modes of communication allow for more positive impression management (e.g., through message editing) and less ‘leakage’ of undesirable affect or attitude through non-verbal cues (Walther, 1996). Assuming that actors are motivated to be perceived positively by others (even those they disagree with), the capacity for positive impression management in longer interactions could further de-escalate hostile orientations, facilitate positive impressions, and further reduce instances of flaming. Future research should assess whether the difference in flaming between the communication media replicates during extended interactions.
Conclusion This study is a well-powered, empirical assessment of the CFO Model as an account for the effect of communication medium on aggression. Contrary to predictions, there was more aggression in the cue-impoverished condition, CMC, than in the cue-rich condition voice condition. These results suggest that the CFO Model is a poor account for the effect of communication medium on negative moral behaviours.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 47
Chapter 4: Self-Disclosure in Computer Mediated Communication Study one found limited evidence for the Cues-Filtered-Out (CFO) approach to a negative moral behaviour in the context of CMC. While the absence of non-verbal cues may have produced the dehumanisation effect, the poverty of cues does not easily explain the reduction in aggression. One possibility is that CFO theory may be a better explanation of positive moral behaviour.
Several studies have found that the qualities of CMC will often facilitate and exaggerate positive impressions of others (Tong & Walther, 2015; Walther, 1996; Walther et al., 2013; Walther & Tong, 2014). Specifically, the ambiguity generated by the absence of non-verbal cues has been shown to promote uncertainty-reducing strategies, such as over- reliance on stereotypes and expectations (Walther, 2007). These strategies, when coupled with the self-selective presentation offered by CMC (e.g., editing messages, omitting unfavourable qualities), can lead to more positive impressions and prosocial behaviours (such as self-disclosure) than in cue-rich media (Walther, 2007). Given that communication medium also impacts the incidence of positive perceptions and behaviour, one might expect that my application of the CFO account is able to account for certain positive moral behaviours in CMC contexts. In this chapter, I further test the claim that the absence of certain social and contextual cues in text communication changes the perception of others, specifically the perception the of other’s humanness, but in the context of a positive moral behaviour: intimate self-disclosure.
Unlike medium effects on aggression, there is greater consensus in the literature about the medium effects on self-disclosing and intimacy related behaviours, which may provide a better opportunity to test the CFO account. A large body of contemporary evidence points to an effect of medium on self-disclosure in experimental research, including a systematic review (Nguyen, Bin, & Campbell, 2012). In comparison, there is disagreement on the effects
Chapter 4: Computer-Mediated Communication and Self-Disclosure 48 of medium on aggression, with some researchers arguing that aggressive behaviour occurs at similar frequencies in cue-poor and cue–rich media (Abrams, 2003; Lea et al., 1992; Reid &
Reid, 2005). Given that one goal of this thesis is on understanding the role of humanness in moral behaviours, as a function of cues, I sought to explore this general question in a context where the basic medium effect was less contested.
This chapter also makes a number of improvements on Study 1. First, I draw from another theoretical account of CMC behaviour, the hyperpersonal model, to improve the proposed CFO model. Specifically, the hyperpersonal model provides an explanation of how the absence of cues in CMC affects impression formation. According to the hyperpersonal model, people compensate for the relative lack of information in CMC contexts by over- utilizing available information (Hancock & Dunham, 2001; Walther, 1996). I hypothesize that these compensatory behaviours may affect humanness perceptions and account for changes to moral behaviour in cue-poor environments.
Second, using the hyperpersonal model as an explanatory framework, this chapter more directly examines how the poverty of cues in CMC affects moral behaviour by manipulating the presence of individuating cues. Study 1 allowed cues to freely vary between the voice and text conditions, under the assumption that, on average, there would be fewer cues in the text condition than in the voice condition. Yet, the poverty of cues in the text condition (and richness of cues in the voice condition) was not sufficient to produce changes in moral behaviour in the expected way. Rather than a general effect of medium, instead it may be that medium moderates the way that cues are used. That is, both the medium and the content of communication together determine moral behaviour. There is some evidence to suggest that disinhibited behaviour in CMC environments is determined by a combination of context and medium (Hutchens, Cicchirillo, & Hmielowski, 2015; Kayany, 1998). These studies have found that there is not a simple effect of medium on disinhibited behaviour, but
Chapter 4: Computer-Mediated Communication and Self-Disclosure 49 rather an interaction between an aspect of the social context (e.g., the topic of discussion, salience of outgroups) and the poverty of cues in text interaction. Therefore, Study 2 will assess whether there is an interaction in the medium and the content of cues, by manipulating the presence or absence of cues related to humanness.
Given the lack of evidence for the CFO model in Study 1, I expand the theoretical scope in Study 2 by assessing other prominent CMC theories that may better account for how medium impacts moral behaviour and perception. First, I explore the role of humanness in the hyperpersonal theory; next, I assess the reduced cues (RC) theory and the potential role for the fear of negative evaluation in driving medium effects on positive moral behaviour; I then discuss the social identity and deindividuation effects (SIDE) and the possibility that medium effects on positive moral behaviour are the result of deindividuation of the self and salient social identities. Finally, I explore several potential moderators that may strengthen my application of the CFO account: (1) interaction expectations as an individual difference; (2) social presence; (3) trust; (4) attraction.
Self-Disclosure and Computer-Mediated-Communication
While some CMC interactions may ultimately lead to hostility and flaming, there are also countless highly intimate, warm and positive interactions that occur online (Derks,
Fischer, & Bos, 2008; Walther, 1996; Whitty & Carr, 2006). CMC may paradoxically result in ‘hyperpersonal’ interactions that match or surpass FtF in feelings of intimacy and self- disclosure (Hammick & Lee, 2014; Jiang, Bazarova, & Hancock, 2011; Joinson, 2001;
Nguyen et al., 2012; Stritzke, Nguyen, & Durkin, 2004; Walther, 1996, 2007). I hypothesize that perceptions of humanness may explain online hyperpersonalism and excessive self- disclosure. That is, in some specific circumstances, CMC may actually produce perceptions of humanness that would surpass those made in an otherwise equivalent FtF context, leading to ‘hyperhumanisation’.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 50
Hyperhumanisation and Hyperpersonalism
The hyperpersonal account argues that CMC users take advantage of certain affordances of the medium to improve their relational outcomes in such a way that can lead to greater intimacy than an equivalent interaction in a rich-media (Walther, 1996). According to this account, in cue-poor media (such as CMC) people utilise the absence of certain cues to positively manipulate the way in which others perceive them. For example, text communication is editable (while voice or FtF communication is not); CMC allows more time to compose messages than richer media; and the absence of non-verbal cues masks
‘leakage’ of undesirable affect or attitude. These qualities of text-based communication allow people selectively self-present certain desirable attributes and thus manage how others perceive them.
Furthermore, the model posits that people compensate for the absence of cues by over-exaggerating the importance of those available cues when forming impressions of others
(Walther, 1996). Hancock and Dunham (2001) found that when forming impressions using
CMC, people perceived a target’s personality traits to be both simpler (i.e., participant rated the target on fewer personality traits) and more extreme, than in the comparative FtF condition. Ellison, Heino, and Gibbs (2006) also demonstrate this compensatory behaviour in a study of online dating. The authors found that people over-extrapolate from subtle cues when forming impressions of others. For example, stylistic cues such as grammar, spelling and message length were equally important in perceiving qualities such as character, education, attractiveness as actual message content. Taken together, the capacity to positively manipulate self-image and the exaggerated importance of those available cues is argued to lead to interpersonal intimacy that exceeds richer media.
One possibility is that these hyperpersonal processes may also affect humanness perceptions in cue-poor media; I hypothesize that there are certain contexts of cue-poor
Chapter 4: Computer-Mediated Communication and Self-Disclosure 51 media (the process of) hyperhumanisation may occur. Hyperhumanisation can be thought of as an example of caricatured impression formation that occurs due to the lack of richness in social and contextual cues in CMC contexts. As a result of the over-utilization of available cues in cue-poor media, impressions become both simplified (based on fewer social cues) and exaggerated (each cue is over-weighted) – impressions are, in effect, ‘caricaturized’ (Ellison et al., 2006; Hancock & Dunham, 2001; Tanis & Postmes, 2003; Walther, 1993). When the cues available are indicative of either HN or HU (e.g., language connoting emotion, rationality, or thoughtfulness) in a cue-poor interaction, then caricaturizing may lead to impressions of hyperhumanness (i.e., perceptions of humanness in CMC may exceed similar perceptions in FtF).
More Human in Text: Greater Moral Status If dehumanisation is linked to a lower moral status (as shown in Study 2), the converse may also be true: hyperhumanisation may increase moral status. However, the kind of moral status increase should depend on the kind of humanness that is amplified (Bastian, et al., 2011). If the humanness cues available during communication happen to suggest HN
(rather than HU) then caricaturization will be along this dimension. A target that is perceived to be selectively high on HN is likely to be seen as having high levels of emotionality, warmth, and empathy – partner qualities that have been shown to increase the likelihood of self-disclosure and the development of close relationships (Joinson, Reips, Buchanan, &
Schofield, 2010; Metzger, 2004; Miller, Berg, & Archer, 1983; Wheeless, 1978).
Furthermore, according to the findings of Bastian et al. (2011), people that are rated high in
HN qualities will, in turn, be judged as having high levels of moral patiency. In turn, those high in moral patiency may be perceived as more sensitive to harm and more deserving of positive experiences; thus, increasing the likelihood of pro-social behaviours such as self- disclosure. Therefore, when an actor conveys HN cues in a cue-restricted medium (e.g.,
Chapter 4: Computer-Mediated Communication and Self-Disclosure 52
CMC), the actor will be perceived as having exaggerated HN qualities, greater moral patiency, and, finally, facilitating self-disclosing to the target, when compared to a cue-rich media (e.g., video).
Exaggeration of HU qualities may result in a different kind of moral status shift and thus a different set of behaviours. If the humanness cues are indicative of HU, then the target may be perceived as having selectively high levels of HU and so may not be seen as a moral patient but rather a moral agent with intellectual and rational capacities. One possibility is that caricaturing of these qualities may result in ‘mechanistic hyperhumanisation’ whereby a target is perceived to have such high levels of these HU traits that they are judged as cold and robotic in nature. This may actually decrease the likelihood of self-disclosure. Research from human-robot interactions have shown that robots that are perceived as more warm and human-like (i.e., lower in HU and higher in HN) are trusted more and disclosed to more frequently (Hancock et al., 2011; Hoffman, Birnbaum, Vanunu, Sass, & Reis, 2014).
Therefore, when a target conveys HU cues in a cue-restricted medium (e.g., CMC) the target will be perceived as more uniquely human, of greater moral agency, and, in turn, will engender less self-disclosure, when compared to a cue-rich medium (e.g., video).
According to this account, we would expect a main (and positive) effect of medium on humanness ratings (for both HN/HU). In addition, the account also predicts a main and positive effect of HN manipulation on self-disclosure, such that there will be more self- disclosure in the high HN conditions than the low HN conditions. Furthermore, this theory predicts an interaction effect: when the target is presented via text and also has cues indicating high HN, the positive effect of HN on self-disclosure should be greater than when the target is presented via video. In comparison, this theory predicts a negative main effect of
HU on self-disclosure. There should also be an interaction between HU and condition when predicting self-disclosure; the negative effect of HU on self-disclosure should be greater
Chapter 4: Computer-Mediated Communication and Self-Disclosure 53 when in the text condition than in the video condition.
Alternative Theoretical Accounts of Self-Disclosure
Reduced Cues (RC) Theory and Self-Disclosure
Although the introduction of medium and cue interaction effects adds an important nuance to the CFO model not explored in Study 1, I wanted to also engage with numerous other plausible accounts of disclosure in the current study. Each of the theories examined predict the same effect of medium on self-disclosure, however, the theories vary in the mechanisms by which medium affects self-disclosure. The subtle differences are outlined below.
Similar to my application of the CFO perspective, the RC theory focuses on the role of the absence of social and contextual cues in CMC in affecting behaviour. While the CFO perspective argues that the absence of cues in the text condition leads to changes in the perception of the target, such as the exaggeration of available information, the RC theory argues that the absence of cues leads to changes in self-perception. According to the RC theory, cues that typically indicate social context and communicate associated behavioural norms are absent in text communication, thereby reducing fear of negative evaluation and leading to disinhibited behaviour (Culnan & Markus, 1987). The RC theory argues that without non-verbal cues, communicating partners are less able to convey feedback to their partners during conversation that reinforce social norms (Dubrovsky et al., 1991). As a result, people become less self-conscious and, in turn, disinhibited. Specifically, people communicating via text are argued to experience less public self-consciousness, or concern about the evaluation of others (Joinson, 2001). Freed from fear of negative evaluation, actors in CMC may disclose more often and more intimate information than in an equivalent FtF interaction (Dubrovsky et al., 1991). In a study of self-disclosure amongst socially anxious
Chapter 4: Computer-Mediated Communication and Self-Disclosure 54 adults, visually anonymous (no web cam) CMC was compared to non-anonymous CMC (web cam) (Brunet & Schmidt, 2008). A greater number of self-disclosures were expressed when actors were visually anonymous, compared to the non-anonymous condition. Likewise, when self-consciousness was directly manipulated, lower public self-consciousness lead to greater levels of spontaneous self-disclosures in CMC (although no non-CMC condition was compared) (Joinson, Woodley, & Reips, 2007). Together, these studies suggest that the absence non-verbal cues in CMC (particularly visual cues) reduces self-awareness, in turn facilitating disinhibited behaviour, such as self-disclosing.
Therefore, according to the RC theory in the absence of non-verbal cues (such as in
CMC) there will less self-consciousness, when compared to a cue-rich medium (such as video). This reduction in self-consciousness will, in turn, lead to disinhibited behaviour including more self-disclosure compared to the video condition. This effect is not expected to be mediated by humanness.
Social Identity and Deindividuation Effects Model (SIDE)
The SIDE model also provides an explanation for increased tendency to self-disclose in cue-rich compared to cue-lean interactions. Like the RC and CFO accounts, this model also focuses on the role of social and contextual cues in understanding medium affects. Also similar to the RC theory, the absence of cues is thought to lead to changes in self-perception, however, the SIDE model argues that these changes are along the dimension of personal versus group identification. The SIDE model argues that the absence of visual cues in some media (e.g., CMC) emphasizes people’s social identity and minimises their personal identity
(Lea et al., 2001). According to the model, there is, on average, less individuating information about the target in cue-poor media, the absence of such information minimises differences between the group members and the self and leads the actor to categorise both themselves and the target in terms of their shared social identity rather than individual
Chapter 4: Computer-Mediated Communication and Self-Disclosure 55 identities (Lea et al., 2001). The net result is that people perceive themselves and the target in terms of stereotypical group features (Spears, 2017). As the shared social identity is salient in cue-poor contexts, the actor is more likely to behave in ways consistent with the group’s norms, compared to cue-rich context (e.g., FtF or video). Thus, the SIDE model predicts that there will be greater self-disclosing behaviour in cue-poor media only if there are salient local norms associated with self-disclosing. Unacquainted individuals meeting via CMC tend to disclose more as a means of reducing uncertainty, compared to offline interactions (Tidwell
& Walther, 2002); thus, it is reasonable to expect that uncertainty-reducing behaviours, like self-disclosing, may be a social norm for CMC where both parties are (assumedly) motivated to get to know one another.
Therefore, according to the SIDE model, there will be more self-disclosure when a target is presented via a lean-media (e.g., text), compared to rich-media (e.g., video). This increased self-disclosure will be explained by reduced self-consciousness, which leads to greater perceived similarity between the self and the target. As is the case with RC, this theory does not expect any effect of humanness.
Summary of Main Theories
This study addresses three major theoretical accounts of CMC behaviour that have proposed explanations for medium effects on self-disclosure: (1) the hyperpersonalism account with humanness extension; (2) the RC account; and (3) the SIDE model.
The hyperpersonal account with humanness extension predicts that the absence of the majority of non-verbal cues in text, compared to video, will lead participants to exaggerate whatever cues are available in text. When the available cues indicate high HU or HN, those humanness dimensions will be exaggerated, resulting in ‘hyperhumanisation’ or humanness perceptions that exceed perceptions of the equivalent stimuli presented via video. When this hyperhumanisation is along the HN dimension of humanness (animalistic
Chapter 4: Computer-Mediated Communication and Self-Disclosure 56 hyperhumanisation), people are perceived as warmer and more emotional, facilitating interpersonal intimacy (such as, self-disclose). In contrast, when this hyperhumanisation is along the HU dimension (mechanistic dehumanisation), people are perceived as cold and robotic in nature, inhibiting interpersonal intimacy and self-disclosure.
The RC account predicts that actors presented with a text stimulus have reduced self- consciousness and so are less concerned with negative social feedback. This, in turn, leads to become disinhibited, increasing their tendency to self-disclosed (compared to those in the video condition). Therefore, this account hypothesizes that there is a main effect of medium on disclosure and an indirect of medium on disclosure via self-consciousness but makes no predictions about humanness.
Finally, the SIDE model predicts that when a stimulus is present in text, social identities become more salient and personal identities are minimised. As a result, the participant tends to define the self and the target in terms of a shared social identity.
Therefore, the SIDE model predicts that participants will have greater perceived-similarity in the text condition, compared to the video condition. As a shared social identity is salient in the text condition, participants will be more likely to follow local social norms associated with this identity. In the case of online dating (the context in the current study), we might expect that these group norms will be self-disclosing more frequently to get to know the target. Like the RC, this model also predicts more self-disclosure in the text condition than in the video condition. However, in addition, to the main effect there will also be an indirect effect of medium on self-disclosure via perceived-similarity.
Possible Moderators, Mediators, and Control Variables Study 1 illustrated that the reduced cues in restricted media cannot alone account for changes to moral behaviour. As a result, I include a number of exploratory moderator, mediator, and control variables in an effort to better account for online moral behaviour.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 57
More contemporary theorising on the effect of communication media on behaviour has emphasized the importance of contextual factors and individual differences in explaining online behavioural/perceptual phenomena (such as self-disclosure). For example, Kayany
(1998) found that the frequency of flaming in online contexts markedly differed depending on whether members shared political, religious, or cultural beliefs, thus suggesting that certain behavioural phenomena may be explained by both social context and individual differences. Hutchens et al. (2015) identified a role for social norms and context in intentions to flame. Namely, in online political debates, perceptions that flaming is socially accepted predicted intentions to flame. Similar nuances have been suggested for explaining medium effects on self-disclosure: Schouten, Valkenburg, and Peter (2007) found that expectations about the effect of reduced cues in online interactions and trait-level self-consciousness predicted self-disclosure in CMC. Specifically, those that are more self-conscious (at the trait-level) felt that the absence of cues in CMC affected behaviour more than those with low self-consciousness, which, in turn, lead to more disinhibition and self-disclosure. Given the possible role of the attributes of the social context and individual differences, I will explore a number of variables that may better improve my account of any medium effects on self- disclosure.
Pre-Interaction Expectations First, I explore the possibility that expectations as an individual difference may moderate any relationship between medium and self-disclosure or humanness. I explore the possibility that people hold stereotype-like expectancies about how others tend to behave online - and these expectancies may moderate any effect of medium on humanness perceptions, and in turn, self-disclosure.
Online interactions, relative to offline interactions, are particularly susceptible to the influence of pre-interaction expectancies (Epley & Kruger, 2005; Walther & Tong, 2014).
Chapter 4: Computer-Mediated Communication and Self-Disclosure 58
Many of the social and contextual cues available in typical offline interactions are absent when interacting online, rendering online interactions more ambiguous than their offline counterparts. People compensate for the ambiguity of the medium with various uncertainty- reduction strategies (Ramirez et al., 2002; Tidwell & Walther, 2002), such as the application of pre-interaction expectancies. Epley and Kruger (2005), for example, showed that participants using CMC relied more on expectancies of specific targets when forming impressions than those listening to voice recordings. The authors further argued that the ambiguity of the medium required participants to over-utilise expectancies, as it was the one of the few sources of information in the lean medium. Likewise, the hyperpersonal model posits that the qualities of CMC will often facilitate confirmation and exaggeration of expectancies, particularly when the expectancies are positive (Tong & Walther, 2015;
Walther, 1996). Provided this evidence for the importance of expectations in CMC, expectations will be examined as a moderator.
Specifically, this study will assess whether people hold expectations about how people generally behave when communicating via CMC and whether this impacts behaviour.
The omnipresence of the internet (a context that requires CMC) implies that most people are familiar with online interactions and are likely to have developed some expectations for how others generally behave on this platform. There is evidence that people hold at least three types of expectations, expectations about others (1) likelihood to behave aggressively, (2) deceitfully and (3) behave in disinhibited ways. Specifically, people report very high expectations that others will be deceitful and misrepresent themselves online; one qualitative survey found that 90% of the sample expected others to misrepresent their appearance online and even quoted a participant stating that “everyone lies on the internet.” (Drouin, Miller,
Wehle, & Hernandez, 2016, p. 141; Ellison et al., 2006). Similarly, online aggression
(flaming) is thought to be so pervasive online that it is a social norm to flame in some online
Chapter 4: Computer-Mediated Communication and Self-Disclosure 59 communities (Moor, Heuvelman, & Verleur, 2010; Postmes et al., 1998). Self-disclosure appears to be perceived as very pervasive and people vary in the extent to which they conceptualise this as a positive or a negative behaviour. The results of some surveys report self-disclosure with negative connotations, such as a recent Pew poll shows that ⅓ of millennials perceive others to ‘over share’ online (Lenhart, 2015). Thus, these three dimensions of expectations were studied in this exploratory aspect of the study.
Given that text is more ambiguous because there are fewer social and contextual cues available, we would expect that participants will more heavily utilise their expectancies for the interaction to inform their behaviour and perception in the text condition; while in the video condition there will be more non-verbal social and contextual cues and thus participants will rely less on the expectancies when forming an impression of the target and composing a message to the target. Therefore, I expect that expectations will have a larger effect on self-disclosure and perceptions of humanness/moral status when the target is presented via text compared to video.
Social Presence Another theory that provides an account for how the absence of cues in CMC changes behaviour is social presence theory. As defined in Study 1, social presence is the degree to which an actor experiences the other person as salient during an interaction (Short, Williams,
& Christie, 1976). When someone experiences high social presence, they feel connected and engaged in communication with another actor. In comparison, low social presence may result in individuals perceiving the interaction as if only with the computer and not another human agent. CMC interactions are argued to lack social presence due to the absence of social cues that would ordinarily make the interactive partner salient in an offline interaction (Short et al.,
1976). Building on the RC account, lowered social presence, in turn, has been associated with reduced anxiety about negative social evaluation and increased self-disclosures (Joinson &
Chapter 4: Computer-Mediated Communication and Self-Disclosure 60
Paine, 2007; Joinson et al., 2007; Tourangeau, Couper, & Steiger, 2003). Therefore, we might expect that social presence will be, on average, less in the text condition than in the video condition and may account for medium effects on self-disclosure. Further, according to social presence theory, we might expect that social presence mediates the effect of medium on self-consciousness (from the RC theory).
Trust The relationship between trust and self-disclosure is straightforward: trust is critical in deciding when to share personal information with others (Collins & Miller, 1994; Corcoran,
1988; Steel, 1991; Wheeless & Grotz, 1977). When intimate information is shared, the discloser may fear a ‘reverse halo-effect’ where the disclosed-to person generalises about other weaknesses the discloser might have (Wenburg & Wilmot, 1973). More recent research has also demonstrated a link between trust and self-disclosure in online contexts, including
CMC (Chang & Heo, 2014; Joinson et al., 2010; Metzger, 2004; Sheldon, 2009; Taddei &
Contena, 2013; Tait & Jeske, 2015; Valenzuela, Park, & Kee, 2009). Given that trust is an antecedent of self-disclosure, the hypothesized relationships between humanness/communication media and self-disclosure may only be observed when target is also trusted.
Trust is also related to perceptions of humanness. In human-robot interactions, robots that feature more human-like features (e.g., responsiveness, eye-contact, ‘personality’) are trusted more than robots that do not have human-like attributes (meta-analysed by Hancock, et al. 2011). Likewise, in CMC contexts, avatars with more anthropomorphism are more trusted and more attractive than avatars without anthropomorphism (Koda & Maes, 1996;
Nowak, 2004; Nowak & Biocca, 2003; Nowak & Rauh, 2005; Wexelblat, 1997). Thus, if
CMC leads to hyperhumanisation, then exaggerated perceptions of trust may lead to increased self-disclosure. Therefore, one possibility that I will explore is that trust mediates
Chapter 4: Computer-Mediated Communication and Self-Disclosure 61 the relationship between hyperhumanness and self-disclosure (i.e., for the CFO account only).
Attraction Another important antecedent variable important to self-disclosure is attraction.
According to predicted outcome value theory (Sunnafrank, 1986), people are motivated to self-disclose when they judge self-disclosures to positively benefit the relationship, for example, by increasing intimacy and indicating their trust in the recipient. Therefore, perceptions of attractiveness may drive self-disclosure via a desire to develop an interpersonal relationship. Consistent with this theorising, a number of studies have found that people are both more willing to self-disclose to someone that they initially find attractive and tend to find those that self-disclose more attractive than those who don’t self-disclose
(reviewed in FtF interactions by Collins & Miller, 1994). For example, Ramirez, Walther,
Burgoon, and Sunnafrank (2002) and Levine (2000) both argue that the desire to seek information through reciprocal self-disclosure in CMC environments is sometimes influenced by initial perceptions (including attraction). In studies of Facebook friendships, social attraction and trust has been shown to be both an antecedent of self-disclosure (Craig, Igiel,
Wright, Cunningham, & Ploeger, 2007; Sheldon, 2009) and a consequence of self-disclosure
(Craig & Wright, 2012). Thus, in CMC contexts, people tend to be attracted to those that self- disclose and self-disclose more to those that they are attracted to. As available cues tend to be weighted more heavily during impression formation that occurs in CMC than in richer media, self-disclosure may be perceived as more attractive in text than in richer media, leading to more self-disclosure reciprocation. In the context of the current study, differences in attraction across media may account for medium effects on self-disclosure. This potential mediator could apply to any of the hypothesised account (i.e., CFO, RC, SIDE).
Chapter 4: Computer-Mediated Communication and Self-Disclosure 62
Study Overview
In order to assess which of the above accounts best explains the hypothesized impact of medium on self-disclosure, I performed a study with the following design and procedure.
This study assessed whether in cue-poor media (i.e., text) the presence of humanness
(HN/HU) cues leads to: (1) ‘hyperhumanisation’ or humanness perceptions that exceeds that of an equivalent stimulus presented via a cue-rich media (video); and (2) whether hyperhumanisation has a positive downstream effect on intimacy-related behaviours (i.e., self-disclosure). Participants viewed one of four dating profiles that had been manipulated to either include cues related to high or low HN or HU perception.
Study stimuli were developed in two phases: first, potential content for the stimuli was extracted content from real OKCupid profiles. The potential content was then pilot tested for perceptions of humanness. This content was then used, in combination with stimuli used in Bastian et al. (2011), to form the profiles used in the main study. Next, the complete profiles were piloted to ensure that humanness perceptions were successfully manipulated.
Stimuli (dating profiles) were presented via a cue-poor (i.e., text) or cue-rich (i.e., video) communication medium. Thus, this study used a 2 (HN: high or low) x 2 (HU: high or low) x 2 (medium: text or video) between-subjects design. Three main theories are examined by this design; (1) hyperhumanness, (2) reduced cues (RC), (3) social identity deindividuation effects (SIDE).
Hypotheses All three theories predict that there should be a main effect of medium on self- disclosure, such that there will be more disclosure in the text condition than in the video condition (H1). However, the theories vary in what mechanism accounts for the effect of medium on self-disclosure.
The hyperpersonal account with humanness extension predicts that the absence of
Chapter 4: Computer-Mediated Communication and Self-Disclosure 63
(many) non-verbal cues in text, compared to video, will lead participants to exaggerate whatever cues are available in text. When the available cues indicate high HU or HN, participants will exaggerate their perceptions along these dimensions, resulting in hyperhumanisation or humanness perceptions that exceed perceptions of the equivalent stimuli presented via video. Thus, this theory predicts a main effect of medium on humanness ratings (for both HN/HU). Specifically, a target presented in text will be perceived as more human than the same target presented via video (H2), as the participant is compensating for the relative lack of cues in text versus video.
When hyperhumanisation occurs along the HN dimension of humanness (animalistic hyperhumanisation), participants are expected to self-disclose more frequently and intimately, as those high in HN have qualities conducive to interpersonal intimacy (e.g., warmth). Therefore, there will also be a main effect of HN manipulation on self-disclosure, such that there will be more self-disclosure in the high HN conditions than the low HN conditions (H3). Furthermore, this theory predicts an interaction effect: when the target is presented via text and also has cues indicating high HN, the positive effect of HN on self- disclosure should be greater than when the same target is presented via video (H4).
In comparison, the HU manipulation should have a negative effect on self-disclosure, as those high in HU are perceived as cold and robotic Therefore, this theory predicts a negative main effect of HU on self-disclosure (H5). Similar to HN, there should also be an interaction between HU and condition when predicting self-disclosure; the negative effect of
HU on self-disclosure should be greater when in the text condition than in the video condition
(H6).
The RC theory argues that CMC filters out relational information including cues that communicate social norms and feedback about appropriate/inappropriate behaviour, causing participants to have less public self-consciousness and become more willing to self-disclose.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 64
Therefore, this theory predicts that there will a main effect of medium on self-consciousness, such that there is less self-consciousness in the text condition than in the video condition
(H7). Self-consciousness will then mediate the effect of medium on self-disclosure (H8).
The SIDE model argues that CMC filters out individuating information, which makes salient shared group identities and obscures personal identities. As a result, people perceive greater similarity between the target and themselves when in the text than in video.
Therefore, this model predicts that there will be a main effect of medium on perceived- similarity, such that there will be more perceived-similarity in the text condition than in the video condition (H9). Furthermore, according to this account, perceived-similarity will mediate the effect of medium on self-disclosure (H10).
Stimuli Development
Sourcing Stimuli First, I sourced four targets from the stimuli used in Bastian et al. (2011). Bastian et al. (2011) used short descriptions of four different actors that systematically varied in HN and
HU ratings in a 2 (HN: high or low) x 2 (HU: high or low) experimental design. Each description drew on trait terms either related to high or low humanness ratings (Haslam, Bain,
Douge, Lee, & Bastian, 2005). Each description taken from Bastian et al. (2011) included 15 traits (10 positive and 5 negative) and was exactly 83 words in length. These stimuli served as the basis for the stimuli to be used in the main study.
Modifying Stimuli to Create Dating Profiles Next, these short descriptions were modified, by adding new content and changing the layout, to appear more similar to a dating profile. These changes, specifically adding content and changing the format, was for the purpose of (1) increasing ecological validity, and (2) strengthening the humanness manipulation.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 65
These modifications were determined by the content and format of real OK Cupid profiles. First, the Bastian et al. (2011) stimuli were inserted into the description section of an
OK Cupid profile (see Figure 6 for an example).
Pilot-Testing and Including Additional Content Next, additional content was added to mimic the appearance of a real OK Cupid
Profile. In real life, OK Cupid users are prompted to include various personality and demographic attributes when creating a profile (e.g., use of drugs, race, income, education), these attributes are then displayed in their profile. Given that people use group stereotypes to make judgements about a person’s humanness (Bastian et al., 2011; Brown-Iannuzzi, McKee,
& Gervais, 2018), including such demographic attributes in the experimental stimuli could strengthen the humanness manipulation. These attributes were pilot tested for their humanness ratings using Amazon’s Mechanical Turk participants (described below). The attributes that were rated the highest and lowest on HN and HU were then included in the relevant target stimulus (OK Cupid profile). Then, I produced a male and female version of each four profiles by changing the gender of the pronouns.
The complete profiles were then pilot tested for humanness ratings to ensure that humanness was manipulated in the desired direction.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 66
Figure 6. An example of a text condition stimulus for the high HN/high HU condition.
Producing the Video Stimuli To produce the video stimuli, six actors (3 male, 3 female) were recorded
reading each of the four target descriptions verbatim, resulting in 24 video stimuli. I
instructed the actors:
Imagine that you are the person who wrote the profile. Imbue your words with all of
the thoughts, emotions, and substance that the writer him/herself felt. Read it as if you
were actually coming up with the lines naturally off the top of your head rather than
reading from a transcript. Speak as naturally as you would if you were in the midst of
Chapter 4: Computer-Mediated Communication and Self-Disclosure 67
a real conversation. 2
The text descriptions using the same visual presentation as an OK Cupid profile (i.e., colours, layout) for ecological validity. For realism, a profile picture was added to each text profile: a still-image was taken from the recording of each video and included on the text profile. The outcome of this process was 24 versions of the text stimuli and a matching 24 versions of the video stimuli. To summarise: 4 target descriptions manipulating humanness dimensions x 2 communication media (text or video) x 6 actors reading each description or included as a profile image.
Pilot Study One
73 Mechanical Turk participants (Mage = 44.3, SDage = 13.7; 31 males) rated a number of attributes (e.g., age, occupation; see complete list in Figure 7) on humanness. Participants rated the extent to which a typical member of each attribute group (e.g., people in their 20s) was viewed as ‘culturally refined’, ‘rational or logical’, or ‘lacking self-restraint’ (reversed; following the methods of Bastian et al., 2011). Participants also rated the attribute groups on aspects of HN: the extent to which group members were viewed as ‘emotionally responsive’,
‘warm towards others’, and ‘rigid and cold’ (reversed). Each item was rated from 0 - 100 on a sliding scale, where 0 (much less than the average person) and 100 (much more than the average person).
Average HN and HU composite scores were calculated for each attribute (Overall:
3 MHU = 61 SDHU =14; MHN = 57 SDHN =10.8) . These scores are displayed in Figure 7. The attribute rated highest and lowest on either HN or HU was then included as content in the
2 Instructions were taken from similar stimuli development described in Schroeder and Epley (2016) 3 A technology failure lost the raw data for this pilot (only aggregate retained), as a result Chronbach’s alpha could not be calculated.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 68 stimuli (dating profiles) used in the main study, along with the traits from Bastian et al.,
(2011).
Attributes identified as reflecting low HU and were included for the experimental stimuli were: people who do drugs regularly, people who dropped out of high school, people that are aged somewhere in their early 20s. Items low in HN included: Muslim, investment banker, and laboratory technician. Items high in HU included: medical doctor, people who have a PhD, university professor. Finally, those items high in HN included: primary school teacher, caregiver at an elderly home, people who own a dog.
These attributes were then combined with the content from Bastian et al., (2011) to produce four dating profiles. I changed the gender of the pronouns to create an equivalent female and male version of each description. Profiles were kept as consistent as possible in other ways, for example, word count for each description was +/- 10% of one another, visual aspects of the layout were equivalent.
Figure 7. The distribution of HU and HN scores for each social category. Reference lines have been added to the mid-point of the scale (HN = 50; HU = 50) and colours have been
Chapter 4: Computer-Mediated Communication and Self-Disclosure 69 added for ease of interpretation. Social categories are coloured according to which quadrant of HN/HU scores their ratings fall within.
Pilot Study Two The four text-profiles where then rated for humanness with a second sample to verify that the final text stimuli successfully manipulated humanness (both HN/HU). Sixty-seven
Mechanical Turk workers rated all four profiles for humanness in a randomised order.
Humanness was measured using the humanness scale specified in (Bastian et al., 2011; HU =
.85, HN = .86). No demographic data was collected.
Average HN and HU scores were calculated for each of the profiles (Overall, MHU =
36, SDHU = 16; MHU = 36, SDHU = 13). Two mixed-effects models were fit with HU (0 = low,
1 = high) and HN (0 = low, 1 = high) as fixed effects (interacting) and participant as a random effect predicting either HN ratings or HU ratings. All the main and interaction effects were significant for both models indicating that the manipulations successfully influenced humanness ratings (Table 5), however, the HN and HU manipulations were not orthogonal.
In other words, the HN manipulation affected both HN and HU ratings, and vice versa. The humanness manipulations were inversely related, that is, the HN manipulation negatively affected HU ratings (and vice versa). This is consistent with some previous research that has found a negative relationship between the humanness dimensions (Bain, Park, Kwok, &
Haslam, 2009; Bastian et al., 2011).
Least squares means (or marginal means) were estimated (Table 6) and graphed for ease of interpretation (Figure 8 and Figure 9). Tukey post-hoc tests assessed whether the stimuli significantly differed in HN and HU scores. The results show that the profiles are differentiated in HN/HU in the intended pattern: the profile intended to be high HU/high HN had the highest scores of HN (M = 80, SD = 14) and HU (M = 80, SD =11), and the low
HN/low HU profile had the lowest HN (M =64, SD = 14) and second lowest HU (M = 51, SD
Chapter 4: Computer-Mediated Communication and Self-Disclosure 70
= 14) scores. Importantly, this implies that the manipulations were successful, and the profiles vary in humanness in the anticipated ways.
Figure 8. Least squares (LS) means for HN ratings for each of the experimental stimuli. Note.
Boxes indicate the LS mean error bars indicate the 95% confidence interval of the LS mean.
Means sharing a letter are not significantly different (Tukey-adjusted comparisons).
Chapter 4: Computer-Mediated Communication and Self-Disclosure 71
Figure 9. Least squares (LS) means for HU ratings for each of the experimental stimuli. Note.
Boxes indicate the LS mean error bars indicate the 95% confidence interval of the LS mean.
Means sharing a letter are not significantly different (Tukey-adjusted comparisons).
Table 5. Mixed-Effects Models for Assessing the Effectiveness of the Humanness
Manipulation
Fixed-Effects Dependent Measures
HN ratings HU ratings
Intercept 64.25 [61.1, 67.3] 50.56 [47.62, 53.49]
HN Condition 8.03 [4.28, 11.79] -4.31 [-7.71, -0.91]
HU Condition -15.384 [-19.14, 0.20] 19.35 [15.95, 22.75]
HNcondition*HUcondition 23.075 [17.76, 28.39] 14.35 [9.54, 19.16]
Note. Values are unstandardized coefficients and 95% bootstrapped confidence intervals. Bolded values are considered significant, as the confidence intervals do not contain 0.
Table 6. Marginal Means and Pairwise Comparisons for the Manipulation Check HU Ratings HN Ratings HU Manipulation Low High Low High HN Low 50.56a 69.91b 64.25 b 48.87 a Manipulation High 46.24a 79.95c 72.29 c 79.98 d Note. Superscript letters reflect significance of the pair-wise comparisons. Means sharing a superscript letter are not significantly different, at p= 0.05 alpha.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 72
Main Study Methods
Participants 325 Mechanical Turk workers (Age; M = 36.4, SD = 10.9) completed this study for a financial incentive. The sample was composed of 41.7% females (97% heterosexual) and
58.3% male (98% heterosexual).
Power calculations were made using effect size estimates from previous experimental studies of self-disclosure and medium. Three studies have experimentally tested the effect of medium on self-disclosure and have found small-medium (d = .37; Coleman, Paternite, &
Sherman, 1999), medium (d = 0.56; Tidwell & Walther, 2002), and large effect sizes (d = .94;
Joinson, 2001). I used the smallest effect size for a conservative estimate of the necessary sample size to achieve 80% power to detect the effect of central importance; an interaction between communication medium and either the HN or HU manipulation. A sample size of
323 participants was required to achieve 80% power to detect a 2-way interaction.
The sample was predominantly liberal, 46% of participants identified as (in general) liberal, 33% as conservative, and 21% as neither conservative or liberal. The sample was
61% socially liberal (identified as either slightly, moderately, or strongly liberal on social issues), 27% of the sample identified as (slightly, moderately, or strongly) conservative, and
12% were neither conservative or liberal. Economic political orientation was evenly distributed: 38% identified as (slightly, moderately, or strongly) liberal on economic issues,
45% identified as (slightly, moderately, or strongly) conservative on economic issues, and
17% identified as neither conservative nor liberal. The average household income was
$51,070 (SD= 39,693) and ranged from $0 to more than $ 365,000. The average life satisfaction was the scale centre, “neither satisfied nor dissatisfied”.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 73
Study Design Data were collected in eight experimental conditions for a two (HN: high or low) x two (HU: high or low) x two (communication medium: video or text) between-subjects design.
Procedure and Measures Participants first reported information about their demographics, including sexual preference. Participants were then presented with one of four possible online dating ads (high
HN/HU, low HN/HU, high HN/low HU, or low HN/high HU), in either video or text format, in the gender that matched their sexual preference. They were asked to carefully read or listen to/watch the profile and compose a message to the author to introduce themselves.
Specifically, participants were told:
We would now like you to write a response to [the target's name] to introduce yourself. Imagine that you came across this profile on a real dating website, and you are now writing a message to [the target's name] with the intent to get to know them. You may write whatever and however much you perceive as appropriate.
Self-Disclosure Participants composed their response in a comment box embedded into the survey.
There were no time or word limits on their response; participants were able to write as little or as much as they wished.
Participant's messages to the target were coded for their depth and breadth of self- disclosure (the main dependent variables) in accordance with the methods of Joinson (2001).
Breadth of disclosure was operationalized as the number of characters in each response (M =
244.16, SD = 187.83). Depth of disclosure was scored using four trained research assistants, each scoring all answers on a 1-10 scale for the depth of intimacy. Each research assistant was given a definition of intimacy as ‘information that makes the discloser feel vulnerable in
Chapter 4: Computer-Mediated Communication and Self-Disclosure 74 some way, for instance, emotionally vulnerable’ (definition taken from Joinson, 2001). Depth and disclosure values were averaged across the raters and used in the main analyses.
Two-way random effects models with absolute consistency assessed the inter-rater reliability, where rater and ratee (i.e., self-disclosure answers) were factors. This type of statistical model accounts for variation that stems from both the raters and the disclosures by modelling these effects as ‘random’. Finally, I used absolute agreement, which is a more conservative estimate compared to consistency agreement (Koo & Li, 2016). Results of the
ICC show that there was excellent inter-rater reliability: the average measure ICC(2,k) was
0.99, with a 95% confidence interval from 0.98 to 0.99 (F(199,597)= 75.7, p<.001; M = 2.5,
SD = 1.66).
Trust Next participants rated the extent to which they trusted the author of the dating profile. The measure of trust was adapted from Hall, Cohen, Meyer, Varley, and Brewer
(2015). This included six attitudinal and quasi-behavioural items that participants responded to on a scale from 1 (definitely not) to 5 (definitely yes). Participants were asked to indicate whether they thought the target (a) was benevolent, (b) had integrity, (c) had the ability to be trustworthy, and (d) was trustworthy, as well as (e) whether they would lend the target money and expect to get it back, and (f) whether they would trust the target with a sensitive secret
(M = 3.14, SD = 0.83, = .86).
Attraction Participants then rated how attractive they found the author of the dating profile. The attraction scale (McCroskey & McCain, 1974) included three subscales: social attraction
(e.g., “I think he/she could be a friend of mine”; M = 4.48, SD = 1.628, =.75), physical attraction (e.g., “I think he/she is quite handsome/pretty”; M = 4.047, SD = 1.636, =.85), and task attraction (e.g., “I couldn't get anything accomplished with him/her”; M = 4.92, SD =
Chapter 4: Computer-Mediated Communication and Self-Disclosure 75
1.40, = .82). All items were measured on a 7-point Likert scale: 7 (strongly agree) to 1
(strongly disagree). The subscales were correlated between r =.32 – .55. Despite the relatively high correlation, the factor structure was consistent with the theorised structure.
The latent structure of the attraction items was estimated with an exploratory factor analysis (EFA) with maximum likelihood estimates and proxmax rotation. Primary factor loading was restricted to .4 or above and items were excluded that had communality lower than .2 (Child, 2006). A three-factor solution resulted, where items loaded consistent with a priori specifications (McCroskey & McCain, 1974). The first factor (social attraction) explained 25% of the variance, the second factor (physical attraction) 24% of the variance, and the third, (task attraction) explained 21%. All remaining items had a primary loading of
0.4 or above and no cross loading (above 0.3).
Humanness and Moral Status Next, participants rated the profile's author for humanness and moral status, using the same scales as in Study 1. Both forms of humanness had acceptable reliability (HU: M =
4.43, SD = 1.01, = .77; HN: M = 4.44, SD = 0.95, = .81). Although the humanness subscales were correlated (r = 0.41) the subscales were treated as distinct for the main analyses, consistent with theoretical distinctions. Likewise, the moral agency (M =3.85, SD =
0.76, =.79) and moral patiency (M = 4.87, SD = 1.29, =.80) were correlated (r = .44) but were treated as distinct, consistent with theory.
Social Presence Participants then completed a number of measures related to the effect of the medium.
First, participants completed a social presence measure. The IPO-SPQ (De Greef &
Ijsselsteijn, 2001) is the most widely used measure of social presence. The scale combines two different methods for measuring social presence: first, participants rate the medium
(either text or video) on a series of bipolar scales (e.g., impersonal - personal), then the
Chapter 4: Computer-Mediated Communication and Self-Disclosure 76 participants rate the extent to which they agree or disagree with a series of statements related to the medium (e.g., “One gets no real impression of personal contact with the people at the other end of the link”) using a 7-point Likert scale (M = 4.53, SD = 0.52, = .83).
Self-Consciousness Next, participants completed the public self-consciousness subscale of the revised self-consciousness scale (SCS-R) (Martin & Debus, 1999). This scale includes nine items
(e.g. “I’m always trying to figure myself out”; M = 2.52, SD =0.67, = .81). Responses were given on a 4-point scale 1 (not at all like me) to 4 (a lot like me).
Perceived-Similarity
Participants then completed the Ten-Item Personality Inventory (TIPI) for a broad and brief measure of their perceptions of the target (Gosling, Rentfrow, & Swann Jr, 2003).
Participants answered this both in terms of themselves (M = 4.11, SD = 0.41) and their perceptions of the target (M = 4.11, SD = 0.43). I then subtracted each item answered for the target from matching item answered for the participant’s own personality. I then then summed the absolute values of the differences to create a composite self-other difference for each participant (M = 0, SD = 0.47). This provides a measure of perceived similarity, with scores closer to 0 indicating more similarity. All items were measured on a 7-point Likert scale: 7 (strongly agree) to 1 (strongly disagree).
Online Expectations Finally, participants completed a novel measure of expectations for CMC. Participants were asked, “Think about when you communicate with others online. When you communicate with someone are they more likely to be ‘...’ online?”
The blank was filled by seven terms which captured expectations of others behaviour: dishonesty, misrepresentation, disclosure, argumentative, flaming, non-judgemental, and
Chapter 4: Computer-Mediated Communication and Self-Disclosure 77 friendly. Responses were measured on fully labelled 7-point Likert scales: 1 (extremely likely) to 7 (extremely unlikely). A final item related to expectations of online disinhibition.
Disinhibition was first defined for participants in plain language terms “Disinhibition is where we have fewer inhibitions and so are more likely behave freely without car…”.
Participants were then asked to rate how often people are disinhibited online using a 5-point scale: 5 (always) to 1 (never).
The latent structure of the expectation items was estimated with an exploratory factor analysis with maximum likelihood estimates and proxmax rotation4. Initial tests of factorability indicated that an EFA was appropriate for these items 5. First, the items friendly and non-judgemental were negatively correlated with the other items and so were reverse scored before proceeding. Primary factor loading was restricted to .4 or above and items were excluded that had communality lower than .2 (Child, 2006). As a result, the disclosure and disinhibition items were excluded for poor communality (.126, .102 respectively) and the non-judgemental item was removed for poor loading (.325) on the first extracted factor. A two-factor solution resulted, where the first factor explained 40.68% of the variance, the second factor 15.90% of the variance All remaining items had a primary loading of .4 or above and no cross loading (above .3). The final factor loadings, communalities, Cronbach’s alphas are shown in Table 7. These items were labelled expectations of aggressive behaviour
(factor 1) and deceptive behaviour (factor 2) given that the items within each subscale relate to either aggressive behaviour (e.g., argue, be unfriendly or unkind) or deceptive (e.g.,
4 A promax solution was preferred as this provided the most defined factor structure as compared to oblimin and varimax.
5 The Kaiser-Meyer-Olkin measure of sampling adequacy was 0.658, above the recommended value of 0.6, and Bartlett’s test of sphericity was significant (c2 (28) = 406.1 p < .05). Six of the 8 items were correlated at least 0.3 with at least one other item. Taken together, these results indicate reasonable factorability of the 8 items.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 78 dishonest, misrepresenting), respectively. The descriptive statistics for these final subscales are presented in Table 8; subscales were correlated at r = .45.
Table 7. Factor Loadings and Communalities
Aggressive Deceptive Communality Behaviour Behaviour
Are more likely to argue online? .89 .65
Are more likely to be aggressive, abusive or .86 .82 unkind online?
Are more likely to be friendly? (reverse scored) .44 .77
Are more likely to misrepresent who they are in .94 .86 terms of their personality, physical appearance, traits or qualities?
Are more likely to be dishonest? .81 .16
Note. Factor loadings < .3 are suppressed
Table 8. Descriptive Statistics for the Two Online Expectations Factors
No. of items M (SD) Cronbach's Alpha
Aggressive Behaviour 3 3.36 (0.45) .76
Deceptive Behaviour 2 2.87 (0.026) .88
Demographics Finally, short demographics were assessed. Participants nominated their household income using an open response form (M= 51,071, SD= 39,693). Next, I measured life satisfaction on a fully labelled 7-point Likert scale where 1 (completely dissatisfied) to
(completely satisfied); (M = 4.89, SD =1.74). Religiosity was measured in two ways: first, participants were asked whether they identified with any religion or denomination with a
Chapter 4: Computer-Mediated Communication and Self-Disclosure 79 binary scale (yes or no). Next, participants rated their religiosity. Participants were asked “To what extent do you identify yourself as a religious person?” and responded with a 5-point
Likert scale, 1 (not at all) to 5 (very much so), (M = 2.32, SD = 1.44). Nationality, age, and gender were described using open-ended responses. Finally, participants identified their political orientation using three items. Participants were asked “To what extent would you describe yourself on each of the following items… liberal issues/economic issues/in general” and responded using a fully labelled 7-point Likert scale, 1 (strongly liberal) to 7 (strongly conservative).
Results
Data Analysis Strategy The data analysis was conducted in four parts: (1) first, a manipulation check assessed the success of the humanness manipulation, (2) main hypotheses were tested, (3) alternative accounts of online behaviour were tested, (4) finally, I explored potential moderators.
For the main analyses, mixed-effects models were fitted with stimuli as a random effect and the three experimental conditions as fixed effects. By fitting a mixed effects model instead of a three-way ANOVA, I am able to statistically account for any effect of actor on the dependent variables and render the responses to these stimuli as generalizable to other, similar stimuli (Baayen, Davidson, & Bates, 2008). ANOVAs rely on averaging across stimuli to obtain cell means and ignore systematic variation between experimental stimuli
(Judd, Westfall, & Kenny, 2012). This variation can contribute to statistically significant mean differences that may not replicate in studies with different stimulus samples and can contribute to type 1 error (Judd et al., 2012; Rietveld & van Hout, 2007). By treating the effect of stimulus as random, I am statistically accounting for the fact that the stimuli used here are a sample from a population of other stimuli that could have been used in this study
(or may be used in future replications). Therefore, I model (or control for) the random effect
Chapter 4: Computer-Mediated Communication and Self-Disclosure 80 that these stimuli have on my outcome measures. Mixed-effects models also have a number of other advantages over ANOVAs: data can be incomplete or unbalanced across conditions, these models can easily accommodate continuous or categorical predictors, and avoids loss of information by averaging across stimuli (Judd, Westfall, & Kenny, 2012).
The main analyses will be variations on a model with three dichotomous fixed-effects:
(1) HN (high or low); (2) HU (high or low); (3) medium (video or text) and a random effect for stimuli. These effects will be allowed to interact, and potential moderators or mediators will be included where the hypothesis requires. This is summarised below:
푆푒푙푓 − 퐷푖푠푐푙표푠푢푟푒 ~ 퐶표푛푑푖푡푖표푛(퐻푁) ∗ 퐶표푛푑푖푡푖표푛(퐻푈) ∗ 퐶표푛푑푖푡푖표푛(푀푒푑푖푢푚) ∗
푀표푑푒푟푎푡표푟 + 푅푎푛푑표푚 퐸푓푓푒푐푡
Correlations and Descriptive Statistics First, to explore the structure of the data set, I produced a correlation matrix with all measured variables (Table 9) and descriptive statistics (Table 10). On inspection of the descriptive statistics, I performed a log-transform on the breadth measure to rectify severe non-normality that may impact main analyses (Shario-Wilk W = 0.82, p < .01; skew = 2.37; kurtosis = 10.69).
Initial examination of the correlations suggests that self-disclosure (breadth and/or depth) is positively associated with many of the measured variables, notably, there was no correlation with task/physical attraction, self-consciousness (contrary to predictions of RC), or perceived similarity (contrary to predictions of SIDE). There was a positive correlation between (depth and breadth.log of) self-disclosure and HN ratings, and a positive correlation between depth of self-disclosure and HU ratings.
Variables related to person perception (humanness, attraction, moral status) were positively correlated with one another. Also notable, perceived similarity was associated with very few variables – only weakly and negatively with HN ratings. Likewise, expectations
Chapter 4: Computer-Mediated Communication and Self-Disclosure 81 were negatively associated with social and physical attraction, implying that pre-interaction expectations influence perceptions of the target.
Finally, polychoric correlations were calculated between medium and each of the dependent measures. Polychoric correlations are appropriate for handling categorical data
(e.g., communication medium). (Holgado–Tello, Chacón–Moscoso, Barbero–García, & Vila–
Abad, 2010). The results are shown in Table 11. There were no significant relationships between communication medium and the dependent measures at the zero-order level.
However, these relationships will be systematically explored, including statistically accounting for the effect of stimuli and participants, in the following analyses.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 82
Table 9. Correlational Matrix for Dependent Measures
Variable 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1. Trust 2. Social Presence .10 3. Condition -.01 -.06 4. Self-Consciousness .10 .06 -.03 5. MP .44** .19** .03 .18** 6. MA .24** .28** -.01 .17** .47** 7. Perceived Similarity -.09 -.02 .06 .06 -.05 -.05 8. HU .39** .08 -.04 .03 .07 .13* -.03 9. HN .40** .11 -.07 .07 .33** .21** -.14* .28** 10. Depth .09 .12* -.07 .01 .21** .18** -.02 .13* .15** 11. Breadth .05 .15** -.06 .01 .21** .21** -.01 .06 .09 .60** 12. Breadth.Log .09 .16** -.02 .01 .27** .24** .00 .09 .14* .55** .83** 13. Aggressive Expectations -.08 .04 .07 .09 .01 .04 .01 -.09 -.05 -.08 -.07 .02 14. Deceptive Expectations -.18** .22** -.08 .14** .03 .16** .08 -.09 -.06 .04 .09 .15** .45** 15. Physical Attraction .30** .05 .03 -.01 .21** .04 -.09 .28** .21** .09 .00 .01 -.12* -.14** 16. Task Attraction .60** .19** .01 .04 .32** .28** -.07 .60** .32** .07 .13* .17** .01 -.00 .32** 17. Social Attraction .43** .08 .04 .06 .30** .19** -.05 .42** .50** .16** .10 .12* -.15** -.16** .44** .55** Note. * indicates p < .05. ** indicates p < .01.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 83
Table 10. Descriptive Statistics for Measured Variables
Variable Mean SD Skew Kurtosis
Self-Disclosure Depth 2.50 1.66 2.02 4.43
Self-Disclosure Breadth 244.16 187.83 2.37 10.69
Self-Disclosure Breadth (log transformed) 2.27 0.35 -1.19 5.10
HN 4.44 0.95 -0.09 0.34
HU 4.43 1.01 0.25 -0.40
MP 4.87 1.29 -1.07 1.91
MA 3.85 0.76 -0.32 -0.36
Public Self-Consciousness 2.52 0.67 -0.01 -0.40
Perceived Similarity 0.00 4.70 0.40 0.51
Social Presence 4.53 0.52 0.03 2.15
Aggressive Expectations 4.34 1.46 -0.25 -0.39
Deceptive Expectations 5.13 1.34 -1.05 1.15
Physical Attraction 4.05 1.43 -0.15 -0.65
Task Attraction 4.92 1.24 -0.44 -0.16
Social Attraction 4.48 1.43 -0.34 -0.60
Trust 3.14 0.83 -0.19 -0.37
Chapter 4: Computer-Mediated Communication and Self-Disclosure 84
Table 11. Polychoric Correlations Between Medium and Dependent Measures
Correlation Coefficient p Value
Trust -.01 .80
Social Presence -.07 .34
Self-Consciousness -.04 .61
MP .04 .63
MA -.01 .94
Perceived-Similarity .07 .35
HU -.05 .47
HN -.09 .26
Depth -.08 .23
Breadth -.07 .30
Breadth.log -.03 .74
Aggressive Expectations .09 .21
Deceptive Expectations -.11 .10
Physical Attraction .03 .58
Manipulation Check and Medium Effects on Humanness
The success of the manipulation was assessed by fitting two mixed-effects models with either HN or HU ratings of the target as dependent measures, with the dummy-coded
HN (high or low), HU (high or low), and medium (video or text) conditions as predictors
(interactions between all three factors), and actor as a random effect. Results of the model are displayed in Table 12 and means for HN and HU in Table 13 and Table 14, respectively. If the humanness dimensions were successfully and independently manipulated, each condition will only have a significant (main) effect on the relevant humanness dimension and there will be no main or interaction effect on the other humanness dimension.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 85
Table 12. Mixed-Effects Model for Assessing the Effectiveness of the Humanness Manipulation
Fixed-Effects Dependent Measures HN ratings HU ratings Intercept 4.72 [3.56, 9.80] 4.08 [0.34, 6.41] HN Condition -0.07 [-0.96, 0.84] -0.58 [-1.35, 0.23] HU Condition -0.65 [-1.46, 0.20] 1.40 [0.54, 2.18] Medium Condition -0.18 [-0.58, 0.19] -0.11 [-0.50, 0.24]
HNcondition*HUcondition 1.28 [0.15, 2.53] 0.09 [-1.03, 1.31]
Medium*HNcondition 0.01 [-0.54, 0.57] 0.34 [-0.19, 0.89]
Medium*HUcondition 0.17 [-0.36, 0.66] -0.14 [-0.69, 0.40]
Medium*HUcondition*HNcondition -0.27 [-1.04, 0.46] -0.27 [-1.05, 0.48] Note. Values are unstandardized coefficients and 95% bootstrapped confidence intervals. Bolded values are considered significant, as the confidence intervals do not contain 0.
Table 13. Means and SD for HN Rating at Different Levels of the Manipulations Condition HN HU Medium Mean SD Low Low Text 4.54 0.91 High Low Text 4.48 0.63 Low High Text 4.06 0.89 High High Text 5.01 1.03 Low Low Video 4.36 0.8 High Low Video 4.31 0.98 Low High Video 4.05 1.07 High High Video 4.74 0.88
Table 14. Means and SD for HU Rating at Different Levels of the Manipulations Condition HN HU Medium Mean SD
Low Low Text 3.98 0.84 High Low Text 3.77 0.77 Low High Text 5.26 0.94 High High Text 4.86 1.04 Low Low Video 3.88 0.75 High Low Video 4.00 0.75 Low High Video 5.00 0.88 High High Video 4.67 0.85
Chapter 4: Computer-Mediated Communication and Self-Disclosure 86
Overall, the manipulation was successful for HU and partially for HN. When predicting HU ratings, there was a main (positive) effect of HU condition on HU ratings and no effect of HN condition on HU ratings. This suggests that the HU manipulation successfully affected HU ratings, independent of the HN manipulation.
However, there was no main effect of the HN condition on HN rating, nor was there a significant main effect of HU condition on HN ratings. Thus, neither the HN nor the HU manipulation, alone, was able to influence HN ratings. Specifically, there was a significant two-way interaction between HN and the HU manipulations; such that the highest HN rating occurred when both the HN manipulation included high HN cues and the HU manipulation included high HU cues. In all other conditions, the HN rating did not significantly differ. To clarify this interaction, simple slopes were plotted (Figure 10), collapsing across the medium manipulation for ease of interpretation. Overall, these results imply that the HU manipulation was successful and any significant results in the main analyses pertaining to the HU manipulation can be interpreted as the consequence of the presence of high or low HU cues.
The HN manipulation is less straightforward, as there was only a significant interaction and no main effects. The HN manipulation only affected HN ratings in combination with the HU manipulation, such that highest HN ratings occurred at high levels of both the HN and HU manipulation. This means that any effect of the HN manipulation in the analyses must be interpreted as an effect of either high HN and HU cues or low HN cues.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 87
Figure 10. Simple slopes for HN ratings for high and low levels of the HN and HU manipulations
More importantly, these mixed-effects models reveal that medium did not significantly affect HN or HU ratings, thus, there is no support for the hyperhumanism (H2).
It was hypothesised that a target presented in text will be perceived as more human than the same target presented via video (H2), as the participant is compensating for the relative lack of cues in text versus video. However, the target was rated equally human – for both HN and
HU – when presented via text or video.
Hyperhumanisation While there is no direct effect of medium on perceptions of humanness, inconsistent with the hyperhumanness account, there may be partial support for this account. Specifically, there may be a main effect of the humanness manipulations on self-disclosure or an interaction between humanness and medium on self-disclosure.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 88
In order to assess whether the humanness manipulation and communication medium influenced self-disclosure, I ran mixed-effects models with either depth, breadth, or breadth.log as the dependent measure, HN (high or low), HU (high or low), and medium
(video or text) as fixed-effects with three way interactions, and actor as a random-effect
(Table 15). There were no significant main or interaction fixed-effects for depth: the condition (humanness perception) nor the medium influenced the participant’s depth of self- disclosure. However, medium was a significant predictor of breadth, such that messages had less breadth in the video condition than in the text condition. This main effect of medium on self-disclosure breadth and breadth.log is consistent with the hypotheses of the hyperpersonal, RC, and SIDE accounts (H1).
There was no evidence for hypotheses H3 or H4, which predicted that HN both directly facilitate self-disclosure and as well as interact with the effect of medium on self- disclosure. Instead, the HN manipulation had no effect on self-disclosure nor significant interaction with medium. As the high HN condition was both high in HN and high in HU cues, any (hypothesized) positive effect of high HN cues on self-disclosure may be supressed by the (hypothesized) negative effect of high HU cues on self-disclosure.
However, there is support for the hypotheses related to HU (H5 and H6). These results suggest that messages sent in the high HU condition had marginally less breadth than messages sent in the low HU condition. There was also a positive interaction between HU and medium. To aid in the interpretation of this interaction, simple slopes were plotted
(Figure 11). These results show that the (negative) effect of the HU manipulation on breadth of self-disclosure is greater in the text condition than in the video condition. Consistent with hypotheses (H6), when in the text condition participants were more sensitive to the humanness cues when deciding how much to disclose, compared to those in the video condition. The results of the mixed models are presented in Table 15.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 89
Thus far, results for the hyperpersonalism theory are mixed. There was no support for
H2; that is, there was no main effect of medium on humanness (HN or HU, see Table 12).
Thus, the results imply that medium by which the target is presented does not impact humanness. However, the interaction between the HU manipulation and medium is consistent with predictions from hyperpersonalism. That is, the effect of the HU manipulation on self- disclosure was greater in the text than in video condition.
To ensure that it was differences in HU (manipulation) affecting self-disclosure in the above described model (and not some confounding variable), an additional model was fitted using HU/HN ratings as fixed effects instead of the HU/HN manipulation (Table 15). There was no significant main effect of HU ratings on self-disclosure (breadth or depth), nor any interaction between HU ratings and medium on self-disclosure. This suggests that it is not
HU that is affecting self-disclosure, but rather some other confounding factor incidentally manipulated in the HU manipulation. Therefore, some other confounding variable manipulated with the HU condition is affecting self-disclosure (both in the main effect and interacting with condition).
Chapter 4: Computer-Mediated Communication and Self-Disclosure 90
Table 15. Mixed Effects Models That Assessed Hyperpersonal/Hyperhumanness Hypotheses
Dependent Variable Fixed Effects Breadth Breadth.log Depth Intercept 293.41 2.54 2.95 [256, 533] [2.32, 2.76] [2.25, 4.45]
Medium Condition -94.81 -0.16 -0.44 (Text = 0, Video = 1) [-173, -6.47] [-0.30, -0.02] [-1.23, 0.24] HU -180.59 -0.13 -0.75 (Low = 0, High = 1) [-351, 18.00] [-0.73, -0.005]# [-0.23, 1.00] HN -112.54 -0.23 -0.51 (Low = 0, High = 1) [-281, 77.3] [-0.56, 0.10] [-0.23, 1.11] HN Rating - 0.08 0.055 [-0.11, 0.41] [-1.13, 1.30] HU Rating - -0.17 0.15 [-0.18, 0.29] [-1.26, 0.94] Two-Way Interactions HN*HU 89.01 0.16 0.50 [-189, 340] [-0.35, 0.065] [-1.75, 2.79] Medium*HN 72.90 0.14 0.26 [-44.8, 177] [-0.09, 0.35] [-0.71, 1.41] Medium*HU 110.30 0.22 0.29 [-17.3, 227] [-0.001, 0.46]# [-0.82, 1.30] HN rating * HU rating - 0.05 -0.01 [-0.06, 0.04] [-0.18, 0.29] Medium * HN rating - -0.13 -0.12 [-0.50, 0.25] [-1.88, 1.61] Medium * HU rating - 0.32 0.00 [-0.36, 0.36] [-1.35, 1.99] Three-Way Interactions Medium*HN Ratings *HU - -0.03 0.007 Ratings [-0.07, 0.08] [-0.40, 0.33]
Medium*HU*HN -70.81 -0.12 -0.18 [-217, 112] [-0.42, 0.22] [-1.64, 1.18]
Note. Values are unstandardized coefficients and 95% bootstrapped confidence intervals. Bolded values are considered significant, as the confidence intervals do not contain 0. # = marginal significance as the t-value exceeds 1.92 but the confidence interval includes 0.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 91
Figure 11. Simple slopes for breadth.log ratings for high and low levels of the HU manipulation and medium.
Altogether, these analyses provide partial support for hyperpersonalism in impression formation in CMC: there was no effect of medium on humanness (H2); the HN manipulation had no main effect on self-disclosure (H3) and did not interact with medium (H4) to effect self-disclosure; there was a marginal (negative) effect of the HU manipulation on self- disclosure breadth (H5); and consistent with H6, the effect of the HU manipulation was larger in text than in the video condition. However, when the model was refit with HU and HN ratings instead of the dummy-coded manipulation, no effects were significant, suggesting that the differences in HU is not affecting self-disclosure. Taken together, these results suggest that people are more sensitive to some cues pertinent to self-disclosure when interacting via a lean media (i.e., text) compared to a rich media (i.e., video). However, it’s not clear from these models what cues are driving self-disclosure, as differences in humanness ratings do not account for medium effects on self-disclosure. Further, the results show no evidence of hyperpersonalism (or exaggerated humanness perceptions) in text versus video, as theorised by the hyperpersonalism account.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 92
Reduced Cues Theory Given the lack of support for hyperhumanness as an explanation for medium effects on self-disclosure, I tested whether there was support for the predictions made by the RC theory. The RC theory predicts that there is less self-consciousness in the text condition, compared to the video condition, due to the filtering out of non-verbal social cues (H7). This reduction in self-consciousness is hypothesized to mediate the positive effect of communication medium on self-disclosure (H8).
Again, these hypotheses were assessed with mixed-effects models (Table 16). First, to assess H7, a model was fit predicting self-consciousness, where medium was a fixed effect and stimulus was a random effect. There was no support for the hypothesis: there was no effect of medium on self-consciousness, implying that self-consciousness was equivalent in the text and video conditions.
Next, two additional models were required to assess H8: one predicting breadth of self-disclosure, and, separately, another predicting depth of self-disclosure. Medium and self- consciousness were fixed effects (with no interaction) and stimulus was a random effect. The
‘causal mediation’ package in R (described here: Tingley, Yamamoto, Hirose, Keele, & Imai,
2014) was used to estimate the direct effects of medium on self-disclosure, indirect effects via self-consciousness, the total effect on self-disclosure, and the proportion of the total effect that is the indirect effect from the mixed effects models specified above. This process is theoretically equivalent to mediation estimated with linear regression (e.g., Hayes, 2017), except this method also allows the specification of a random effect (in this case, stimulus).
Results show that there was no evidence to support H8, as neither the direct or indirect effects were significant (Table 17). Together, these analyses demonstrate that: medium did not impact participant’s self-consciousness (Table 16), self-consciousness did not impact self-disclosure (independent of medium) (Table 16), and there was no indirect
Chapter 4: Computer-Mediated Communication and Self-Disclosure 93 effect of medium on via self-consciousness (Table 17). Thus, there was no support for the RC account of self-disclosure in CMC.
Table 16. Mixed Effects Model fitted to assess H7 and H8
Dependent Variable
Fixed Effects Self-Consciousness Depth Breadth.log
Intercept 2.55 [2.41, 2.68] 2.55 [2.41, 2.68] 2.28 [2.07, 2.48]
Medium -0.05 [-0.19, 0.09] -0.05 [-0.19, 0.09] 0.004 [-0.05, 0.06]
Self-Consciousness 0.04 [-0.58, 0.17] -0.015 [-0.09, 0.06]
Table 17. Estimated Mediation Effects of Self-Consciousness on Medium and Self-Disclosure
Depth of Self-Disclosure Breadth.Log of Self Disclosure
Estimated Effect [95% CI]
Indirect Effect of Medium on -0.0001 [-0.03, 0.03] p=.99 -0.0003 [-0.006, 0.00], p=.88
Disclosure (Via Public Self-
Consciousness)
Direct Effect of Medium on -0.21 [-0.77, 0.13], p=028 -0.017 [-0.09, 0.06], p=.65
Disclosure
Total Effect -0.20 [-.05, 0.13], p=.22 -0.017 [-0.09, 0.06], p=.65
Proportion Mediated -0.001 [-0.28, 1.16], p=.98 0.0007 [-0.43, 0.63], p=.96
Note. Estimates were produced with 1,000 quasi-Bayesian Monte Carlo simulations
Social Identity and Deindividuation Effects
Finally, I tested whether the SIDE model could account for the medium effect on self- disclosure. I assessed whether there was greater perceived-similarity in the text condition than in the video condition (H9) and if perceived-similarity mediated the effect of medium on
Chapter 4: Computer-Mediated Communication and Self-Disclosure 94 self-disclosure (H10) (Table 16). The same analysis method used for H7 and H8 was used for
H9 and H10. Again, there was no support for H7 or H8 (
Table 19). Participants in the video and text condition did not differ in their perceived- similarity, and perceived-similarity did not indirectly affect self-disclosure via medium. As was the case for RC, there was no evidence to support the claims of the SIDE model relating to self-disclosure.
Table 18. Mixed Effects Model fitted to assess H7 and H8
Dependent Variable
Fixed Effects Perceived-Similarity Depth Breadth.log
Intercept -0.81 [-2.60, 0.90] 2.85 [2.15, 3.57] 2.29 [2.18, 2.42]
Medium 0.54 [-0.45, 1.68] -0.21 [-0.57, 0.16] -0.02 [-0.09, 0.06]
Perceived- 0.0002 [-0.004, 0.04] 0.0003 [-0.008, 0.009]
Similarity
Note. Values are unstandardized coefficients and 95% bootstrapped confidence intervals. Bolded values are considered significant, as the confidence intervals do not contain 0.
Table 19. Estimated Mediation Effects of Assumed-Similarity on Medium and Self-Disclosure
Depth of Self-Disclosure Breadth.Log of Self Disclosure
Estimated Effect [95% CI]
Indirect Effect of Medium on 0.0008 [-0.03, 0.02] p = .96 0.0002 [-0.006, 0.01], p = .96
Disclosure (Via Perceived
Similarity)
Direct Effect of Medium on -0.20 [-0.56, 0.14], p = .14 -0.017 [-0.09, 0.06], p = .65
Disclosure
Total Effect -0.21 [-.06, 0.14], p = .14 -0.017 [-0.09, 0.06], p = .69
Proportion Mediated -0.00003 [-0.35, 0.31], p = .99 0.0004 [-0.52, 0.79], p = .98
Chapter 4: Computer-Mediated Communication and Self-Disclosure 95
Note. Estimates were produced with 1,000 quasi-Bayesian Monte Carlo simulations
Exploratory Analysis Given the lack of support for my hypotheses, I conducted exploratory analyses to identify what other factors may explain the effect of condition on breadth of self-disclosure.
Specifically, I explore the role of three factors expected to affect self-disclosure in CMC: (1) social presence, (2) trust, (3) attraction, and (4) expectations for the interaction.
Social Presence One possibility hypothesized is that social presence is, on average, less in the text condition than in the video condition. This difference in social presence across medium may account for differences across media in self-disclosing behaviour. A mediation model was fitted to test the possibility whether social presence mediated the effect of medium on self- disclosure (Table 20). There was no significant mediation in the resulting model, suggesting that social presence does not directly affect self-disclosure nor mediate the relationship between medium and self-disclosure.
Table 20. Estimated Mediation Effects of Social Presence on Medium and Self-Disclosure
Breadth.Log of Self Disclosure
Estimated Effect [95% CI]
Indirect Effect of Medium on -0.0005 [-0.007, 0.01], p = .84
Disclosure
(Via Social Presence)
Direct Effect of Medium on Disclosure -0.014 [-0.09, 0.06], p = .71
Total Effect -0.014 [-0.09, 0.06], p = .71
Proportion Mediated 0.0009 [-0.83, 0.53], p = .97
Note. Estimates were produced with 1,000 quasi-Bayesian Monte Carlo simulations
Chapter 4: Computer-Mediated Communication and Self-Disclosure 96
Trust6 Next, a model was fitted to test the possibility that the relationship between medium and self-disclosure was moderated by trust. Given that trust is an important antecedent of self-disclosure, the hypothesized relationship between communication media and self- disclosure may only be observed when target is also trusted.
The results (see Table 21) show that trust is a significant predictor of breadth.log, consistent with theorising, however, there was no interaction between trust and medium when predicting either breadth or depth of self-disclosure. Therefore, the relationship between medium and self-disclosure is not impacted by how trusted the target is.
Table 21. Mixed Effects Model fitted to Assess Trust as a Moderator
Dependent Variable
Fixed Effects Depth Breadth.log
Intercept 1.75 [0.73, 2.72] 2.04 [1.83, 2.24]
Medium 0.69 [-0.70, 2.19] 0.24 [-0.05, 0.57]
Trust 0.28 [0.0007, 0.57] 0.07 [0.012, 0.14]
Trust * Medium -0.30 [-0.73, 0.13] -0.08 [-0.18, 0.008]
Note. Values are unstandardized coefficients and 95% bootstrapped confidence intervals. Bolded values are considered significant, as the confidence intervals do not contain 0.
Attraction Next, attraction was explored as a possible mediator of the effect of medium on self- disclosure. One possibility is that targets were perceived as more attractive in text than in video, leading to more self-disclosure in text than in video. There were only zero order correlations between self-disclosure (breadth) and social/task attraction (Table 9), therefore,
6 Although it was also hypothesized that trust mediated the relationship between hyperhumanization and self-disclosure, there was no evidence of hyperhumanization in impression management in text and so this possibility was not explored.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 97 for simplicity, these analyses excluded physical attraction. There were no significant effects in the resulting models (Table 22), suggesting that attraction does not mediate any effect of medium on self-disclosure.
Table 22. Estimated Mediation Effects of Attraction on Medium and Self-Disclosure
Breadth.Log of Self Disclosure
Model for Task Attraction Estimated Effect [95% CI]
Indirect Effect of Medium on Disclosure 0.0009 [-0.012, 0.01], p = .82
(Via Task Attraction)
Direct Effect of Medium on Disclosure -0.015 [-0.09, 0.07], p = .69
Total Effect -0.014 [-0.09, 0.07], p = .74
Proportion Mediated 0.007 [-1.55, 1.71], p = .96
Model for Social Attraction
Indirect Effect of Medium on Disclosure 0.004 [-0.012, 0.01], p = .44
(Via Social Attraction)
Direct Effect of Medium on Disclosure -0.018 [-0.09, 0.06], p = .63
Total Effect -0.014 [-0.09, 0.06], p = .71
Proportion Mediated -0.017 [-2.01, 1.71], p = .89
Note. Estimates were produced with 1,000 quasi-Bayesian Monte Carlo simulations
Online Expectations Finally, the possible moderating effect of expectations was assessed. Expectations may be more heavily utilised in impression formation in the text condition than in the video condition, as a compensatory action for the lack of social cues in the text condition.
Therefore, expectations may have a larger effect on self-disclosure when the target is presented via text compared to video.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 98
Two mixed-models were fitted to explore this possibility. Expectations for aggression, expectations for deception, and medium were (interacting) fixed effects, predicting self- disclosure (breadth.log or depth). There was a significant effect for the expectations for aggression on both self-disclosure measures when predicting self-disclosure depth (
Table 23). Expectations for aggression had a negative main effect on both the breadth and depth of self-disclosure. That is, the more that a participant expected other people to be aggressive when communicating online, the less inclined they were to disclose both in intimacy (depth) and in quantity (breadth).
Table 23. Mixed Effects Models for Expectations of Interactions and Self-Disclosure
Breadth (log) Depth
Main Effects Intercept 2.52 [1.19, 3.81] 3.07 [-3.0, 8.09]
Condition -0.09 [-5.35, 0.17] 0.49 [-2.45, 3.79]
Expectations for Aggression -0.08 [-.17, -.004] -0.46 [-0.86, -0.09] Expectations for Deception -0.003 [-.064, .057] -0.125 [-0.44, 0.17]
Two-Way Interactions
Aggressive Expectations * 0.013 [-.002, .030] 0.06 [-0.004, 0.13] Deceptive Expectations Aggressive Expectations * 0.062 [-0.13, 0.25] 0.05 [-0.80, 0.83] Condition Deceptive Expectations * -0.01 [-0.14,0.11] -0.31 [-0.98, 0.31] Condition
Three-Way Interactions
Aggressive Expectations * -0.005 [-0.04, 0.02] 0.03 [-0.12, 0.17] Deceptive Expectations * Condition
Chapter 4: Computer-Mediated Communication and Self-Disclosure 99
Discussion Overall, this study did not find clear support for any of the RC, or SIDE accounts of self-disclosure and only partial support for the hyperpersonal/CFO account. As predicted by all theories, the quantity (or breadth) of self-disclosure was greater in the text condition than in the video condition. However, none of the theories examined were able to identify what aspect of the medium drives this effect. There was mixed support hyperpersonalism in impression formation in CMC as an explanation of the medium effect on self-disclosure: while there was no effect of medium on humanness (H2) and the HN manipulation had no main or interaction effect (with medium) on self-disclosure (H3/H4), there was a marginal
(negative) effect of the HU manipulation on self-disclosure breadth (H5) and the effect of the
HU manipulation on self-disclosure was greater in text than in the video condition (H6).
Taken together, these results suggest that certain cues relevant to self-disclosure were more influential when presented in text than in video, consistent with hyperpersonalism. However, it’s not clear which cues were exaggerated in text compared to video, as there was no evidence of exaggerated humanness perceptions in text versus video, as theorised by the hyperpersonalism account.
There was no support for the RC account of self-disclosure. Self-consciousness was equivalent in the text and video condition and self-consciousness did not mediate the effect of medium on self-disclosure. There was also no support for the SIDE model: the video and text condition did not differ in their perceived-similarity, and perceived-similarity did not indirectly affect self-disclosure via medium.
Finally, exploratory analyses found that social presence, trust, nor attraction were unable to explain medium effects on self-disclosure. However, the results of this study provide preliminary evidence to suggest that pre-interaction expectations influence self- disclosing behaviour in online environments.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 100
Self-Disclosure Effects Consistent with previous empirical findings, there was more breadth of self-disclosure in the cue-poor condition (text) than in the cue-rich condition (video) (Antheunis,
Valkenburg, & Peter, 2007; Barak & Gluck-Ofri, 2007; Coleman et al., 1999; Joinson, 2001;
Kiesler, Zubrow, Moses, & Geller, 1985). This implies that some quality of the medium has an effect on self-disclosure; unfortunately, the results of this study were unable to clarify what qualities of the medium underpin self-disclosing behaviour.
The main hypothesis, derived from hyperpersonal theory and cues-filtered-out accounts of CMC behaviour, provided a partial explanation for the medium effect on self- disclosure. According to the hyperpersonal theory, people will compensate for the absence of certain non-verbal cues in the text condition by exaggerating the relative importance of verbal cues when forming impressions of others (Walther, 1996). Thus, when humanness cues are available in the text condition, those cues will be weighted more heavily when perceiving the target than when that same target (and same humanness cues) is presented via video. As a result, the target should be perceived as more human when humanness cues are presented in the text condition. However, there was no interaction between medium and cue (HN or HU) on humanness perceptions, thus perceptions of humanness were not exaggerated in the text condition, or, in other words, there is no hyperhumanisation.
Consistent with hyperpersonalism, there was evidence that participants exaggerated some aspect of the target in the text condition compared to the video condition, as the HU manipulation in the text condition had a greater effect on self-disclosure than in the video condition. Specifically, the negative effect of HU on self-disclosure was greater when in the text condition than in the video condition. Therefore, certain cues were more influential when determining self-disclosure for those in the text condition than those in the video condition.
However, it’s not clear what aspect of the HU manipulation is more influential in the text condition than in the video condition. Additional analyses found that ratings of humanness
Chapter 4: Computer-Mediated Communication and Self-Disclosure 101
(HN and HU) did not predict self-disclosure directly nor interact with medium to predict self- disclosure. It is possible that some other, confounding variable, manipulated incidentally with
HU, was exaggerated in the text condition (relative to the video condition). Therefore, an unmeasured variable associated with HU and self-disclosure was exaggerated in text
(compared to video) leading to more (and fewer) self-disclosures in the text condition at high and low levels of the HU manipulation.
The RC account was also unable to clearly account for this self-disclosure result.
According to this theory, the absence of non-verbal cues in the text condition impedes the communication of social context and associated norms, which results in reduced self- consciousness, less awareness of expected behaviour, and disinhibited behaviour (Joinson,
2001; Suler, 2004). However, there was no evidence that the self-consciousness of the participant changed as a function of medium: participants were equally self-conscious when in the text and video conditions and self-consciousness did not mediate the effect of medium on self-disclosure. Thus, there is no evidence to suggest that participants felt less concerned about negative evaluation of others or less aware of the social norms for behaviour when communicating with the target presented as a text, compared to the target presented as a video. Therefore, self-consciousness and disinhibition does not account for this effect of medium on self-disclosure.
An alternative interpretation of the negative result for self-consciousness is that the design of this study allows the participant to behave without expectation of evaluation.
Participants submitted a response to the target without necessarily expecting a response, thus implying that they will not be evaluated by the target. This may have resulted in a floor effect for self-consciousness. Instead, the mean value for self-consciousness was near the mid-point of the scale (M = 2.57, SD = 0.67, on a 1 - 4 scale). Therefore, it is unlikely that this negative result is due to the study design availing concerns of social evaluation.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 102
The SIDE model was also unable to explain the effect of medium on self-disclosure.
The SIDE model argues that the absence of visual cues in some media emphasizes people’s social identity and minimises their personal identity (Reicher, Spears, & Postmes, 1995).
According to the model, the absence of individuating information in the text condition (in particular, the filtering of some visual cues) leads the participant to categorise the target and themselves according to a shared group identity (i.e., online daters), and emphasizes expectations for behaviour in dating sites. Therefore, participants may have been more likely to self-disclose in the text condition because they were more sensitive to group norms of expected self-disclosure in online dating forums, compared to those in the video condition.
However, there were no differences in perceived similarity across the communication media, nor did perceived similarity mediate the effect of medium on self-disclosure. This result implies that the text condition did not filter out individuating cues that would emphasize a shared social identity. Thus, the SIDE model is unable to account for this self-disclosure effect.
One possible explanation of the medium effects on self-disclosure is a systematic difference in ecological validity between the text and video conditions. That is, the effect of communication medium on self-disclosure may be confounded by ecological validity. Online dating typically utilises primarily text as a communicative medium, for example, static images are paired with text descriptions of the person and the two people send text messages to communicate. In comparison, a video dating profile is an atypical format for online dating.
Thus, the text condition has a higher ecological validity than the video condition. The text condition may have appeared more consistent with participant’s experiences of and expectations for online dating than the video condition, making salient norms regarding how people should behave in online dating environments. As noted earlier, self-disclosure is often expected in text-restricted contexts where two unacquainted individuals are getting to know
Chapter 4: Computer-Mediated Communication and Self-Disclosure 103 one another (Tidwell & Walther, 2002). Likewise, reciprocating self-disclosures after a communicative partner has self-disclosed (e.g., in the form of a dating biography) is also a norm for behaviour (Joinson, 2001). Thus, participants in the text condition may have found the dating profile more representative of real dating profiles, making salient norms related to self-disclosure, in turn leading to greater breadth of self-disclosure compared to the video condition. However, according to this explanation, we would also expect that the ecological validity of the text condition might also increase the salience of expectations related to online interactions, yet there were no significant interactions between expectations and condition
(Table 23).
Unusually, there was no result of medium on intimacy (depth) of disclosure, despite previous empirical work finding such an effect (Coleman et al., 1999; Joinson, 2001). It is possible that the study design was not conducive to studying intimate self-disclosures.
Specifically, this study featured one-way interaction; participants were not reciprocally interacting with our target here, but instead, only formed initial impressions based on a dating profile. Previous studies that have found a depth effect have studied self-disclosure in a real, bi-directional interaction (Coleman, Paternite, Sherman, 1999; Joinson, 2001). Furthermore, despite efforts to make the page as ecologically valid as possible, participants remained, presumably, aware that they were involved in a study when composing their message to the target. As a result, participants may not have felt comfortable sharing intimate information that they might share when messaging a real dating profile. On inspection of the distribution of depth scores, a floor effect is apparent (M = 2.5, SD = 1.66, scale ranges from: 0 - 10). This supports the possibility that the study design was not appropriate for the measurement of intimate self-disclosures. This floor effect may also explain why social presence did not vary between conditions, that is, because the participants did not feel that they were realistically embedded in a real interaction.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 104
Humanness Perceptions As noted above, there was no effect of medium on humanness ratings (neither main effect nor interaction with the manipulation), this may be the consequence of the study design. Hyperhumanisation effects may require bi-directional interaction and time to manifest. Social Information Processing theory of CMC argues that the text communication is less efficient when compared to richer media (such as FtF or video), causing interpersonal processes in text to become ‘retarded’ (Walther & Burgoon, 1992). Therefore, one possibility is that given bi-directional interaction (and so more time to communicate humanness), humanness perceptions in the text condition would surpass the video condition.
Although medium did not have a negative effect on humanness perception (i.e., hyperhumanisation), it is surprising that there was not, instead, a positive effect of medium on humanness (i.e., dehumanisation). Schroeder and Epley (2016) found that the addition of non-verbal cues in voice facilitated ratings of humanness compared to text. Likewise, study 1 partly replicated the results from Schroeder and Epley (2016); targets presented via audio were perceived as more uniquely human than those same targets presented in text. Thus, in lieu of hyperhumanisation, a dehumanising effect of text was expected.
One possibility is that the addition of visual cues in the video condition actually undermined the humanising effect of voice identified in Study 1 and Schroeder and Epley
(2016). The visual cues in the current study may have either made the stimuli less believable or the visual cues may have conferred qualities low in humanness. Schroeder and Epley
(2016) found that video and voice presentations of a target did not differ in humanness ratings, implying that visual cues do not confer any additional humanness beyond the humanising effect of voice. While the authors found that visual cues did not confer any additional humanness to the effect of voice, in the current study, the visual cues may have instead undermined the effect of the voice cues on humanness.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 105
An important methodological difference between the two studies may explain the differing effect of presenting a target via video on humanness perception. Schroeder and
Epley (2016) recorded the video and audio of people describing their own emotion experiences and then transcribed this to text. The current study had the opposite order of stimuli development: the text stimuli were composed first (to manipulate humanness) and second, actors read the transcriptions aloud. As a result, the transcriptions were not the actor’s own words, and this may have made the video recordings (1) less realistic and/or (2) conferred qualities low in humanness, such as low emotionality. Therefore, those presented via video may have been perceived as less human than expected, according to previous research, because their recordings were not imbued with emotionality (a quality related to high humanness). Alternatively, these video recordings were not perceived as authentic representations of the target and thus any high humanness cues conveyed in the video may not have been attributed to the person.
If authenticity of the cue-rich condition affected perceptions in Study 2, we might also expect this to apply to Study 1. Although Study 1 also had actors read predetermined transcriptions, the richer the media in non-verbal cues, the more people leak unintended nonverbal cues that affect person perception (Walther, 1996). As such, people may have leaked more unintended non-verbal cues in the video condition in Study 2 than in the voice condition in Study 1. For example, people in the video condition may have had stiff posture or failed to use non-verbal cues consistent with the expression of emotion (Ekman, 2004), undermining the ecological validity of the stimuli and/or perceptions of humanness.
However, those only recording their voice (such as in Study 1) would be able to hide this stiff posture and other visual cues inconsistent with emotionality and authenticity.
Chapter 4: Computer-Mediated Communication and Self-Disclosure 106
Expectations While online self-disclosure could not be explained by any of the measured CMC theories, the novel expectations scale did significantly predict self-disclosing behaviour.
Specifically, the extent to which participants expect others to behave aggressively online significantly predicted both reduced intimacy and less breadth of self-disclosure. The effect of expectancies for aggression on self-disclosing implicates the possibility that expectancies may have an adaptive role in informing the approach to a new social situation. Those who expected more aggression may psychologically distance themselves during an interaction in a pre-emptive attempt to reduce negative affective consequences for the expected aggressive behaviour, leading them to engage less with the target (i.e., disclose less). Studies of construal level theory in the offline domain suggest that more psychological distance will unwanted negative affect changes, such as less distress when watching violent media
(Williams & Bargh, 2008). Although it is unknown whether psychological distance may be adaptively implemented in interpersonal contexts, it has been found to be adaptively implemented in intra-personal contexts. When attempting to self-reflect, adaptive use of psychological distancing can allow self-reflection in a way that ‘buffers’ the experience of negative affect (e.g., Ayduk & Kross, 2010; Kross, Ayduk, & Mischel, 2005). According to this logic, people may equally use adaptive psychological distance when expecting aggressive behaviour, thereby ‘buffering’ themselves from the negative affective consequences, leading them to be less willing to self-disclose. Therefore, the results of this study provide preliminary evidence to suggest that expectations influence the behaviour and experience of the online domain, when approaching a new social situation.
Surprisingly, there was no interaction between medium and expectations when predicting self-disclosure. Previous research has shown that expectations are a source of information that can compensate for the absence of individuating information in text, thus expectations should be more important when someone is presented with text than with video
Chapter 4: Computer-Mediated Communication and Self-Disclosure 107
(Epley & Kruger, 2005; Walther & Tong, 2014). The ambiguous nature of text interactions should increase the utility of these expectancies when deciding whether to engage with the target or not. However, this study did not reveal any evidence for different effects of expectancies in different media. Instead, expectations were equally important to self- disclosures in both video and text conditions. One possibility is that the two conditions were equally ambiguous, that is, the non-verbal cues in the video condition did not reduce uncertainty associated with forming an initial impression of someone. This may be related to
(assumed) low ecological validity of the video condition: participants may have felt that the videos were not genuine and thus did not utilise the additional non-verbal cues present in the video (but not in text) when forming their impression of the target. Instead, both the video and text condition required participants to draw on their general expectations for online interactions to inform their perception of the target and behaviour. This may also explain why there are no mean differences in the perception of the target. That is, if the non-verbal cues in the video condition were not utilised to inform impression formation because they were not perceived as genuine, this would also explain why there were no mean-differences in the perception of the target. As a result, future studies should test whether these expectancies are more or less utilised in real interactions in cue-poor and cue-rich media.
Conclusion The results of this study imply that communication medium does affect some aspects of perception and behaviour (i.e., self-disclosure). Specifically, self-disclosure was greater when perceiving someone via text compared to video. Likewise, there was evidence that some aspect of person perception was exaggerated in the text condition, relative to the video condition. That is, the HU manipulation, which negatively predicted self-disclosure, had a greater effect on self-disclosure in the text condition than in the video condition.
Unfortunately, this study was unable to identify what aspect of the manipulation was
Chapter 4: Computer-Mediated Communication and Self-Disclosure 108 exaggerated in text and accounted for the effect on self-disclosure. Additional analyses revealed that neither HN nor HU ratings predicted self-disclosure or varied with medium.
Further, there was no evidence to support any of the alternative theories of online behaviour.
These results raise questions as to whether cues-filtered-out theories of CMC behaviour are applicable to positive moral behaviours in CMC.
Instead, I find evidence that expectations for online interactions may, in part, account for moral behaviours in online contexts. Expectations of aggressive behaviour in online interactions predicted both the amount of self-disclosure and the intimacy of that disclosure.
These results suggest, instead, that some medium effects, including an increased tendency to self-disclose in CMC environments relative to FtF interactions, are not due to the absence of cues in CMC but are the result of individual differences (e.g., expectations). In other words, I the way in which people perceive others and behave online appears to be driven by qualities of the perceiver, and not necessarily qualities of the target or the medium.
Chapter 5: Discussion of Stream 1 109
Chapter 5: Discussion of Stream 1 Across two studies there was limited support for a CFO model when applied to two specific moral behaviours: self-disclosure and flaming. Central to my application of the CFO model to online moral behaviour is the claim that text communication filters out the majority of non-verbal social and contextual cues available in richer media (e.g., video, FtF, voice) and this absence of non-verbal cues in text leads to changes in person perception, including humanness perception, and moral behaviour. My application of the CFO model to a negative moral behaviour, flaming, posited that text would filter out non-verbal cues important to humanness perception, resulting in dehumanisation, the denial of moral status, and moral disengagement that licenses flaming. However, results were paradoxical: while there was dehumanisation in the text condition and dehumanisation was weakly related to flaming, and there was more flaming in the voice condition. These results imply that some aspect of medium affected moral behaviour, although neither CFO nor any other major theoretical account of online behaviour was able to explain what particular quality of the medium was driving the effect of medium on flaming.
Study 2 provided a second test of the CFO, but in the context of a positive moral behaviour, self-disclosure. In addition, Study 2 allowed for the exploration of alternative accounts of CMC behaviour. Again, there was mixed support for the hypotheses: while there was more self-disclosure in the text condition, it was unclear what aspect of the medium was driving this effect on self-disclosure. Although the human uniqueness manipulation significantly (and negatively) predicted self-disclosure and the effect of this manipulation on self-disclosure was greater in the text condition than in the video condition, follow-up analyses reveal that it was not humanness ratings driving this effect. Instead, some other aspect of the manipulation influenced self-disclosure. Further, all other prominent CMC theories examined in this study (SIDE, RC) were also unable to identify what aspect of the
Chapter 5: Discussion of Stream 1 110 medium influenced self-disclosure. Thus, the evidence suggests that medium exerts an effect on person perception such that qualities related to self-disclosure and humanness are exaggerated in cue-poor media and this has downstream effects for moral behaviour.
However, similar to Study 1, Study 2 was also unable to identify what quality of person perception was impacted by medium to affect self-disclosure.
Medium Effects on Moral Behaviour Across two studies there is evidence that medium does affect both positive and negative moral behaviours. Study 1 demonstrated that listening to the voice of someone who disagrees with your political opinion can lead to more aggression than reading an equivalent text passage. While Study 2 shows that, under certain circumstances, reading a dating profile can lead to more self-disclosure than watching a video of an equivalent dating profile. As noted previously, although findings of medium effects of flaming and self-disclosure are mixed, the results of these studies add to the CMC literature that confirms that medium can affect certain online behaviours (Abrams, 2003; Antheunis et al., 2007; Barak & Gluck-Ofri, 2007; Cho &
Kwon, 2015; Coleman et al., 1999; Joinson, 2001; Kiesler et al., 1985; Liao, Bazarova, &
Yuan, 2018; Moor et al., 2010; Verma, Nitin, & Srivastava, 2016). These results also add to the moral psychology literature by highlighting the possibility that the medium by which a target is presented, or people are interacting can affect moral behaviours and morally relevant perception.
Taken together, these the two studies indicate that the absence of cues in text can have positive consequences for behaviour, rather than the negative implications that many researchers have suggested (Hmielowski, Hutchens, & Cicchirillo, 2014; Jane, 2015; Lapidot-
Lefler & Barak, 2012; Sproull & Kiesler, 1991; Suler, 2004). Study 1 found that the cue-poor condition (text) lead to less aggression than the cue-rich condition (voice) and Study 2 found that the cue-poor condition (text) lead to more breadth of self-disclosure, a positive moral
Chapter 5: Discussion of Stream 1 111 behaviour, compared to the cue-rich condition (video). Although these are only two examples of online behaviour in two constrained circumstances (i.e., online dating and political discussion forums), these results have positive implications for behaviour in CMC contexts.
Many researchers and lay people have argued that CMC has negative implications for interpersonal behaviour (Hmielowski et al., 2014; Jane, 2015; Lapidot-Lefler & Barak, 2012;
Sproull & Kiesler, 1991; Suler, 2004). For example, a number of studies have shown that when those of opposing political beliefs interact online, communication can result in aggression and escalation of their beliefs (Barberá, Jost, Nagler, Tucker, & Bonneau, 2015; Hmielowski et al.,
2014; Hutchens et al., 2015). These results instead highlight that lean-media, such as CMC or text-based communication, can facilitate intimacy related behaviour like self-disclosure and even reduce aggression in political discussion.
Medium Effects on Humanness Despite identifying medium effects on moral behaviour in both Study 1 and Study 2, there was limited support for the hypothesized explanatory role of humanness. Specifically, the absence of non-verbal cues in text was hypothesised to lead to systematic changes in humanness perception that would, in turn, affect moral behaviour. However, in both studies humanness did not, or minimally, accounted for the relationship between medium and behaviour, suggesting that there is limited support for humanness as an account for online moral phenomena.
Further, there was limited support for the CFO approach to understanding medium effects on behaviour. Self-disclosure results were partially explained by aspects of the CFO and hyperpersonalism perspective: the HU manipulation had a larger effect on self-disclosure in the text condition than in the video condition, consistent with the claim that available cues are exaggerated during impression formation when people are presented in text. However, there was no direct effect of medium on any measured aspect of person perception; thus, the
Chapter 5: Discussion of Stream 1 112 exact mechanism by which medium affected self-disclosure is unclear. The results for Study
1 are also inconsistent with current theories of CMC behaviour. One possible explanation that may provide some explanation for both behaviours are expectations about the interaction.
The Role of Expectations for Flaming
A theory from the negotiation’s literature, the communication orientation model, may provide some explanation for findings of Study 1 (Swaab, Galinsky, Medvec, & Diermeier,
2012b). The communication orientation model resolved a similarly paradoxical finding within the online-negotiation literature. In some circumstances, the presence of verbal or visual channels had positive consequences for negotiations (compared to text-only communication), while other times those same cues hindered negotiations and group decision making (Swaab et al., 2012). The model argues that richer communication mediums are only associated with positive negotiations if the interactants have a cooperative or neutral orientation. When people have a neutral orientation the presence of more social-contextual cues will reduce their uncertainty of their partner’s intentions and so support the establishment of rapport. When people have a cooperative orientation, positive negotiation will result irrespective of the media richness. Swaab et al., (2012) argue that those with a cooperative orientation are motivated to share information about their preferences, priorities and interpret others’ actions as attempts to cooperate. These are ambiguity-reducing behaviours that will clearly establish their cooperative intentions, even in a cue-poor medium.
Importantly, when people have a non-cooperative orientation, then more social-contextual cues will actually impede negotiation. According to Swaab et al. (2012), when confronted with extreme opposition, the richness of the medium will intensify negative emotions and escalate an actor’s non-cooperative predispositions. As a result, people are more likely to interpret one another’s actions as hostile attacks and so utilise competitive tactics to defend
Chapter 5: Discussion of Stream 1 113 their own interests. Applied to a flaming scenario: if the participant enters an interaction with a non-cooperative orientation, they are more likely to perceive a target as having hostile intentions. This negative perception of the target will be exaggerated when interacting in a richer medium, increasing the likelihood that they respond with flaming. In the context of my study 1 methods, I strategically provided participants with the extreme opposition on a heated political topic. I aimed for the participant to perceive the out-group member as threatening to their own values to facilitate flaming behaviour. As such, it is reasonable to assume that the threatening out-group member likely generated a non-cooperative, conflict orientated mindset for the participants. When the out-group member voiced their opinion, rather than writing, the extra social-contextual cues may have intensified the oppositional nature of the opinion.
Participants in the voice condition would then be more likely to perceive the target as threatening and so respond defensively. Hence, the increased tendency to flame in our verbal condition may be explained by an interaction between the participant’s non-cooperative orientation and the paralinguistic cues available in the voice condition, but absent in the text condition.
This model provides an explanation for the inconsistency for Study 1’s result and the two other across-media tests of flaming. Castellá et al. (2000) found that there was a significant decrease in the prevalence of flaming with increased richness of the communication channel (CMC, Video, FtF). Participants were required to complete, in groups, a ‘moon landing’ ordering task. This task requires participants to imagine that they have crash-landed on the moon and (as a group) rank the importance of a list of survival items. It is unclear what the participant’s orientation may have been as the authors did not include the participant’s provided instructions. Yet, given that the task was not competitive participants were likely to be cooperative or neutral. Consequently, the additional social-
Chapter 5: Discussion of Stream 1 114 contextual cues provided by the video and FtF condition exerted a positive influence on their negotiation outcomes and decreased the likelihood of flaming.
The interpretation is less clear in the second across-medium study and accounting for these results requires a divergence from the communication orientation model (Swaab et al.,
2012). Unlike Castellá et al. (2000), Lapidot-Lefler and Barak (2012) included a competitive communication topic: participants were randomised to dyads and instructed to debate whom of the two should receive a lifesaving drug. The communication was either: (1) anonymous or identified, (2) visible by webcam or not (side-on view only), (3) or involving eye contact or not. It was found that the absence of eye contact was the largest contributor to the presence of flaming. According to the communication orientation model, we may expect that the presence of eye contact would actually increase flaming in this competitive scenario. Indeed, Swaab et al. (2012) explicitly refer to findings that eye contact exacerbates negative feelings and poor negotiation outcomes (Swaab & Swaab, 2009).
One possibility is that there are two competing factors that contribute to aggression in the study. Specifically, while the presence of additional social and contextual cues (i.e., eye contact) may have facilitated aggression, the identifiability of the eye contact condition instead suppressed anti-normative behaviour (i.e., flaming). Flaming is a non-normative behaviour that consists of explicit aggressive communication; this implies that flaming will be less likely to occur when there is the potential for negative social evaluation. The anonymity in this study was limited to whether the participants had personalised usernames or not. Thus, the participants were not strictly anonymous as they had webcams broadcasting them to their partner in both the eye contact and visibility condition. The presence of a webcam in the visibility and eye contact conditions likely increased participant’s self- consciousness and, in turn, increased inhibition of flaming (Joinson, 2001).
Chapter 5: Discussion of Stream 1 115
Therefore, I suggest that anonymity is a necessary, but not sufficient condition for flaming to occur. When the anonymity condition is met, the interaction between media richness and communication orientation will predict whether flaming occurs or not. This precondition was met in the current study as actors were explicitly told that they were unidentifiable, thus this accounts for why Study 1 found more flaming in the richer medium, while Lapidot-Lefler and Barak (2012) did not.
Although this model has only been tested on negotiation outcomes, there is some tentative evidence for the relationship between non-cooperative orientations and flaming. The majority of documented instances of flaming have occurred in contexts where there is political discussion (Kayany, 1998; Laineste, 2013; Molaei, 2014; Santana, 2014). It may be that political topics provide a salient means of identifying ideological differences; when these differences are perceived as threatening their own belief system they may respond with a hostile orientation, increasing the likelihood of flaming. Consistent with this, Hutchens et al.
(2015) found that direct challenges to a participant’s political beliefs were mostly strongly associated with intention to flame.
The application of the communication orientation model to flaming implies that flaming is not a function of a cue-poor medium per say but is constrained to a specific circumstance within online interactions. This may explain why there has been such variance in the prevalence of flaming (e.g. Kayany, 1998; Moor et al., 2010; Reid & Reid, 2005;
Swaab et al., 2012) with some researchers even claiming that flaming is so infrequent as to not deserve research attention (Abrams, 2003; Lea et al., 1992). Future studies should test the claim that flaming is an interaction between participant anonymity, orientation and medium richness.
Chapter 5: Discussion of Stream 1 116
The Role of Expectations for Self-Disclosure
Study 2 also found that expectations, but as an individual difference, were important for self-disclosure. Of all measured variables, only expectations, moral agency, and trust significantly predicted self-disclosure, despite testing a number of CMC theories and potential moderator/mediators of medium effects on self-disclosure. The expectation result is particularly notable given the presumed role of expectations in Study 1 and its novel nature.
Results show that the extent to which participants expect others to behave aggressively online significantly predicted both reduced intimacy and less breadth of self-disclosure. These results point to a largely overlooked area of research for the cyber-psychology literature: the effect of individual differences in online behavioural phenomena, including online self- disclosure. While there has been a plethora of research that examines possible mechanisms that underlie these online phenomena, much of this research has favoured studying situational factors such as the medium or social context (e.g., each of the theories addressed in Study 2).
Relatively few papers have examined individual differences that contribute to online behaviour (Alonzo & Aiken, 2004; Hammick & Lee, 2014; Schouten et al., 2007). The results of Study 2 demonstrate that expectations of other’s behaviour in the medium should be considered when attempting to understand online self-disclosure. The theorising from
Study 1 and my application of the communication orientation model, suggest that there may be an interaction between the medium and expectations such that different expectations have different effects for cue-poor versus cue-rich media (Swaab et al., 2012). Study 2 provides evidence that expectancies have downstream consequences for how people behave when approaching new online situations and is fertile ground for future research. Chapter 5: Systematic Review 117
Stream 2: The Effects of Medium on Moral Judgement
Research Agenda for Stream 2
The first stream of this thesis has revealed that there is limited evidence for the CFO account of medium effects on behaviour/perception. Furthermore, predictions derived from other prominent theories in the CMC literature were also largely unsupported. Across two high-powered studies, predictions derived from CFO, SIDE, hyperpersonalism, and RC had limited (and in some cases, no) support. Furthermore, the main theory tested, CFO, failed to account for moral behaviours of either a positive or negative valence. These results suggest that the CFO model with humanness extension is a poor account of medium effects on moral behaviour. As a result, the exact nature of how medium relates to morality is unclear.
The second half of the thesis will thus take a more basic approach, moving away from
CMC-related theorising and research. CMC was a useful starting point to examine how medium might affect morally relevant perception and behaviour because there is a mature literature to inform my examination of medium effects for the moral domain. However, given the lack of support for CMC theories in Study 1 and 2, in Stream 2 of this thesis I instead utilise moral psychology theories to frame my exploration of medium effects. Further, the second stream also moves away from CMC as a context for studying medium effects. CMC is one specific example of a cue-poor communication medium, Stream 2 considers the more generalizable context of stimulus presentation medium.
To date, moral psychology has relied disproportionately on text-based stimuli in the development of theories and in the testing of empirical research questions. Reviews by
Boccia et al. (2017) and Chapman and Anderson (2013) of subsets of the moral psychology literature suggest that up to 90% of studies on moral judgement have exclusively relied on text-stimuli. Further, many of the most influential theories in moral psychology have been developed with a near-exclusive reliance on text-stimuli; for example, the dual process model Chapter 5: Systematic Review 118 of moral judgement was developed using text depictions of trolley problems (Greene,
Sommerville, Nystrom, Darley, & Cohen, 2001; Lotto, Manfrinati, & Sarlo, 2014) and moral foundations theory (MFT) was refined using various text-based self-report instruments
(Clifford, Iyengar, Cabeza, & Sinnott-Armstrong, 2015; Graham et al., 2011; Lotto et al.,
2014). However, text stimuli lack many of the social and contextual cues available in everyday interactions that directly influence moral processes (Cannon, Schnall, & White,
2011; Proeve & Howells, 2006). Therefore, the over-reliance on text stimuli in moral psychology may have resulted in the mischaracterisation of moral psychological processes.
Consistent with this possibility, the results of Study 1 and Study 2 provide evidence that the medium in which a stimulus is presented influences moral behaviour (e.g., self-disclosure and aggression) and perception (e.g., humanness). However, the exact nature of medium effects on moral behaviour remain unclear (e.g., directionality of effects, mechanism by which these effects occur). Thus, the type and magnitude of mischaracterisation of moral psychological processes is unknown.
In order to better understand the nature of medium effects on moral psychological processes, Stream 2 of this thesis will explore systematically explore the effect of presentation medium on moral judgement. First, I conduct a systematic literature review to quantify the extent to which moral psychology has relied on text-restricted stimuli and explore differences between text and other media if sufficient instances are uncovered by this literature review (Study 3). Second, I provide the means for researchers to address the possibility that moral psychological processes are mischaracterised by the overreliance on text stimuli by developing a moral film set, the MAAFS (Study 4). Finally, I test this stimulus set against a matching set of text-transcriptions to infer conclusions about the consequence for the over-reliance on text stimuli on the measurement of the moral judgement (Study 5). Chapter 5: Systematic Review 119
Chapter 5: A Systematic Literature Review of Presentation Medium in Moral Psychology Everyday interactions are rich in social and contextual cues that enable and constrain morally relevant behaviour (Burgoon, Guerrero, & Manusov, 2011). For example, people generate non-verbal cues when communicating (e.g., facial expression, speaking pace, voice tone, eye gaze) and aspects of the environment provide contextual cues (e.g., cultural context)
(Sproull & Kiesler, 1986). Presentation media differ in their capacity to convey these social and contextual cues (Culnan & Markus, 1987). For example, video stimuli are rich in both verbal cues (e.g. language) and the non-verbal cues (e.g. facial expressions, voice tone). In comparison, text stimuli often fail to convey many of the cues that are present in face-to-face interaction (Culnan & Markus, 1987; Short et al., 1976).
Growing evidence indicates that the medium used to present a stimulus or complete an experimental task can affect outcomes. For example, a message is more persuasive when presented with a rich medium (video), compared to when that same message is presented using text or audio (Chaiken & Eagly, 1976). Likewise, when working on the same task, teams that communicate using richer media (e.g., voice) report more teamwork behaviour
(e.g., communication, giving feedback) than those using text communication (Fletcher &
Major, 2006). Meta-analytic comparisons of different negotiations have found that audio and visual cues increase the likelihood of positive outcomes when actors have positive expectations for the negotiation, but worsen outcomes when actors have negative expectations (Swaab et al., 2012). The perception of a target also changes depending on what presentation medium is used. When content is held constant, participants rely on stereotypes more when communicating with someone over email (text) compared to voice (Epley &
Kruger, 2005). A similar effect of medium exists when generalising across natural conversations: those who converse using text compared to face-to-face rely more heavily on Chapter 5: Systematic Review 120 their expectations (Walther & Tong, 2014) and exaggerate the importance of available information when perceiving their partner (Hancock & Dunham, 2001).
Importantly, many of the verbal and non-verbal cues, which are sometimes absent in text, are directly relevant to moral judgement. For example, facial expression and voice tone communicate emotions (Keltner, 1996; Simon-Thomas, Keltner, Sauter, Sinicropi-Yao, &
Abramson, 2009) which in turn can elicit empathy (Niedenthal, 2007; Saarela et al., 2007), signalling that a moral transgression has occurred (Pizarro, 2000). Facial and vocal cues can also signal a perpetrator’s remorse or guilt (Proeve & Howells, 2006). Other cues, such as proxemics (physical distance) and kinesics (body language), can indicate the nature of the relationship between actors (Burgoon & Le Poire, 1999) which, in turn, may define what counts as morally acceptable (Simpson & Laham, 2015). Thus, the presence or absence of such cues across media suggests that different media may elicit different moral judgements
(in degree and/or kind).
Evidence for the Effect of Medium in Moral Judgement Consistent with this theorising, there is already some evidence to suggest that presentation medium does affect moral behaviour and perception. For example, participants attribute fewer humanness qualities to a target that is presented using text than voice
(Schroeder & Epley, 2015, 2016). Presentation medium also influences cooperation in economic games: when the same game is played using social and contextually rich media
(voice and video) compared to restricted media (text), participants are more cooperative and rate their partners as more trustworthy, intelligent, and likable (Brosig, Weimann, &
Ockenfels, 2003; Jensen, Farnham, Drucker, & Kollock, 2000). Similarly, emotion, a construct frequently linked to morality (Greene et al., 2001; Rozin, Lowery, Imada, & Haidt,
1999; Schnall, Haidt, Clore, & Jordan, 2008), also varies by presentation media. Multi-modal stimuli (e.g. subtitled film, which includes visual, aural and verbal modalities) tend to elicit Chapter 5: Systematic Review 121 more intense emotional responses than text stimuli – particularly for anger and sadness
(Ferrer, Grenen, & Taber, 2015; Gerrards‐Hesse, Spies, & Hesse, 1994; Westerman, Spence,
& Van Der Heide, 2014). One possibility is that researchers may have underestimated the effect of emotion on moral psychological processes by over-relying on text stimuli - a medium that features a single modality and lacks non-verbal cues.
Presentation medium may not only affect moral judgement quantitatively (e.g., how wrong a transgression is), but also qualitatively (e.g., why a transgression is wrong). Text- stimuli are often more abstract than image or video stimuli because written language requires the reader to draw on his or her mental representation of the stimuli to fill in the blanks.
Video (or images) instead fill in the blanks for an observer by depicting more concrete stimulus features (Amit & Greene, 2012). Abstractness changes a range of psychological variables related to moral judgement, for example, abstract thinking (compared to concrete thinking) is associated with a greater attention to ends versus means (Amit & Greene, 2012), greater value-behaviour consistency (Eyal, Sagristano, Trope, Liberman, & Chaiken, 2009), emphasis on different moral values (Luguri, Napier, & Dovidio, 2012; Napier & Luguri,
2013; Vess, Rogers, Routledge, & Hicks, 2016), and less harsh moral judgements (Gong &
Medin, 2012; Žeželj & Jokić, 2014). A study of virtual-reality trolley dilemmas provides some direct evidence for the effect of presentation medium on moral reasoning. When cue- rich virtual reality sacrificial dilemmas have been contrasted with the same dilemma presented as text-restricted vignettes, participants make significantly different responses
(Patil, Cogoni, Zangrando, Chittaro, & Silani, 2014). Specifically, those presented with virtual reality dilemmas made more utilitarian judgements compared to those presented with text (Patil et al., 2014). The authors argue that in the text condition, the absence of contextual information forces participants to rely more on mental simulation of the moral event, thereby changing their response when compared to a context-rich virtual reality dilemma (Patil et al., Chapter 5: Systematic Review 122
2014). Therefore, violations presented in text may be judged qualitatively differently to the same violations presented via video.
The Overuse of Text May Have Biased Moral Psychology Research As a result, an overreliance on text stimuli may have biased our picture of moral psychology. A central tenant of moral psychology is the importance of harm: the moral proscription to not harm others has repeatedly been demonstrated as underpinning many moral judgements (Cushman, Young, & Hauser, 2006; Gray, Young, & Waytz, 2012;
Greene, Nystrom, Engell, Darley, & Cohen, 2004). Some researchers even theorise that judgements of harm underlie all moral judgements (Gray et al., 2012). However, studies that have used text stimuli have demonstrated the primacy of harm (e.g., Cushman et al., 2006;
Gray et al., 2012; Greene et al., 2004) use text stimuli only. Given that abstractness can make moral principles more salient (Agerström & Björklund, 2009; Eyal & Liberman, 2012; Eyal,
Liberman, & Trope, 2008; Eyal et al., 2009) and text stimuli are more abstract than images/videos (Amit & Greene, 2012), one possibility may be that considerations of harm are more relevant when responding to text stimuli (than less abstract stimuli). If this is the case, reliance on text-restricted stimuli may have resulted in the over-stated importance of harm in our view of moral psychology. Consistent with this, when an economic game (a contextually rich stimulus) was compared to an abstract, text stimuli, participants were more likely to let others be hurt than when presented with the abstract, text stimulus (FeldmanHall et al., 2012).
The authors concluded that when the participant believes their response will have real, concrete consequences (such as, in the economic game) they are more likely to think harming others is permissible than when they think their response is a hypothetical scenario. Thus, the importance of harm in moral judgements may be diminished in more realistic scenarios that are rich with social and contextual cues. This result implies that the cue-impoverished nature of text-restricted stimuli that are currently used in moral psychology may have biased the Chapter 5: Systematic Review 123 moral judgement of participants in such a way that may have led theorists to over-state the importance of harm.
A similar mischaracterisation could have affected moral foundations theory (MFT).
MFT is derived from an evolutionary account of morality yet has been tested using text- heavy methods that fail to represent the kinds of non-verbal stimuli available during the evolutionary development of morality (e.g., Clifford et al., 2015; Graham et al., 2012;
Graham, Haidt, & Nosek, 2009; Rottman, Kelemen, & Young, 2014; Yilmaz, Harma,
Bahçekapili, & Cesur, 2016). According to MFT, evolutionary pressures have shaped human morality: each foundation of morality has emerged because of the adaptive value that it has provided for living in a social environment (Haidt, 2007). The building blocks for morality are thought to have appeared very early on in human development, before the emergence of language and deliberative reasoning capacities (Haidt, 2007). However, only text stimuli have been used to represent and assess the moral foundations. Text stimuli fail to represent the non-verbal social and contextual information that (1) contributed to the development of human morality, and (2) (in part) reflects every day, contemporary morality (Hofmann,
Wisneski, Brandt, & Skitka, 2014). As a result, by developing and testing claims of the MFT with only text stimuli, the foundations may not generalise to more ecologically valid, social and contextually rich (and non-verbal) stimuli. One possibility is that when non-text stimuli are used to describe morality; we may find a different structure of moral foundations.
Text-heavy moral methods may also mischaracterise moral psychological processes by conflating the measurement of psychological processes and language capacities. By definition, text-restricted stimuli require that the participant have a certain level of language ability to both comprehend the stimuli and respond appropriately. However, individual differences in verbal ability, spatial ability, and cognitive style interact with stimuli medium to predict differences in performance on learning tasks and language comprehension Chapter 5: Systematic Review 124
(reviewed in Chun & Plass, 1997, see also Mayer & Massa, 2003). For example, low verbal ability negatively impacts text comprehension but also predicts the efficacy of visual aids.
Those with visualising cognitive styles also perform worse on learning and comprehension tasks when presented with text-restricted stimuli than those with verbalising cognitive styles
(Chun & Plass, 1996a, 1996b). However, there is no difference in performance when visualizers are provided with visual cues or multi-media stimuli (Chun & Plass, 1996a,
1996b). Thus, text-heavy methods in moral psychology may disproportionately impact the language comprehension and responses of those with low verbal, spatial abilities or visualising cognitive styles. Accordingly, variation in moral judgement may be driven in part by variation in language abilities for text stimuli to a far greater extent than for multi-modal stimuli. Non-text stimuli, such as videos, circumvent this concern by providing morally relevant information in both verbal and non-verbal forms, thereby more accurately measuring moral judgement than text stimuli, irrespective of individual variation in cognitive style.
Taken together, there is evidence to suggest that medium may impact the measurement of moral judgment. In turn, this overreliance of text stimuli may have biased our picture of moral judgment and moral psychology. The extent to which moral psychology has been biased by the use of text and the exact nature of any such effects is unclear from my narrative review of the literature. A systematic literature review is one way in which we can quantify the degree to which the literature has relied on text stimuli. Further, using meta- analytic techniques, a systematic review may examine any differences in those studies that have used text stimuli versus studies that have used non-text stimuli.
Overview of the Systematic Review I conducted a systematic review of any empirical study that measured moral judgment published between 2001 to 2017 (Npublications = 314; Nstudies = 718 were included). The moral stimuli used in each study was coded as either text or non-text, along with year of Chapter 5: Systematic Review 125 publication, number of stimuli, number of participants, and use of an ad-hoc or previously validated measure7. I calculate percentage of studies that use text stimuli (only) and compare the text and non-text studies with similar outcome measures.
Methods Due to the size of this systematic review, four research assistants supported the search and coding of the literature. Research assistants were each allocated a search term and a database, screened search results for eligibility (detailed below), and coded the presentation medium of moral stimuli for each included study. I completed a final check of each manuscript against the inclusion criteria and double-coded any study that contained a non- text moral stimulus.
Search Strategies Search terms and key definitions and are summarised in Table 24 and
Table 24. Theoretical Definitions
, respectively. These definitions and the search terms were provided to the research assistants that supported the review, as part of their instructions. Specifically, moral judgement was defined as any implicit or explicit assessment of the moral value of a person(s) or situation(s).
7 While a number of attributes were coded, for the purpose of this thesis, I focus on presentation medium. Chapter 5: Systematic Review 126
Table 24. Theoretical Definitions Variable Definition Examples
Moral judgement Any implicit or explicit Key phrases associated with assessment of the moral moral value in prominent value of a person(s) or theories include intuition, reason, value, good, bad, situation(s). right, wrong, blame, praise punish, deontological, utilitarian, foundation, patient, agent.
Moral stimulus The moral stimulus (or stimuli) is any object or event that is presented to the participant (as part of the study design) to elicit a morally relevant psychological or behavioural response. Presentation medium Presentation medium was defined as either text or non-text. Text stimulus Text was coded as Typical examples of text anything that is presented stimuli include using only written questionnaires, vignettes, language; the written and transcriptions of court language might be on a documents. printed on a piece of paper, presented on a computer screen, or a handwritten note. Non-text stimulus Any stimulus that is not For example, economic presented exclusively with games may include text, which means that a instructions that were non-text stimulus may presented in text, but the have both written text and stimulus also involves a non-text elements. behavioural element.
Chapter 5: Systematic Review 127
Table 25. The Derivation of Search Terms from Prominent Theories and Studies
Theory or Domain of Moral Description Search Term Psychology Research Social intuitionist model Moral judgement can be Intuition (Haidt, 2007) automatic, affect-laden intuitions. Moral foundations theory MFT describes six foundations Foundation (Graham et al., 2012; Graham of moral concern and contends et al., 2009) that people vary in the extent to which they experience moral intuitions for each of the six foundations. Moral reasoning Other theories of moral Rational; reasoned (e.g., Kohlberg, 1966; Monin, judgement have focused on the Pizarro, & Beer, 2007; D. A. contribution that deliberative Pizarro & Bloom, 2003) reasoning plays in moral judgement Dual process model The dual process model Deontological; utilitarian (Greene et al., 2001). attempts to account for both the role of emotion and reasoning in response to moral dilemmas. In particular, this theory defines moral judgement in terms of deontological (rule- based judgements) or utilitarian (outcome-based judgements). Moral dyad theory Moral judgement occurs in Patient; agent (K. Gray et al., 2012) response to the perception of harm only. According to moral dyad theory, moral judgement is the response to a moral agent causing harm to a moral patient.
Attributions of blame Moral judgement may not only Praise; blame; punishment (Bastian et al., 2011; D. A. Moral judgement may involve Pizarro, Uhlmann, & Bloom, not only assessments of the act, 2003) but also the actors. People are motivated to acquire information about the character of others, and morally deviant behaviour is a source of information about an actor’s moral character.
Chapter 5: Systematic Review 128
Two research assistants were assigned to each of PsychInfo and Web of Science and each of the search terms ‘moral*’ and ‘ethic*’8. Research assistants were then assigned five additional search terms, a subset of the search terms drawn from major moral theories:
intuit*, reason*, evaluat*, valu*, good, bad, right, wrong, blam*, prais*, punish*, deont*, utili*, foundation, patient*, agent*
The same keywords were assigned to multiple research assistants to ensure all relevant papers were captured in our search.
Articles were sourced through the major databases for psychology, PsychInfo and
Web of Science, and were accessed from October 2016 to May 2017.
Selection Process Research assistants independently searched their assigned database with their provided terms. The search results in 3,355 papers, before any screening for eligibility. The research assistants initially screened the resulting publications for eligibility according to the study title and abstracts (N=2,285 excluded, N=1,070 remained). After the first screen, the research assistants then assessed the eligibility of each the studies contained in the publications by reading the remaining publications in full (N=592 excluded, N=478 remained in the review). I then reviewed each publication for a final screen of eligibility (using the same eligibility criteria), 51 papers were excluded for failing eligibility criteria, and 113 papers were excluded for duplications.
There were 314 publications included in the systematic review, which contained 718 eligible studies. An overview of the search process and inclusion and exclusion of papers at each step is summarised Figure 12.
8 * refers to a wildcard search Chapter 5: Systematic Review 129
Figure 12. An overview of publication selection and exclusion for the systematic literature review.
Eligibility Criteria I limited the scope of this study, for pragmatic reasons, to adult populations with no clinical abnormalities and studies published in English. Accordingly, I have excluded samples with adolescents or children, incarcerated populations, or abnormal psychology. I surveyed articles from 2001 onwards as this year marks the publications of major papers by
Haidt (2001) and the onset of a ‘renaissance’ of research in moral psychology. Chapter 5: Systematic Review 130
A paper was included in this systematic review if: (1) the stimulus was a morally relevant text, image, dynamic image9, video, sound, virtual reality experience, or behaviour
(e.g., economic game, confederate behaviour, or memory recall), (2) the study was empirical
(i.e., not a theoretical paper); presents participants with moral stimuli and require participants to make a judgement of its moral qualities, (3) the study was disseminated as a conference proceeding, published article in a peer-reviewed journal or dissertation, (4) the study measured moral judgement defined as any implicit or explicit assessment of the moral value of a person(s) or situation(s)10.
Next, to further explore the non-text stimuli presentation media, non-text studies were coded into more refined categories11: text, image, dynamic image, video, sound, virtual reality experience, FtF (e.g., economic game, confederate behaviour), or memory. Where applicable, one study may be coded as a combination of these categories. The coding scheme is summarised in Table 26.
Data Collection Process Any paper included after all screens were then coded (by the same research assistant that retrieved it). Research assistants collected information on the (1) citation information
(year of publication, authors, title); (2) number of participants, (3) number of stimuli, (4) the presentation medium of the stimuli, (5) the morally relevant dependent variable, (6) any previously validated morally relevant questionnaires or measures; however, for this thesis, I
9 Dynamic image is a time-series of still but consecutive images, typically used in neuroimaging studies in lieu of video. 10 This broad inclusion criteria includes both explicit judgements of morality and implicit judgement, meaning that research using a moral intuitionist operationalisation of judgement (i.e., judgements are quick, affect-laden intuitions) are not excluded. For example, Whitton, Henry, Rendell, and Grisham (2014) present participants with moral images and measure facial expressions to clarify the relationship between implicit moral judgement and disgust. Despite not explicitly measuring moral judgement, these studies still extend moral psychological research and thus were relevant to this review.
11 Research assistants were not used for the coding of non-text presentation media Chapter 5: Systematic Review 131 will only explore the results for the coding of presentation medium. Any time the research assistant was uncertain about their coding, they made a note of this paper in the coding table.
I double-coded any study that was marked by the research assistant as well as any study coded as non-text by the research assistant. The coding scheme is summarised in Table 26.
Table 26. Effect Codes, Definitions, and Coding Options. Variable Definition Coding Options
Authors Authors of the publication Open-ended (string) Year Year of publication Open-ended (string) Title Title of publication Open-ended (string) Morally relevant dependent Any assessment of moral Open-ended (string) variable value. An assessment of moral value may be (1) automatic and intuitive, (2) reasoned and rational, (3) involve a judgement of a person; (4) or an act. Previously validated Was the morally relevant Ad hoc, previously- measures dependent measure created validated ad hoc (for this study) or previously validated? Presentation medium The presentation medium Text, non-text used to convey the moral stimulus. Non-text presentation The non-text presentation Text, image, dynamic medium medium used to convey the image, video, sound, virtual moral stimulus reality experience, FtF, memory Number of participants Final number of participants Open-ended (numeric) used in the study analyses, after any exclusions Number of moral stimuli How many moral stimuli Open-ended (numeric) were used in the study to assess the morally relevant dependent measure
Chapter 5: Systematic Review 132
Interrater Reliability In the current study, raters each coded a subset of the included papers. Thus, to calculate the inter-rater reliability, all research assistants were also assigned the same 10% of studies (N = 76), and inter-rater reliability coefficients were generated from this sample.
Two-way random effects models with absolute consistency assessed the inter-rater reliability.
This type of statistical model accounts for variation that stems from both the raters and the studies by modelling these effects as ‘random’ (Koo & Li, 2016). Finally, I used absolute agreement, which assesses if raters assign the same score to each study, rather than the degree to which raters differ in scores (Koo & Li, 2016). Reliability was calculated for presentation medium and, for comparative purposes, the number of participants, and stimuli (presented in
Table 27).
There was a high degree of interrater reliability for coding of presentation medium
(.80). Although this is lower than expected, these are results still within an acceptable range, and thus the data was used for the main analyses. This surprisingly low result reveals that, despite the presumed objectivity of judging text versus non-text, judgements of medium are less clear than anticipated. Number of stimuli has a similar reliability (.82), although in an acceptable range, also lower than expected. In contrast, interrater reliability for number of participants was very close to 1.0, perfect reliability. This result suggests that the raters were competent at coding and that coding of the moral stimuli involved some ambiguity and subjective judgement. One possibility is that raters had difficulty identifying the moral stimuli, as some studies include multiple stimuli (moral and non-moral; see Jackson, Gaetner,
& Batson, 2016). As reliabilities are in an acceptable range, the main analyses will use the coding of these raters.
Chapter 5: Systematic Review 133
Table 27. Inter-rater reliabilities for the coding of the included papers
Presentation medium .80 [.70, .90] F (75, 75) = 8.83, p < .001
Number of participants .999 [.998, .999] F (46, 92) = 28293, p < .001
Number of stimuli .82 [.72, .89] F (65,42) = 18.4 p < .001
Results and Discussion
Reliance on Text Stimuli As predicted, the clear majority of studies exclusively used text stimuli: 92.2% (N =
662) of studies were coded as text stimuli. As noted above, the definition of text was restrictive such that borderline cases were classed as non-text. This definition slightly biases the coding to overestimate the proportion of non-text stimuli, rather than overestimating the proportion of text stimuli. Despite this potential bias in the coding scheme, the number of text stimuli still disproportionately outweighs the number of non-text. There were 56 studies that use non-text stimuli that I assessed in more detail (a table of the non-text studies is included in supplementary materials (Appendix A).
Of the non-text studies, moral images were the most common non-text presentation medium (42.4%, N = 25), followed by video (16.9%, N=10) and FtF (13.6%, N = 8). The proportion of presentation media are summarised in Figure 13. Chapter 5: Systematic Review 134
Figure 13. Proportion of presentation media in the reviewed studies. Note. Left is text and non-text proportions and right is non-text portions of presentation media.
Presentation Medium Over Time Furthermore, the overreliance on text methods is not decreasing over time, despite developments in technology (e.g., access to computers, virtual reality, video and image sharing websites) and methodology (e.g., open access sharing of materials) (Figure 14).
Although the use of non-text methods has increased over time, the increase is proportionate to the number of studies in the field in total. According to the results of this review, the number of publications studying moral judgement has rapidly increased from two publications in 2001 to 123 publications in 2016 (2017 was incomplete). Although the number of non-text has marginally increased during that time, the percentage of studies that utilise non-text methods remains low even in 2016 (2.5%). If the number of studies continues to rise but researchers continue to disproportionately rely on text moral stimuli, the field may Chapter 5: Systematic Review 135 be increasingly at risk of mischaracterisation accordingly to the limitations text-stimuli.
Figure 14. Frequency of Text, Non-Text Stimuli, and Total Number of Studies From 2001 to
2017.
Implications of Over-Reliance on Text Stimuli Unfortunately, meta-analytic techniques are not able to statistically estimate whether presentation medium affects moral judgement because of the heterogeneity in dependent variables combined with the minimal number of non-text papers. For example, Bahnemann,
Dziobek, Prehn, Wolf, and Heekeren (2009) asked participants to make moral judgements but reported only measurements of neurological activation, while Buon, Jacob, Loissel, and
Dupoux (2013) used a binary measure of moral character where actors were categorised as good or bad, and Brambilla and Riva (2017) used a 7-point Likert scale of moral character evaluations. As a result, I will instead descriptively review the possible effects of presentation medium on moral judgement by comparing studies that the studies that directly compare responses as a function of media (N = 5). Chapter 5: Systematic Review 136
Medium Effects on Moral Emotion First, the results of Lee and Gino (2015) show that inducing negative affect was more successful when a video stimulus was used, compared to images. Specifically, there was significantly more negative affect when presented with an aversive moral video (M = 4.61,
SD = 2.20) than aversive moral images (M = 2.70, SD = 1.35; t(278) = 8.98, p > .001). This result could imply that there is an effect of presentation medium on moral emotions, such that richer media (e.g., videos) arouse more intense moral emotions than leaner media (i.e., text or images). One potential interpretation of this result is that the additional non-verbal cues in the video, relative to the image condition, increased arousal. This comparison is limited, however, as the authors did not equate the moral content of the images and videos (this was not the focus of the study). Additional research that controls the moral content of the stimuli across different presentation media is required to assess whether presentation media affects moral emotions.
Medium Effects on Moral Judgement and Behaviour Overall, there was mixed evidence for an effect of medium on moral judgement. First,
Iliev, Sachdeva, and Medin (2012) found that medium does not affect attributions of causal responsibility made from certain types of cues. Specifically, the authors found that visual cues of motion and physical contact (i.e., non-verbal cues absent in text) are relevant to judgements of causal responsibility, but also extend this effect to show that text descriptions of motion and physical contact have an equivalent effect on judgements of causal responsibility. Thereby demonstrating that a verbal description of contact or motion has an equivalent effect on moral judgement as a non-verbal representation of contact or motion.
However, it’s unclear if this effect is limited to attributions made from cues about motion or physical contact, or if this effect extends to attributions made from other cues (e.g., a moral agent expressing guilt or mitigating contextual factors). Chapter 5: Systematic Review 137
There is also evidence that presentation medium affects mentalisation, which can, in turn, affect moral judgement and behaviour. Caruso and Gino (2011) show that presentation media that portray less information, facilitate mentalisation of the moral act, and this, in turn, results in more polarised judgements. The authors found that when participants listened to moral vignettes (with their eyes closed), they made more polarised judgements (moral behaviours were more moral, immoral behaviours were less moral) and were more likely to behave ethically, than when they read the same vignette. The authors also show that this effect of presentation medium was mediated by the increase in mental simulation when listing to the moral vignette. Given that text stimuli tend to be abstract and may facilitate mentalisation by requiring the reader to draw on mental representations to ‘fill in the blanks’.
This study, therefore, suggests that the reliance on text stimuli could have led to more polarised moral judgements in past research and overestimation of effect on moral judgement. This result implies that presenting moral content in a medium that facilitates mental stimulation (i.e., text) will harshen moral judgement.
The use of text stimuli may not only change the harshness of judgement but may also the directionality of moral judgement. FeldmanHall et al. (2012) show that providing more concrete details in a stimulus (and reducing the opportunity for mental simulation) leads to less moral behaviour. Specifically, the authors compared responses to a hypothetical dilemma and participant decisions in an equivalent, real scenarios that included increasing levels of contextual information. Participants were increasingly likely to harm another for their gain when more concrete details of a stimulus were included, but when presented with the same dilemma as a hypothetical text dilemma, participants were (on average) less likely to harm another for personal gain. The results of this study suggest that moral judgements of hypothetical text dilemmas (particularly those that lack contextual information) do not represent the moral behaviours in real situations. Importantly, the overreliance on text stimuli Chapter 5: Systematic Review 138 in the literature may have mispresented the nature of moral behaviours in ecologically rich scenarios. Given the close relationship between moral behaviours and judgements, this might also imply that the use of text stimuli may have led to the misspecification of moral judgements also.
Medium Effects on Moral Character There is some preliminary evidence that medium is relevant to moral character.
Brambilla and Riva (2017) show that an actor was perceived as more immoral when presented via a non-text than a text stimulus. The authors conducted two studies: the first used text stimuli and the second was a conceptual replication/extension with non-text stimuli
(image and text). I compared the effect of the two types of stimuli on moral character judgements by testing if the means in their immoral character condition significantly differ by presentation medium. There was a significant difference such that the actor was judged as more immoral when presented as an image/text (M = 2.02, SD = 1.18), than when presented as text only (M =3.16, SD = 1.43; t (87) = 3.54, p < 0.01). However, this comparison is limited because, as the authors did not intend to compare the effects of presentation medium, the stimuli in study 1 and study 2 are not equivalent and extraneous variables may account for the medium effect on character judgements.
Conclusion The results of this systematic review demonstrate an excessive reliance on text stimuli in moral psychology: 92.2% of the included studies relied on only text stimuli. Unfortunately, it is unclear what the implications of the reliance on text stimuli are due to the limited number of comparisons between studies that have used text stimuli and those that have used non-text stimuli.
The few comparisons that were possible resulted in inconsistent implications. While comparisons between different presentation media in Lee and Gino (2015) and Brambilla and Chapter 5: Systematic Review 139
Riva (2017) suggest that presentation media rich in cues may lead to greater arousal and harsher judgements of moral character, paradoxically, media with fewer cues also lead to more mentalisation and more polarised moral judgements (Caruso & Gino, 2011). This finding is surprising as previous research has linked greater arousal with harsher moral judgements. It’s likely that there are multiple, underlying and (potentially) competing factors of presentation medium that impact morally relevant outcomes. For example, while the presence of additional non-verbal social and contextual cues in some non-text stimuli (e.g., video/image), compared to text stimuli, might lead to more arousal and harsher judgements, the abstract nature of text facilitates mentalisation which also contributes to harsher judgements. Additional research is required to explore the mechanisms that contribute to any medium effects on moral judgement.
This review did not clarify the effect of the overreliance on text stimuli in moral psychology. The comparisons within Lee and Gino (2015) and Brambilla and Riva (2017) are limited as the stimuli are not equivalent in content and thus any differences in moral judgement across presentation media will also result from differences in content.
Furthermore, there were also only three possible comparisons between text and non-text stimuli (including, Iliev et al., 2012), which is an insufficient sample size to draw any conclusions about the effect of presentation medium on the moral judgement or morally relevant factors. Thus, while this review provides a preliminary indication that presentation medium does have an impact on morally relevant outcomes, including judgement, arousal, and behaviour, the exact nature of the effect on these outcomes remains unclear. As a result, a non-text stimulus moral stimulus set with a text version that matches in moral content is required to explore the effect of presentation medium on moral judgement systematically.
Chapter 6: Developing a Film Set 140
Chapter 6: Developing a Moral and Affective Film Set Building on the limitations indicated by the systematic literature review (Study 3), this chapter describes the development of a stimulus set that provides the means for researchers to address the possibility that moral psychological processes are mischaracterised by the overreliance on text stimuli. Specifically, this chapter describes the development of a video stimulus moral film set, the moral and affective film set (MAAFS). I selected video as a presentation medium as it confers numerous advantages for use. First, the multi-modal nature of videos means that they closely approximate the real world, but do not pose the ethical and practical problems associated with placing participants in real, morally compromising situations (Schaefer, Nils, Sanchez, & Philippot, 2010). Second, because videos convey multiple kinds of information (verbal and non-verbal) via multiple channels
(visual, auditory), responses to video stimuli are less likely hinge upon text-related psychological capacities, such as verbal comprehension. Third, videos are an efficient medium for conveying information. Text conveys information using only verbal cues, while videos convey information with both verbal and non-verbal cues. An equivalent text description that includes both the verbal and non-verbal social context would be lengthy, and thus time-consuming to administer. Consequently, text is a less efficient means of communicating information relative to cue-rich channels of communication. Finally, video is potentially a more engaging presentation medium than text. Some researchers have reported that when participants are presented with video rather than text stimuli, they have greater motivation to participate and better attention over longer experimental sessions (Adolphs,
Nummenmaa, Todorov, & Haxby, 2016).
Chapter 6: Developing a Film Set 141
Overview of Stimulus Set Development and Validation My key goal was to develop a video stimulus set of ecologically valid, contextually rich stimuli encompassing a wide range of moral content. I thus used the broadest and one of the most prevalent characterizations of morality in psychological research, moral foundations theory, as a framework for the development of our stimuli (Graham et al., 2011). MFT categorises moral content into six foundations: care, fairness, loyalty, authority, sanctity, and liberty. MFT claims that these foundations represent the evolutionary bases upon which different cultures form systems of moral values (although our focus on MFT does not presuppose the evolutionary relevance of value categories; I use MFT to ensure breadth of coverage of moral content).
I developed the MAAFS using pre-existing video clips hosted on the video streaming website, YouTube. In the video collection phase, Amazon Mechanical Turk (MTurk) participants searched YouTube for potential clips using either vignettes that represented the moral foundations (Cannon et al., 2011; Clifford et al., 2015) or definitions of the moral foundations as search prompts. Participant-selected videos were assessed by the researchers on a broad set of initial inclusion criteria (details provided below) and the researchers manually searched for additional video clips to fill gaps in the sampling space. Selected videos were then rated by an independent sample of participants, in the video validation phase, on a range of moral dimensions. These validated videos were assessed against a second set of inclusion criteria (detailed below). The retained and rated videos (N = 69) formed the final video set. I provide URL links to the final videos to avoid infringing the terms and conditions of YouTube or copyright for distributing content (osf.io/8w3en). An overview of this process is presented in Figure 15.
Chapter 6: Developing a Film Set 142
Figure 15. An overview of the development of the MAAFS including video collection and video validation phases.
Method: Video Collection
Participants 175 participants from MTurk participated in the video collection phase (63 male, M age= 32.8, SDage = 10.1). The sample was highly educated on average: 85% of participants had some level of college education. No other demographics were collected.
Procedure and Materials Participants were asked to search YouTube for videos that represented either the provided moral vignettes or moral foundation definitions. Ninety-eight moral vignettes were drawn from previously validated text stimulus sets described in (Cannon et al., 2011; Clifford et al., 2015), described in Table 28. I also used moral foundation definitions (one definition per foundation) as alternative search prompts to broaden the search (definitions in Table 29).
Chapter 6: Developing a Film Set 143
One hundred and twenty-six participants were presented with moral vignettes as prompts; 49, with foundation definitions.
Participants were either presented with 10 randomly selected vignettes or two moral foundation definitions. Participants presented with vignettes were asked to search for a video clip that “most completely represents the content of each statement”, while those presented with foundation definitions were asked: “please find a video that you believe would make most people think of [moral foundation].” Participants were told that the video clip: (1) must be one minute or less in length, (2) must be hosted on YouTube, (3) must not contain obscene or offensive content (e.g., pornographic content), (4) must not include text as a central feature, (5) must be a moral transgression and not a praiseworthy action, (6) must be of actual scenes, events, people and real objects (not animations). Participants were then required to submit a URL link to a YouTube video for each vignette or foundation definition.
Participants were instructed that they could submit clips that were conceptually similar to the moral vignettes if an exact video match could not be found. Finally, participants were told that they could describe a video (e.g. a scene from a specified movie) if they were able to recall an appropriate video from memory but could not source a URL.
Table 28. A complete list of vignettes used in the development of the Moral Video Set
Original Moral Moral Vignette Author Foundation Cannon, Care Someone threw something at a dog that was barking (c) Schnall, and Loyalty Someone betrayed his family (c) White (2011) Authority Someone made fun of traditional things. (c) Authority Someone threw something at a politician she disliked. (c) Care Someone pinched a baby’s nose until it cried. Care Someone punched someone who bumped into him at a bar. Someone made cruel remarks to an overweight person about his Care appearance. Someone shot and killed an animal that is a member of an endangered Care species. Care Someone stepped on an ant hill, killing thousands of ants.
Chapter 6: Developing a Film Set 144
Care Someone hurt someone’s feelings by making fun of them. Fairness Someone took more than his share of the profits. Fairness Someone gave raises only to the employees he liked. Fairness Someone cut in front of some people in line waiting to buy tickets. Fairness Someone cheated in a game of cards. Fairness Someone hired only people of his own race. Fairness Someone marched in a “white power” KKK rally. Someone refused to help a friend move, after the friend had just helped Fairness him the week before. Loyalty Someone criticized her own country on a foreign television program. Loyalty Someone gossiped about a friend at work. Loyalty Someone bet against his home football team. Loyalty Someone broke of all contact with his family. Loyalty Someone burned his country’s flag at a protest rally. Loyalty Someone wore the opposing team’s strip to a home match. Authority Someone was disobedient to all authority figures. Authority Someone failed to fulfill the duties of his role. Authority Someone tried to create chaos and disorder at a party. Authority Someone became legally separated from her parents. Authority Someone cursed his parents to their face. Authority Someone insulted the royal family. Authority Someone made an obscene gesture to his boss. Sanctity Someone rarely showers and always smells bad. Sanctity Someone eats in the same place she goes to the bathroom. Sanctity Someone signed a piece of paper selling his soul on a dare. Sanctity Someone got a tattoo of a swear word on her neck. Sanctity Someone chose to have a surgery that split his tongue in two. Sanctity Someone ate an unwrapped chocolate bar that he found in the dustbin. Sanctity Someone injected drugs into his arm with a syringe. Sanctity Someone wrote “666” in hymn books and bibles in a church pew. Harm Someone tortured a stray cat. (e) Harm Someone stuck a pin into a child’s palm (e) Someone threw out a box of election ballots in order to help his favourite Fairness candidate win. (e) Someone stole money from a poor person and gave it to a rich person for a Fairness laugh (e) Ingroup Someone renounced her citizenship. (e) Ingroup Someone left her group of friends and got a new group. (e) Purity Someone cooked and ate his dog after it died of natural causes. (e) Clifford, Care A teenage boy chuckling at an amputee he passes by while on the subway. Iyengar, A woman commenting out loud about how fat another woman looks in her Cabeza, and Care jeans.
Chapter 6: Developing a Film Set 145
Sinnott- Care A man quickly cancelling a blind date as soon as he sees the woman. Armstrong Care A girl laughing when she realizes her friend's dad is the janitor. (2015) Care A girl saying that another girl is too ugly to be a varsity cheerleader. Care A teenage girl openly staring at a disfigured woman as she walks past. Care A boy making fun of his brother for getting dumped by his girlfriend. Care A man loudly telling his wife that the dinner she cooked tastes awful. A man telling a woman that her painting looks like it was done by Care children. Care A girl telling a boy that his older brother is much more attractive than him. Care A man laughing at a disabled co-worker Care A boy throwing rocks at cows that are grazing in the local pasture. Care A zoo trainer jabbing a dolphin to get it to entertain his customers. Care A man lashing his pony with a whip for breaking loose from its pen. Care A girl shooting geese repeatedly with a pellet gun out in the woods. Care A boy placing a thumbtack sticking up on the chair of another student. A woman spanking her child with a spatula for getting bad grades in Care school. Fairness A student copying a classmate's answer sheet on a makeup final exam. A runner taking a shortcut on the course during the marathon in order to Fairness win. Fairness A soccer player pretending to be seriously fouled by an opposing player. Fairness A referee intentionally making bad calls that help his favoured team win. A judge taking on a criminal case although he is friends with the Fairness defendant. Fairness An employee lying about how many hours she worked during the week. Fairness A boy skipping to the front of the line because his friend is an employee. Fairness A woman lying about the number of vacation days she has taken at work. Fairness A professor giving a bad grade to a student just because he dislikes him. Liberty A man telling his fiancé that she has to switch to his political party. Liberty A man telling his girlfriend that she must convert to his religion. Liberty A mother telling her son that she is going to choose all of his friends. Liberty A man forbidding his wife to wear clothing that he has not first approved. Liberty A mother forcing her daughter to enrol as a pre-med student in college. Authority A girl repeatedly interrupting her teacher as he explains a new concept. Authority An intern disobeying an order to dress professionally and comb his hair. Authority A teenage girl coming home late and ignoring her parents' strict curfew. Authority An employee trying to undermine all of her boss' ideas in front of others. Authority A player publicly yelling at his soccer coach during a playoff game. Authority A man secretly watching sports on his cell phone during a pastor's sermon. A group of women having a long and loud conversation during a church Authority sermon. Authority A man turns his back and walk away while his boss questions his work. A star player ignoring her coach's order to come to the bench during a Authority game.
Chapter 6: Developing a Film Set 146
An employee joking with competitors about how bad his company did last Loyalty year. A coach celebrating with the opposing team's players who just won the Loyalty game. Loyalty A mayor saying that the neighbouring town is a much better town. Loyalty A man leaving his family business to go work for their main competitor. Loyalty A teacher publicly saying she hopes another school wins the math contest. Loyalty The class president saying on TV that her rival college is a better school. A Hollywood star agreeing with a foreign dictator's denunciation of the Loyalty US. Loyalty A college president singing a rival school's fight song during a pep rally. Sanctity Two first cousins getting married to each other in an elaborate wedding. Sanctity A single man ordering an inflatable sex doll that looks like his secretary.
Someone leaving his dog outside(c) Care
A teacher hitting a student's hand with a ruler (c) Care Fairness Someone cheating in a card game (c) A politician inappropriately using federal tax dollars for his personal Fairness purposes (c) Liberty A father requiring his son to become the same profession as him (c) Liberty A public leader on TV trying to ban a type of clothing (c) Authority A student talking back to the teacher in front of the classroom. (c) Loyalty A man secretly voting against his wife in a local election. (c) Loyalty An American telling foreigners that the US is bad. (c) Loyalty Someone publicly giving up his citizenship to the US. (c) Harm A boy telling a woman that she looks just like her overweight bulldog. (e) Harm A man snickering as he passes by a cancer patient with a bald head. (e) Harm A girl telling her classmate that she looks like she has gained weight. (e) Harm A woman throwing her cat across the room for scratching the furniture. (e) A tenant bribing a landlord to be the first to get their apartment repainted. Fairness (e) A boss pressuring employee to buy goods from her family's general store. Liberty (e) Authority A girl ignoring her father's orders by taking the car after her curfew. (e) A staff member talking loudly and interrupting the mayor's speech to the Authority public. (e) Ingroup The coach's wife sponsoring a bake sale for her husband's rival team. (e) Harm A girl laughing at another student forgetting her lines at a school play. (e) Harm A woman clearly avoiding sitting next to an obese woman on the bus. (e) Harm A woman swerving her car in order to intentionally run over a squirrel. (e) Harm A boy setting a series of traps to kill stray cats in his neighbourhood. (e) A woman pressuring her daughter to become a famous evening news Liberty anchor. (e) Liberty A father requiring his son to take up the family restaurant business. (e)
Chapter 6: Developing a Film Set 147
A pastor banning his congregants from wearing bright colours in the Liberty church. (e) Authority A boy turning up the TV as his father talks about his military service. (e) Authority A student stating that her professor is a fool during an afternoon class. (e) A former US General saying publicly, he would never buy any American Ingroup product. (e) The US Ambassador joking in Great Britain about the stupidity of Ingroup Americans. (e) A head cheerleader booing her high school's team during a homecoming Ingroup game. (e) A US swimmer cheering as a Chinese foe beats his teammate to win the Ingroup gold. (e) Purity A man having sex with a frozen chicken before cooking it for dinner. (e) Purity A drunk elderly man offering to have oral sex with anyone in the bar. (e) A man in a bar using his phone to watch people having sex with animals. Purity (e) Purity A woman having intimate relations with a recently deceased loved one. (e) A homosexual in a gay bar offering sex to anyone who buys him a drink. Purity (e) An employee at a morgue eating his pepperoni pizza off of a dead body. Purity (e) A story about a remote tribe eating the flesh of their deceased members. Purity (e) A man searching through the trash to find women's discarded underwear. Purity (e) Note. (c) = vignettes that have been altered from the original published vignette; (e) = vignettes that were removed from the study for ethical concerns.
Chapter 6: Developing a Film Set 148
Table 29. Moral Foundation Definitions Used as Search Prompts in the Development of the Moral Video Set Care/harm This category of values relates to harming or caring for others. Actions are deemed wrong because they concern harming others (physically, psychologically, and/or emotionally). Actions are deemed right because they concern helping, caring for, protecting, or showing compassion towards others. Central concepts include care, compassion, empathy, help, support; harm, cruelty, emotional suffering, weakness/vulnerability
Fairness/cheating This category of values relates to justice, fairness, rights, and reciprocity. Actions are deemed wrong because they cause injustice, unfairness, or lack of reciprocity, and/or violate individual rights. Actions are deemed right because they uphold justice, fairness, or reciprocity, and/or uphold individual rights. Central concepts include fairness, reciprocity, justice, rights, equality, equity, proportionality, honesty; discrimination, cheating, bias, lying
Loyalty/betrayal This category of values relates to loyal duty and commitment to others (often to members of one's group, e.g., family, country, friendship group, team, or some other social group). Actions are deemed wrong because some duty or commitment to others has been neglected. Actions are deemed right because they uphold obligations, duties or loyalties to others. Central concepts include loyalty, unity, solidarity, alliance, ingroup, mateship; betrayal, disloyal, traitor.
Respect/subversion This category of values relates to showing respect to others, and/or showing proper deference/duty to a superior. Actions are deemed wrong because somebody is being disrespectful, and/or is undermining the leadership of the superior in that particular relationship or situation. Actions are deemed right because somebody shows proper respect to others, and/or is showing proper deference to the superior in the particular relationship or situation. Central concepts include respect, esteem, honour, roles, order, leadership, deference; disrespect, disobedience, insubordination, subversion NOTE: "Superior" can relate to the relationship, e.g., a boss is an employee's superior, a parent is a child's superior. "Superior" can sometimes also depend on the situation. E.g., if one person is giving advice/direction/instruction to another, he/she is the "superior" in that particular situation/task Sanctity/degradation This category of values emphasizes the purity and sanctity of human beings. Actions are deemed wrong because they threaten to defile the sanctity and decency of the human body, soul, or spirit. Actions are deemed right because they maintain purity and sanctity. Central concepts include purity, sanctity, decency, piety, sacred, cleanliness, wholesomeness; disgust, unnaturalness, depravity, sin, lewd, defilement. Liberty/oppression This category of values relates to the ability of individuals to act autonomously, free from oppression. Actions are deemed wrong because they undermine the ability of individuals to act freely or autonomously. Actions are deemed right because they uphold the ability of individuals to act freely or autonomously. Central concepts include liberty, freedom, autonomy, emancipation; oppression, tyranny, subjection, servitude.
Chapter 6: Developing a Film Set 149
Results and Discussion: Video Collection There were 742 video submissions in total: 344 videos were identified based on vignette search prompts; 398 videos, on the basis of definition prompts. The first and third authors reviewed each crowd-sourced video and made judgements regarding: (1) how well it represented the original vignette (vignette-based searches only), (2) how well it represented any moral event related to the target moral foundation, and (3) fulfilment of the video criteria.
Videos that were judged as inappropriate or inadequate on the basis of these criteria were removed from the next stage of video validation.
74 videos (Nvignette = 40, Ndefinitions = 34) fulfilled the stringent inclusion criteria.
Participants had more success in identifying videos primarily related to the care (N = 17) and fairness (N = 15), than to loyalty (N = 8), sanctity (N = 10), and liberty (N = 3).
Consequently, I manually searched for video clips for these under-represented domains. I again used the vignettes as a guide and followed the criteria given to MTurk searchers.
Nineteen additional videos were identified to give a total of 93 clips. Videos were assigned an initial ‘associated foundation’ as per the moral foundation classification of the previously validated vignette search prompts (Cannon et al., 2011; Clifford et al., 2015) or the foundation definition.
Methods: Video Validation A validation study was then run to collect normative ratings for the videos on a range of moral and affective dimensions.
Participants Videos were validated using a sample of Australian undergraduates and American
MTurk participants. I restricted MTurk workers to those with approval rates ≥ 90%, and ≥
100 previously approved HITs. After excluding 7 participants for failing attention checks, our final sample comprised 575 participants, including 253 Australian undergraduates and 322
Chapter 6: Developing a Film Set 150
American MTurk workers. The sample was 44% male and had an average age of 29.57 (SD =
12.82). The sample was composed of 14.2% self-identified political conservatives, 26.6% liberals, 13.5% moderates; 45% of participants chose to not to respond to this question.
American participants received a small monetary reward, while Australian participants were undergraduate psychology students that participated for course credit.
Sample size was determined based on a target of obtaining at least 30 ratings for each video on each dimension, although the average number of ratings was considerably higher (M
= 41.7). This number of ratings per stimulus is consistent with the validation procedure in the comparable moral text stimulus set, the moral foundation vignettes (Clifford et al., 2015).
Our sample size also matches or exceeds the sample size of studies that have validated affective video sets (Gross & Levenson, 1995; Philippot, 1993; Schaefer et al., 2010). A comparison between the sample size of the current study and the rating frequency of existing affective stimuli and the moral foundation vignettes is summarised Table 30.
Table 30. A Comparison Between the Rating Frequency and Sample Size of the MAAFs and Comparable Affective Film or Moral Stimulus Sets.
Citation Stimulus Number of Sample Number of Total Set Type Stimuli Size Stimuli Each Number Validated Participant Rated of Ratings Philippot Affective 20 60 12 720 (1993) Film Schaefer et Affective 70 364 10 3640 al. (2010) Film Gross and Affective 78 494 10 4940 Levenson Film (1995) Clifford et al. Moral Text ≈ 308 616 14 - 16 8642 (2015) 9856 MAAFS Moral Film 93 575 10 5750 Note. Clifford et al. (2015) do not report the total number of vignettes validated, only the number of vignettes retained in the final set (N = 132). I have estimated the total number of stimuli validated in their paper by taking median total number of ratings and dividing by 30 (the number of times each vignette was rated).
Chapter 6: Developing a Film Set 151
Procedure and Materials The validation procedure was drawn from previous studies reporting the development of affective video sets (Gross & Levenson, 1995; Philippot, 1993; Schaefer et al., 2010) and text-based moral stimuli (Clifford et al., 2015). Participants were asked to carefully watch a random subset of 10 videos from the pool of 93. After watching each clip, participants rated it on a range of moral and affective dimensions, before moving on to the next clip. Details of all questions asked, and response options are presented in Table 31.
After viewing each video, participants first provided ratings on several moral dimensions typically used in moral psychology research: wrongness, moral foundation relevance, emotional intensity, and punishment. Next, participants rated the discrete emotions that the video induced using the modified Differential Emotions Scale (DES) (Izard, 1993).
This scale has been used for the validation of several affective film sets (Philippot, 1993;
Schaefer et al., 2010) and measures 16 emotions (joy, surprise, anger, disgust, contempt, shame, guilt, fear, interest, sadness, awe, contentment, gratitude, hope, love, pride, and sexual desire). I added one item and altered the disgust DES item to distinguish between moral and core disgust. The original disgust item was changed from “disgust = disgusted, turned off, repulsed” to “disgusted” (captures moral disgust) and another separate item “grossed out”
(captures physical disgust). Prior studies have used this wording to distinguish between core disgust (“grossed out”) and moral disgust (“disgusted”) (Gutierrez, Giner-Sorolla, &
Vasiljevic, 2012; Herz & Hinds, 2013; Hutcherson & Gross, 2011). Participants also rated how funny they found the clip.
Participants then rated how frequently they witness or hear about the kind of moral act displayed by the video in their daily life and how weird the act is (in light of recent critiques of stimulus sampling bias in moral psychology research) (Gray & Keeney, 2015).
Participants next reported whether they have previously seen the video clip and briefly described the actions depicted in each video to ensure both that the clip was free from
Chapter 6: Developing a Film Set 152
technical problems and that the moral action was clearly depicted. Participants further
verified the clarity of the clip and the absence of technical problems by rating each of these
variables on a Likert scale.
Table 31. Summary of the Measures Used to Norm and Validate the Moral Videos
Measured Question Wording Response Scale Source Variable Wrongness How morally wrong is the (1) Not at all wrong - (5) extremely (Clifford et behaviour? wrong al., 2015) Moral Why is the action morally wrong? (1) It violates norms of care or care (Clifford et Foundation Select the main reason. (e.g., unkindness, causing pain al., 2015) Relevance to another) (2) It violates norms of fairness or justice (e.g., cheating or reducing equality) (3) It violates norms of loyalty (e.g., betrayal of a group) (4) It violates norms of respecting authority (e.g., subversion, lack of respect for tradition) (5) It violates norms of sanctity (e.g., degrading or disgusting acts) (6) It violates norms of freedom (e.g., bullying, dominating) (7) It is not morally wrong (8) It is morally wrong but none of the provided choices apply Punishment Should the actor in each clip be (1) Not at all - (5) very much New punished for their behaviour? Emotional How strong was your emotional (1) No emotion – (5) very strong (Clifford et Intensity response to the behaviour depicted al., 2015) in this scenario? Discrete How did watching the clip make (1) Not at all – (5) very (Philippot, Emotion you feel? Rate each of your 1993; emotions below: Schaefer et (1) Interested, concentrated, alert al., 2010) (2) Joyful, happy, amused *item (3) Disgusted* altered to (4) Fearful, scared, afraid distinguish (5) Anxious, tense, nervous (6) Disdainful, scornful, between contemptuous core and (7) Surprised, amazed, astonished moral (8) Warm-hearted, gleeful, elated. disgust (9) Loving, affectionate, friendly taken from (10) Guilty, remorseful (Gutierrez (11) Moved
Chapter 6: Developing a Film Set 153
(12) Satisfied, pleased et al., 2012; (13) Calm, serene, relaxed. Herz & (14) Ashamed, embarrassed. Hinds, (15) Grossed out* 2013; (16) Angry, irritated, mad Hutcherson (17) Sad, downhearted, blue & Gross, 2011) Commonness How often do you see or hear (1) Never – (5) constantly (Clifford et about actions like the one al., 2015) described in this scenario in the media or your daily life? Weirdness How atypical [i.e., weird, strange, (1) Not at all atypical – (5) very (Gray & unusual] are the actions or events atypical Keeney, in this clip? 2015) Previous Have you seen this clip before? (1) Never, (2) possibly, (3) at least New Exposure once, (4) more than once Humour How funny was the behaviour (1) Not at all – (5) extremely New depicted in the clip? Clip Clarity How clear did you find the events (1) Completely unclear - (7) New in the clip? completely clear Clip Clarity Describe the behaviour depicted Open response New in the clip in one sentence. Technical Did you have any technical (1) Yes [please specify] - (2) no New Problems problems displaying the clip?
Results and Discussion: Video Validation Three videos were reported as causing technical difficulties and so were removed
from the final video set. Videos were excluded from the final stimulus set if more than 20%
of participants selected the option “the clip is not morally wrong” when asked to select a
description of why the clip was morally wrong. Twenty-one videos were removed on this
criterion, leaving 69 videos conveying content deemed morally wrong. Summary descriptions
of the final video set are presented in Table 32 and detailed descriptive statistics for each
video are described in Appendix B.
Chapter 6: Developing a Film Set 154
Table 32. Summary Descriptions of the MAAFS
Video Video Description Moral Uniqueness Average Average Foundation Score Wrongness Arousal 1 A player insults and hits his coach Authority 29 4.5 3.5 2 A basketball player yells at his coaches Authority 3 3.8 3.1 3 An employee destroys her boss' laptop and Authority -12 2.8 2.9 office 4 Child swears at their guardians Authority 18 3.0 3.0 5 A kid sues her parents to pay for her Authority -34 3.3 2.9 education. 6 Students disrespecting teacher Authority 4 4.1 3.4 7 Children disrespect deaf mother Authority -9 3.3 2.8 8 A young man swears at police Authority 2 3.1 3.5 9 A boy is forcefully subdued by police due to Authority 12 3.3 2.4 fighting 10 A man disrespects the judge when he is on Authority 45 1.6 2.5 trial 11 A student disrupts the class Authority 56 3.1 2.5 12 Basketballers disrespect their coach Authority 43 3.5 2.5 13 Women gossip at work Care -16 4.3 2.8 14 Teacher hits a student with a ruler Care 71 3.9 3.1 15 Guys wont date a girl because she’s Care -16 2.7 2.4 overweight 16 People make fun of an overweight woman Care 23 2.8 2.2 17 Kids bully another kid for being overweight Care 7 4.5 3.9 18 Someone throws a shoe at a dog Care 91 2.1 1.8 19 Two girls fight each other Care 94 2.5 2.4 20 Someone throws a shoe at President George Care 11 2.9 2.4 Bush 21 Child is abandoned on the side of the road Care 59 2.0 2.1 22 A disfigured man was bullied on Instagram by Care 3 1.8 2.7 the athlete Shaq 23 A hunter kills an endangered rhino. Care 16 4.1 3.0 24 People are starving animals to death Care 48 2.1 2.0 25 A man is disrespectful toward his adoptive Care -14 3.7 3.1 parents 26 A mother yells at child Care 71 3.8 2.8 27 Someone purposefully trips fleeing refugees Care 29 3.5 3.1 28 A man punches a pregnant woman in the Care 60 1.6 1.4 stomach 29 A teenager disrespects her mother Care 3 4.3 3.3
Chapter 6: Developing a Film Set 155
30 Police in riot gear forcefully deal with Care -7 3.9 3.4 protesters 31 A man jumps a desk and punches a security Care 57 3.0 2.6 guard 32 A police officer assaults a woman Care 22 3.3 2.7 33 A man yells insults at his grandma Care 16 3.6 2.7 34 A bigger boy bullies a smaller boy Care 2 2.9 2.4 35 A man shoots at people Care 25 3.2 2.7 36 Protesters are beaten by police Care 9 3.8 2.7 37 A rich man steals money from a homeless Fairness 0 3.1 2.8 person 38 A man refuses to hire a woman, because she is Fairness 49 1.8 2.4 a woman 39 Someone is not hired for a job because of their Fairness 77 2.9 2.7 ethnicity 40 Vote are rigged during an election Fairness 85 3.8 3.0 41 Ballots are destroyed by setting them on fire Fairness 9 3.9 3.2 42 Woman lies to blind man about the value of Fairness 0 4.2 3.6 money bills 43 Man lies about a disability to get extra welfare Fairness 42 4.5 3.7 payments 44 A woman intentionally dents cans of food in Fairness 60 4.2 3.3 order to get a discount on the product. 45 A man backs out of a bet during a pool Fairness 0 4.4 3.8 competition 46 A guy fakes an illegal tackle to try and get a Fairness 38 3.3 2.6 free kick 47 A boy in a hurdle race cheats and runs around Fairness 79 2.7 2.2 the hurdles. 48 A man cuts a line so that he can get tickets Fairness 78 3.4 2.5 before other people 49 A man cheats on a game show Fairness 91 3.4 3.6 50 Student cheats in test Fairness 77 2.9 2.5 51 Lance Armstrong admitting to drug cheating Fairness 64 2.6 1.9 52 A woman lied to put her husband in jail Fairness 0 4.6 3.6 53 A man steals a bike Fairness -5 4.4 4.3 54 People rob an Apple computer store Fairness 0 4.0 3.1 55 A girl is forced to wear what her boyfriend Liberty 30 3.9 3.1 wants 56 A young girl is forced to marry an old man Liberty -37 4.3 3.7 57 The Chinese government censors the internet Liberty 37 4.5 3.7 for the Chinese citizens 58 African people are sold into slavery Liberty -6 4.4 3.8
Chapter 6: Developing a Film Set 156
59 A female group audition on talent show and Loyalty 80 2.5 2.5 one girl betrays the others for a chance to proceed in the competition 60 A woman catches man cheating Loyalty -5 3.7 3.0 61 A girl is betrayed by her boyfriend to avoid a Loyalty 24 2.7 2.5 criminal sentence 62 A guy cheats his family out of their money Loyalty -2 3.0 2.9 and property 63 Bride kisses best man on wedding Loyalty 13 1.9 2.0 64 Guy admits to cheating on girlfriend Loyalty 4 2.0 2.6 65 There is a theft from an infant's grave Moral Other 3.7 3.2 66 A girl goes to the bathroom and injects herself Sanctity -17 3.1 2.6 with drugs 67 A KKK ceremony Sanctity -53 3.1 2.3 68 A man takes drugs on a bus Sanctity -23 3.9 3.1 69 A woman steals flowers from grave Sanctity -38 4.3 3.8
Summary descriptive statistics for the MAAFS are presented in Table 33 and
distributions are depicted in Figure 16. I used the normative ratings of these 69 video clips to
explore the features of the MAAFS and implications for future researchers. The discussion of
the results will (1) describe the breadth and representativeness of moral content, (2) an
analysis of how uniquely each video portrays each of the moral foundations, (3) explore the
moral and affective features of the MAAFS, (4) a comparison between the performance of
the MAAFS as a mood induction procedure with pre-existing affective film sets, (5) describe
the effect of previous exposure on moral judgement, (6) consider ecological validity, and (7)
an inter-rater reliability analysis.
Chapter 6: Developing a Film Set 157
Table 33. Features of the Stimulus Set: Descriptive and Distributional Measures for each Variable
Item Mean SD Range Kurtosis Skewness Wrongness 3.80 0.55 2.74 - 4.80 -0.98 -0.16 Arousal 3.09 0.52 1.90 - 4.25 -0.64 0.11 Commonness 2.43 0.46 1.60 - 3.31 -0.98 -0.07 Funny 1.37 0.42 1.00 - 2.59 1.13 1.43 Punishment 3.11 0.71 1.41 – 4.38 -0.43 -0.27 Prior Exposure 1.29 0.52 1.00 - 3.63 9.42 3.03 Clarity 6.23 0.46 5.08 - 6.97 -0.15 -0.72 Weirdness 2.80 0.74 1.00 - 4.60 -0.44 0.03 Interested Concentrated Alert 2.67 0.23 2.02 - 3.15 0.29 -0.42 Joyful Happy Amused 1.38 0.22 1.00 - 2.19 1.70 1.09 Disgusted 2.89 0.60 1.64 - 3.96 -0.83 -0.27 Fearful Scared Afraid 1.82 0.51 1.09 - 2.97 -0.77 0.50 Anxious Tense Nervous 2.09 0.54 1.14 - 3.18 -1.01 0.18 Disdain Scornful Contempt 2.61 0.48 1.67 - 3.97 0.01 0.24 Surprised Amazed Astonished 2.20 0.36 1.45 - 2.88 -0.57 -0.14 Warm-hearted Gleeful Elated 1.26 0.14 1.00 -1.56 -0.24 -0.17 Loving Affectionate Friendly 1.24 0.15 1.00 - 1.57 -0.54 0.06 Guilty Remorseful 1.46 0.22 1.06 - 1.95 -0.26 0.33 Moved 1.59 0.32 1.09 - 2.33 -0.64 0.42 Satisfied Pleased 1.28 0.17 1.00 - 1.81 0.44 0.37 Calm Serene Relaxed 1.40 0.19 1.03 - 1.91 0.38 0.27 Ashamed Embarrassed 1.84 0.23 1.27 - 2.47 -0.09 0.04 Grossed out 1.88 0.45 1.05 - 3.03 0.14 0.51 Angry Irritated Mad 2.77 0.56 1.64 - 3.97 -0.84 -0.04 Sad Downhearted Blue 2.19 0.59 1.20 - 3.53 -0.84 0.33
Chapter 6: Developing a Film Set 158
Figure 16. Box-plots of averages for each video in the MAAFs for moral judgements, arousal.
Inter-Rater Reliability Video rating reliability was assessed using a mixed effects model. Typical methods of inter-rater reliability (e.g., ICCs, Cohen’s Kappa) could not be computed for this data set due to the completely random allocation of subsets of videos to raters. Instead, I ran a set of linear mixed effects models with crossed random effects (i.e., with random intercepts for participant and video) with the intercept as the only predictor. These models allowed us to partition variance into: (1) variance attributable to raters (i.e., random intercept variances for participant), (2) to videos (i.e., random intercept variances for video), and (3) residual
Chapter 6: Developing a Film Set 159
variance (summarised in Figure 17). The proportion of variance attributed to stimuli can be interpreted as inter-rater reliability.
The variance attributed to the videos ranged between 20 – 40% for judgements of wrongness, punishment, arousal, weirdness, and commonness. Importantly, wrongness and punishment ratings were associated with the highest proportion of stimulus-explained variance, suggesting reliable ratings of wrongness and punishment across the MAAFS. The variance attributed to the stimuli was very low for the positively-valenced emotions (e.g., friendly, pleased, relaxed, amused). This might imply that these positive emotions are highly subjective dimensions, relative to moral judgements. However, as the MAAFS contains moral transgressions only and not morally praiseworthy acts, the very high proportions of rater-variance may be driven by floor effects for these dimensions. There was very little variance across the videos in positive emotions, with most of these dimensions averaging between 1.0 – 2.0 (on a 1 - 5 scale), with very low standard deviations (e.g., 0.2).
Consequently, a substantial amount of the variance in positive emotion ratings may be driven by the raters or extraneous factors, rather than the videos.
Although there are no independent benchmarks for this type of reliability analysis, the rules of thumb proposed by Cicchetti (1994) for inter-rater correlations provide one possible comparison. According to these guidelines, these reliabilities range from “fair” (.40 ≤ ICC <
.60, as in the case of moral judgment) to “poor” (ICC < .40, as in the case of all other dimensions). However, it should be noted that (1) these guidelines were intended for the evaluation of clinical assessment instruments (which may be composed of multiple items), and that, (2) to our knowledge, ICCs are neither reported for validation studies of existing video sets (e.g. 64) nor textual moral stimulus sets (e.g. 2), making it difficult to provide a sufficiently similar reference point for comparison. Furthermore, these variances are also comparable to the ICCs described in Crone et al. (2016).
Chapter 6: Developing a Film Set 160
Figure 17. Proportions of variance attributable to video, rater, and residual error, sorted in order of variance attributable to video
Breadth and Representativeness of Moral Content Although all foundations were represented by multiple videos, the individualising foundations (care and fairness) were best represented: 24 clips were classified as care violations, 18 as fairness, 12 as authority, five as sanctity, six as loyalty, and four as liberty.
One video was primarily classified as ‘moral – other’.
This distribution of moral content is similar to the distribution found by experience sampling of everyday moral behaviour (Hofmann et al., 2014). In a large experience sampling survey (N = 1252), harm was by far the most common type of moral behaviour
Chapter 6: Developing a Film Set 161
experienced (50.6%), followed by fairness (13.9%), while the binding foundations were relatively uncommon experiences (5.6% authority, 5.2% sanctity, 4.8% loyalty, and 3.3% liberty). The MAAFS has a similar distribution: videos are primarily represented by the individualising foundations (34.8% harm violations and 26.1% fairness) and fewer videos represent the binding foundations (17.4% authority, 7.2% sanctity, 8.7% loyalty, and 5.8% liberty). Despite differences in methodology, the similarity in distributions of moral foundations suggests that the MAAFS samples types of moral acts at a similar frequency to which they occur outside the laboratory.
The goal of this stimulus set development exercise was not to develop a moral foundations video set but was rather to use MFT to select videos covering a broad range of moral content. However, I acknowledge that some researchers may be interested in studying each of the foundations in isolation and thus may wish to select videos that uniquely represent single foundations. To address this need, I calculated a uniqueness score for each video. To calculate this score, I took the percentage frequency that a given video was categorised as belonging to the target foundation and subtracted the percentage frequency that the video was categorised as belonging to any other moral foundation. A uniqueness score of
100 would indicate that all participants categorised the video as belonging to the target foundation, while a uniqueness score of -100 indicates that no participants categorised the video as belonging to the target foundation. Uniqueness scores for each video are available in
Table 32 and distributions of these scores within each foundation are displayed in Figure 18.
Across all videos, uniqueness scores ranged from -53 to 94 (M = 22.2; SD = 36.7). The overall distribution of uniqueness scores demonstrates that videos vary in the extent to which they uniquely represent moral foundations. Care, fairness and loyalty each had high maximum values, implying that at least one video in each foundation had very high, positive uniqueness scores. Importantly, each of care, fairness, loyalty, authority and liberty have at
Chapter 6: Developing a Film Set 162
least two videos with positive uniqueness scores, indicating the presence of videos in these foundations that predominately (if not exclusively) represent each foundation. Sanctity videos tend to overlap with the ‘moral other’ category and, thus, have low uniqueness scores. I suggest that this overlap demonstrates poor folk understanding of what defines sanctity, or a mismatch between folk and theoretical definitions. Although these videos are judged as morally wrong, participants don’t clearly categorize these videos into the sanctity foundation.
Figure 18. Box-plots of uniqueness scores for videos categorised into each moral foundation
Several videos were found to have uniqueness scores that were negative or close to 0
(particularly sanctity, liberty, and authority), which implies that there is overlap across the moral foundations. I generated a confusion matrix as a way of further exploring the variability in how the moral videos were categorised into moral foundations by participants
Chapter 6: Developing a Film Set 163
(Figure 19). A confusion matrix allows us to answer questions about how the moral foundations overlap in our videos, such as, are the fairness videos also often rated as care violations? Are the moral foundations uniquely represented by videos, or do videos represent multiple moral foundations? A confusion matrix is typically used as a way of visualising the accuracy of a classification procedure. The rows of the matrix represent how the data should be classified according to some ground truth (here, the modal moral content category describing why the content of the video was morally wrong), while the columns represent how the data was actually classified (here, the average percentage of times each moral content category was selected).
To create the confusion matrix the videos were first categorised by the moral foundation that participants most frequently selected as representative of the video’s moral content. This categorisation is represented by the row labels. Within each of these categories
(e.g., within the first row, representing videos whose modal moral content category was care),
I then calculated the average proportion of each content category such that each row sums to
1. For example, for videos with a modal content category of care (summarised in row 1), care was selected 62% of the time, and liberty 14% of the time. To look at "fairness" videos, you simply apply the same logic to row 2 (and so-on for each other category). The diagonal running from the top left to the bottom right represents how uniquely each of the moral foundations are represented in the MAAFS. According to this method, sanctity is the least uniquely represented foundation. Only 30% of participants categorised the sanctity transgressions as sanctity violations. Videos categorised as sanctity transgressions were most frequently classified as morally wrong but for a reason not represented by the moral foundations (28%). There is substantial overlap between certain foundations, videos that were categorised as liberty transgressions were also frequently categorised as fairness violations
(16% of the time). Care had some overlap with all the moral foundation, implying that some
Chapter 6: Developing a Film Set 164
participants perceived care violations across many videos. This is unsurprising, perceptions of harm and care are tightly linked to wrongfulness in western culture (Buchtel et al., 2015).
Care/harm has even been argued to be a superordinate moral domain, with all moral actions involving harm or care (Gray, Young, et al., 2012). In our data set, videos that were categorised as any of the moral foundations were rated as a care violation by at least 8% of participants. I conclude from the confusion matrix that individual videos generally convey information about multiple moral foundations. The rich variety of social and contextual cues available in videos likely signal more than one type of moral transgression. I suggest that the overlap across the domains may imply that when moving from text to richer media (such as video) moral content is not as easily delineated into discrete foundations. This may be more problematic for some domains than others, such as sanctity and liberty.
Chapter 6: Developing a Film Set 165
Figure 19. Confusion matrix where the moral videos (N = 69) are categorised according to the most frequently selected moral foundation and the proportion with which each alternative foundation was selected is shown. Row labels represent the modal moral content category describing why the content of the video was morally wrong. Columns represent the average percentage of times each moral content category was selected. The diagonal represents how uniquely each of the foundations are represented within the MAAFS, the off-diagonal reflects overlap between the moral foundations.
Moral and Affective Features
First, the MAAFS contains stimuli that clearly convey moral transgressions. As expected, the stimulus set had a high mean (3.80, on a 5-point scale) and minimum value
(2.74) for wrongness ratings. Clarity ratings were similarly distributed, with a high mean
Chapter 6: Developing a Film Set 166
rating (6.23, on a 7-point scale) and minimum value (5.08), indicating that moral transgressions are clearly conveyed by the MAAFS videos.
Arousal was near-normally distributed across the video set, with most videos clustering at the mid-point of the scale (mild arousal), although some videos evoked either very high or low arousal. This is consistent with our expectation that the moral content presented in video format would be effective at inducing (at least some) arousal, but also permits sampling across the arousal spectrum. Arousal was strongly and positively correlated with both wrongness and punishment (|Table 34).
The final set contains videos that can induce several morally relevant emotions. The distributions of discrete emotions are visualised in Figure 20.
Figure 20. Distributions of averages for each video in the MAAFs for discrete emotions
Chapter 6: Developing a Film Set 167
Other-condemning emotions were successfully induced across the video set. Across the MAAFS, there were high mean values for disdain, anger and moral disgust. There are individual video clips in the database that induced (on average) “a lot” of disdain, anger, and disgust (equivalent to the highest point on the scale). Certain videos were also effective at inducing shame, fear, physical disgust, sadness, surprise and anxiety (detailed in Appendix
B). Overall, the stimulus set elicited these negatively-valanced emotions to a similar degree to that of pre-existing affective-film sets (detailed below).
Of the discrete emotions, other-condemning emotions were most strongly correlated with moral judgement (see Table 34). There was a large, positive correlation between the other-condemning emotions and moral judgement, such that videos that were rated as very wrong or very punishable also elicited high levels of disdain, anger, and moral disgust. Fear, physical disgust, sadness, surprise, and anxiety were also moderately and positively correlated with both wrongness and punishment. Shame and guilt were only correlated with wrongness judgements and not punishment judgements.
Participants also felt engaged when watching the MAAFS videos.
‘Interested/concentrated/alert’ had the highest minimum value of the elicited emotions
(minimum = 2.02), suggesting that most videos evoked some interest from participants. This may imply that the cue-rich quality of videos as a communication medium creates an engaging way of conveying moral content. There was a moderate, positive correlation between the extent to which the video evoked interest and wrongness ratings (Table 33).
Videos were normed on funniness as there is some evidence that violations that elicit laughter may be judged differently (including less wrong, Warren & McGraw, 2015); the range (1.00 – 2.59) allows researchers to select videos on a variety of dimensions while controlling for funniness. Overall, the videos exhibited a positive skew in the ratings of
Chapter 6: Developing a Film Set 168
funniness: only one video exceeded an average rating of 3.0 (associated with a label of
“somewhat funny”). The remainder of the videos ranged from 1.00 – 2.63, with the majority
(61 of 69) falling between 1.0 and 2.0. Perhaps unsurprisingly, funniness was negatively correlated with judgements of wrongness (r = -.54), punishment (r = -.47), and arousal (r = -
.56).
The MAAFS could delineate between moral disgust and physical disgust. Recent research has shown that disgust is not unitary: moral and physical disgust are distinct (but correlated) variables (Chapman & Anderson, 2013). These forms of disgust are distinguishable at the level of individual videos. To quantify this, I calculated a moral disgust
– physical disgust (mean) difference score. Scores ≤ 0 reflect videos that primarily evoke physical disgust and scores ≥ 0 reflect videos that primarily evoke moral disgust. Seven
MAAFS videos (10%) primarily evoked physical disgust and 62 videos (90%) primarily evoked moral disgust (range: -0.27 – 2.03). I explored whether it was moral disgust or physical disgust that was associated with moral judgement by regressing judgements of wrongness onto both types of disgust. Moral disgust was the only significant predictor of wrongness judgements (Bmoral = 0.863, pmoral < 0.01, Bphysical = -0.75, pphysical = 0.478; F(2, 68)
= 63.150), with an equivalent pattern of results when regressing punishment onto each type of disgust (Bmoral = 0.629, pmoral <0.01, Bphysical = -0.149, pphysical= 0.331; F(2, 68) = 12.92, VIF =
2.17, tolerance = .47).
Comparison with Affect Induction Films The use of this stimulus set is not limited to moral researchers; affective researchers may wish to use the MAAFS to induce discrete emotions. To this end, I compare the performance of the MAAFS against three frequently cited affective film sets that have also normed videos using the DES or a very similar measure (Gross & Levenson, 1995; Philippot,
1993; Schaefer et al., 2010). For each of these stimulus sets, I took the highest (video-level)
Chapter 6: Developing a Film Set 169
mean for each discrete emotion reported by the authors (smallest means were not reported by all authors). In other words, I selected the best performing film clip for each of the discrete emotions. These values were then transformed to a common scale (1 – 7) and compared against the highest (video-level) mean in the MAAFS. These comparisons are visualised in
Figure 8. Unsurprisingly, the other affective sets out-perform the MAAFS on the induction of positive emotions. The affective films in (Gross & Levenson, 1995; Philippot, 1993; Schaefer et al., 2010) were selected to induce both positive and negative emotions, whereas the
MAAFS was selected to represent moral transgressions and so was expected to induce only negative emotions. The MAAFS performs particularly well when inducing contempt and disgust, with comparable maximum values to the (50, 54) stimulus sets. When comparing the
MAAFS and the stimulus sets in Gross and Levenson (1995); Philippot (1993), the MAAFS performed slightly worse on fear, sadness, and anger, with approximately one scale point difference. The MAAFS performed comparably to the most recently published stimulus set,
Schaefer et al. (2010); the MAAFS out-performed Schaefer et al. (2010) on anger and sadness and had similar maximum values for fear, disgust, and overall negative affect. This is encouraging as our measurement of discrete emotion was most similar to the method used by
Schaefer et al. (2010). While some of the difference in the scores of the MAAFS and the stimulus sets (Gross & Levenson, 1995; Philippot, 1993) could be the result of measurement error, score differences with Schaefer et al. (2010) are more likely to represent real differences in the films capacity to induce emotion. Overall, these results suggest that the
MAAFS is able to evoke negatively-valenced emotions to a similar degree to the stimulus sets in (Gross & Levenson, 1995; Philippot, 1993; Schaefer et al., 2010), but is unable to evoke positively-valenced emotions. Therefore, the MAAFS is suitable for use in affective research that seeks to induce discrete, negative emotions.
Chapter 6: Developing a Film Set 170
Figure 21. A comparison between affective film sets’ and the MAAFS’ capacity to induce discrete emotions. Values reflect the video-level average for the best performing film clip for each of the discrete emotions reported by the respective authors.
Chapter 6: Developing a Film Set 171
Previous Exposure to the Clips It is possible that participants may have had some prior exposure to some MAAFS
videos as the stimulus set contains movie/television video clips. Thus, I assessed the naivety
of participants to these videos and whether previous exposure influences judgments. First,
more than 90% of videos were rated as a < 2.0 (on average) for previous exposure, which
equates to “never seen before”. Second, I assessed if previous exposure affected moral
judgements (Table 34). Previous exposure was not correlated with any of the moral
dimensions, but there were small-moderate, positive correlations with some positive emotions
and clip clarity.
Table 34. Bivariate Correlations Between the Affective and Moral Ratings
1. 2. 3. 4. 5. 6. 7. 8. 1. Wrongness 2. Arousal .844** 3. Commonness -.166 -.176 4. Funny -.544** -.559** -.178 5. Punishment .904** .744** -.174 -.465** 6. Prior Exposure -.020 .004 .047 .112 -.088 7. Clarity .179 .164 -.123 .142 .030 .266* 8. Weirdness .363** .358** -.826** .131 .354** -.038 .143 9. Interested, Concentrated, Alert .360** .368** -.048 -.240* .266* .146 .277** .190 10. Joyful, Happy, Amused -.436** -.529** -.100 .739** -.431** .428** .212 .028 11. Disgusted .809** .831** -.128 -.627** .689** -.077 .107 .280* 12. Fearful, Scared, Afraid .575** .636** .113 -.534** .554** -.011 -.265* .167 13. Anxious, Tense, Nervous .571** .673** .199 -.613** .505** .075 -.211 .075 14. Disdain, Scornful, Contempt .797** .757** -.078 -.588** .686** .049 .076 .179 15. Surprised, Amazed, Astonished .443** .423** -.541** .080 .382* .077 .319** .639** 16. Warm-hearted, Gleeful, Elated . 041 -.127 .003 .122 -.014 . 370** -.102 .017 17. Loving, Affectionate, Friendly .166 -.009 .165 -.105 .083 .376** .126 -.112 18. Guilty, Remorseful .403** .409** .170 -.444** .197 .284** .165 -.099 19. Moved .505** .565** .108 -.475** .343** .273** .206 .019 20. Satisfied, Pleased -.071 -.232 .082 .178 -.079 .347** .121 -.088 21. Calm, Serene, Relaxed -.198 -.416** 0.100 .226 -.277* .362** -258* -.211 22. Ashamed, Embarrassed .307* .357** -.082 -.332** .174 .085 .228 .025 23. Grossed, out .552** .570** .011 -.501** .427** -.069 .043 .199 24. Angry, Irritated, Mad .785** .830** -.137 -.669** .700** -.010 .142 .214 25. Sad, Downhearted, Blue .669** .742** .009 -.641** .492** .136 .129 .117 Note. * p<0.05, **p<0.01, df = 67
Chapter 6: Developing a Film Set 172
Commonness Some researchers have raised concerns about the lack of ecological validity of typical moral stimuli, such as sacrificial dilemmas (Bauman, McGraw, Bartels, & Warren, 2014;
Gold, Pulford, & Colman, 2014; Hofmann et al., 2014). I addressed this concern by measuring the commonness of the moral action. The distribution of commonness scores suggests that the MAAFs includes a range of stimuli that are rated as commonly experienced:
7 videos were (on average) “sometimes” witnessed or heard about (≥ 3.0), and 43 videos were (on average) “occasionally” witnessed or heard about (≥ 2.0). This range allows researchers to choose (or manipulate) commonness as a key variable.
As mentioned previously, Gray and Keeney (2015) argue that existing sanctity stimuli suffer from a confound with weirdness. I assessed whether weirdness and commonness of the action varied as a function of moral foundation using bivariate correlation. I correlated the frequency that each video was categorised into each moral foundation with weirdness and commonness: commonness was not correlated with the (frequency of) categorisation into any moral foundation, but videos that were deemed weird were more frequently categorised as sanctity (r(67) = .329, p = .006). Less weird videos tended to be classified as loyalty violations (r(67) = -.245, p = .043). To further investigate the effect of weirdness and commonness on foundation classification, I regressed the frequency that each video was classified as sanctity onto both weirdness and commonness. The pattern of effects supports the correlational analyses: weirdness significantly predicted sanctity frequency, while commonness was a non-significant predictor (Bweird = 0.597, pweird =0.004, Bfrequency = 0.325, pfrequency = 0.113; F(2, 66) = 5.44, VIF = 3.15, tolerance = 0.32). These analyses suggest that the sanctity violations videos are not unusually uncommon but tend to be judged as weirder than violations in other foundations.
Weirdness, but not uncommonness, was correlated with moral judgement. Weirdness was associated with more wrongness, punishment, arousal, and less commonness (Table 34).
Chapter 6: Developing a Film Set 173
Commonness was not associated with moral judgement, despite the large correlation with weirdness (Table 34). According to Gray and Keeney (2015), weirdness is behaviour that is both uncommon and non-normative. Thus, it may be that only the non-normative aspect of weirdness (and not uncommonness) is morally relevant.
Possible Applications for the MAAFS The MAAFS have a wide range of possible applications for psychological research.
These videos can be used as the direct object of moral judgement, as a complement to text- vignettes. The cue-rich and dynamic nature of these clips allows researchers to explore a variety of interpersonal moral constructs such as, judgements of the victim’s/perpetrator’s moral character, attributions of blame or causality, intentionality, and empathy, in a non-text medium.
Researchers can use the normative ratings and video descriptions in Appendix B to strategically select videos that either manipulate or control for moral constructs of interest.
For example, if a researcher was interested in punishment of sanctity violations, videos could be arranged in descending order for (1) sanctity categorisations and (2) punishment.
Researchers may also wish to make use of algorithms that allows stimuli to be programmatically selected according to these normative ratings (Armstrong, 2012;
Constantinescu, 2017; Huber, 2017; van Casteren, 2007). For example, SOS (Armstrong,
2012) and Match (van Casteren, 2007) are software packages that select optimal stimuli from a database (e.g., MAAFS) based on the constraints specified by the experimenter (e.g., weirdness < 3.0).
Moral psychology researchers can use the MAAFS to study the contribution of specific information channels to moral judgement. Researchers can systematically vary cues by presenting participants with the MAAFS videos, audio-only versions of the MAAFS (i.e., no video), videos with no audio, and text-vignette transcription (refer to Chapter 7).
Chapter 6: Developing a Film Set 174
The MAAFS can be used to induce moral emotions and study their effects. For example, certain videos in the stimulus set can induce either moral or physical disgust, providing new ways to study the differential effects of each form of disgust. Likewise, the stimulus set can be used to induce the moral emotions of anger, contempt, and guilt.
Affective scientists can use these videos to induce emotions and study their effects.
The MAAFS has been normed on the same discrete emotions used to validate affective video sets and analyses reveal that the MAAFS performs equally or better at the induction of negative emotion (e.g., anger, guilt, sadness, contempt) when compared with existing affective stimulus sets (Gross & Levenson, 1995; Philippot, 1993; Schaefer et al., 2010)
(detailed analysis in S7). Affective stimulus sets are also typically normed only on emotions
(Gross & Levenson, 1995; Philippot, 1993; Schaefer et al., 2010) and ignore relevant variables that affective scientists may also wish to control or manipulate. The MAAFS videos are normed on a number of other, relevant dimensions, such as previous exposure, weirdness, and wrongness. Typically, affective stimulus sets rely on fictional behaviour from film scenes. The MAAFS presents a novel use of video-sharing technology by sampling fictional and non-fictional behaviours. Thus, the MAAFS expands the current choice of affective films in both number and type of film.
Conclusion Moral psychology has near-exclusively relied on text stimuli in the development and testing of theory. However, text stimuli lack the rich variety of morally-relevant social and contextual cues available in everyday interactions. The reliance on text-based stimuli may have systematically biased empirical research and psychological theories. Consequently, current moral psychology perspectives may not accurately account for moral phenomena in non-text or real-world contexts. I provide researchers with the means to move beyond the limits of text-stimuli by developing a cue-rich moral and affective film set (MAAFS). The
Chapter 6: Developing a Film Set 175
MAAFS includes moral transgressions that are diverse in content, intensity, and elicited emotions. I anticipate that the MAAFS will provide researchers with new insights into current theories and tools to develop a more complete understanding of moral psychology.
Chapter 7: Testing the Effect of Medium on Judgment 176
Chapter 7: Empirically Testing the Effect of Presentation Medium on Moral Judgement
The validation of the moral video set in Study 4 provided the tools to assess if moral judgement changes when moving from social and contextually poor presentation media (e.g., text) to social and contextually rich presentation media (e.g., video). This chapter will use this stimulus set to systematically examine how moral judgement is affected by presentation medium. First, I transcribe the MAAFS to equivalent moral (text) vignettes to ensure that the text and video stimuli are not confounded by differences in moral content. Then, I will then use the MAAFS and the matching text vignettes to (1) directly assess how presentation media affects moral judgement, and (2) identify implications for moral psychology as a field.
The MAAFS and a matching set of text vignettes (described below) were used to experimentally compare the effect of presentation medium (text and video) on various aspects of moral judgement. Study 3 revealed that very little research has examined the role of presentation medium on moral judgement and thus this study is largely exploratory.
However, certain hypotheses about how presentation medium affects moral judgement can be deduced by drawing from previous literature on the effect of non-verbal cues and presentation medium outside the moral domain.
Moral Foundation Categorisations First, presentation medium may affect the categorisation of a transgression into moral foundations. When a moral transgression is presented in text, there are fewer cues (verbal and non-verbal) to convey information about the nature of the transgression, compared to a transgression presented in video. This additional information about the moral transgression in the video condition may change the moral categorization in unexpected ways. In particular, the presence of more individuating information about the moral agent/patient (in a moral video versus moral text) may lead to more variation in the judgement in the categorisation of a transgression into moral foundations. Individuating cues can reduce stereotyping, which, in
Chapter 7: Testing the Effect of Medium on Judgment 177
turn, may result in less consensus in moral judgements and categorisation compared to when people use commonly-held stereotypes to inform their moral judgement. For instance, in the vignette ‘A lawyer tells a girl while in court that her boyfriend blamed all criminal activity on her to reduce his sentence’, the inclusion of lawyer may evoke stereotyped judgements about an actor that is cold, lacking emotion and thus likely to perpetrate a harmful act. As a result, the transgression may be judged, on average, as a harm violation. However, when viewing the moral video, non-verbal cues such as a concerned facial expression may undermine the ‘cold’ stereotype of a lawyer and perceptions of the lawyer as a perpetrator of a harm act. One possibility is that, when presented as a video, the boyfriend’s actions become more salient.
Instead, the boyfriend is perceived as the perpetrator of a disloyal action (betraying her to avoid receiving blame) and so the transgression is judged as a loyalty violation.
H1: There will be greater variation in why a moral action is wrong (i.e., categorisation along the foundation relevance scales) when the action is presented as a video, compared to when it is presented as a text vignette.
Arousal and Empathy The additional social and contextual cues in video, relative to text, may also have consequences for arousal. Several reviews show that the presentation medium influences the intensity and type of emotion experienced (Ellard, Farchione, & Barlow, 2012; Gerrards‐
Hesse et al., 1994; Lench, Flores, & Bench, 2011). Specifically, multi-modal stimuli (e.g. subtitled film, which includes visual, aural and verbal modalities) tends to elicit more intense emotional responses than text stimuli – particularly for anger and sadness (Ferrer et al., 2015;
Gerrards‐Hesse et al., 1994; Westerman et al., 2014). The social and contextual information conveyed by film clips is thought to increase undifferentiated arousal both directly by stimulating psychobiological processes typically aroused in FtF interactions (Kock, 2005)
Chapter 7: Testing the Effect of Medium on Judgment 178
and indirectly through increasing social presence (Riva et al., 2007). Therefore, I expect that moral content presented via video will be more arousing than equivalent moral texts.
H2: Videos will be more arousing than the equivalent text.
Relatedly, presentation media may also impact empathy. Certain social and contextual cues that are typically used in emotion recognition and empathy may be absent in text (e.g., facial expression, body posture, voice tone). Thus, the absence of this social cue (such as in text stimuli) may be a barrier to recognising emotion in others and, in turn, experiencing empathy.
Although the more explicit nature of text could compensate for the absence of non- verbal cues when conveying basic emotions, evidence suggests that more complex aspects of emotion recognition, such as intensity and intention, are more difficult to identify in text
(Derks et al., 2008; Sasaki & Ohbuchi, 1999). Sasaki and Ohbuchi (1999) compared emotion recognition via text and voice during a hostile conflict, while emotion recognition was equivalent in text and audio, angry emotions and perceived negative intent lead to increased aggression in the audio condition only. These results suggest that the non-verbal cues available in the voice condition (e.g., an angry tone) lead to another interpretation of another’s emotional state and a different behavioural response (e.g., aggression). In particular, the intensity of the emotion may be more difficult to recognise when verbal cues are absent. Other evidence shows that emotional recognition is greatest when perceiving a multimodal stimulus rather than unimodal stimuli. For example, empathetic accuracy is highest when both visual and verbal information is available to interactants (Gesn & Ickes,
1999; Hall & Schmid Mast, 2007; Zaki, Bolger, & Ochsner, 2009). Thus, some aspects of emotion recognition are more difficult when non-verbal cues are absent.
Given that empathy requires (at least, some) emotion recognition, the obstruction of emotion recognition may have consequences for the experience of empathy (Decety &
Chapter 7: Testing the Effect of Medium on Judgment 179
Jackson, 2004). Empirical evidence supports this possibility: motor empathy (e.g., unintentional mimicry) is greatest in dyads that can see each other (visual cues), compared to dyads that can hear each other (verbal cues) or both see and hear each other (Richardson,
Marsh, & Schmidt, 2005), and cognitive empathy (i.e., perspective taking) was greater for participants presented with videos of a robot than images of a robot (Zhao, Cusimano, &
Malle, 2016). Taken together with the evidence for a medium effect on emotion recognition, multimodal stimuli may lead to more cognitive and emotional empathy than the equivalent stimulus in text.
H3: Videos will generate greater feelings of empathy (emotional and cognitive) for victims than the equivalent text, as a function of the presence of visual and audio cues.
Wrongness Judgments Judgements of wrongness, a construct frequently linked to arousal, may also vary with presentation medium as a result of changes to empathy and arousal. Previous literature has drawn frequent links drawn between emotion and moral processes (Greene et al., 2001; Rozin et al., 1999; Schnall et al., 2008). For example, moral judgements have been shown to be more extreme when the experience of negative emotion is heightened (Wheatley & Haidt,
2005). Likewise, in certain circumstances, the experience of empathy can harshen moral judgement. According to Blair (1995, 2005), the experience of empathy when observing another suffering (as part of an immoral act) facilities judgement of wrongness. Evidence for this is found in psychopaths, who lack the experience of empathy and make equivalent wrongness judgements when presented with immoral acts and transgressions of social conventions (Blair, 1995, 2005; Young, Koenigs, Kruepke, & Newman, 2012). Therefore, if moral videos induct a stronger affective response (including more empathy), relative to text vignettes, then this should have downstream effects on moral judgement.
H4: Video violations will be seen as more wrong than text, and
Chapter 7: Testing the Effect of Medium on Judgment 180
H5: this difference will be accounted for by differences in empathy and arousal.
Person Perception Another important aspect of moral judgement, person perception, may also be affected be presentation medium. Study 1 (Chapter 2) found that in certain circumstances, there is an effect of presentation medium on humanness judgements. Specifically, audio cues
(present in video/voice and not text) can convey humanness qualities. As a result, when comparing text-restricted stimuli and voice stimuli, participants are less likely to attribute human uniqueness qualities to the author of a text passage (Schroeder & Epley, 2015, 2016).
Therefore, text stimuli will have a dehumanising effect relative to video stimuli.
H6: Participants will rate actors in text as having less human uniqueness than when those same actors presented in film.
Relatedly, if rich stimuli (e.g., video) convey more information about humanness than text stimuli, we might expect that video also conveys more information about qualities tied to humanness. Specifically, when a perpetrator is presented in video format (compared to text), there may be more social and contextual cues present that convey information about a perpetrator's mental state, such as the intentionality of the act and relatedly, causal responsibility (Cushman, 2008). For example, visual cues such as physical motion, directionality, and physical contact can portray causality and intention (Gao, Newman, &
Scholl, 2009; Iliev et al., 2012; Scholl & Tremoulet, 2000). Although some of these cues could also be conveyed verbally (i.e., in text descriptions), there is some evidence that presenting these cues visually may be more informative than presenting verbally.
Neuroimaging studies have shown that there are specialist brain regions that are sensitive to visual input of ‘biological’ motion (or motion of animals/humans) (Grezes et al., 2001). Thus, one possibility is that people are less sensitive to intention/causal information when represented in verbal description of motion rather than visual representation of motion. Given
Chapter 7: Testing the Effect of Medium on Judgment 181
that the perception of motion informs attributions of intentionality and blame, there may be fewer attributions when presented with verbal information about the transgression (i.e., text) when compared to a transgression conveyed with visual cues (i.e., video).
H7: Perpetrators presented in video will be perceived as acting more intentionally than those presented in text.
H8: Perpetrators presented in video will be judged as more blameworthy compared to perpetrators presented in text.
Individual Difference Moderators All of the above-hypothesized channel effects might be moderated by certain individual differences that previous researchers have already found to relate to communication medium. One relevant individual difference may be the variation in how people organise and process information, that is, cognitive style. Developmental and cognitive psychologists have linked cognitive style to presentation medium in decades of research (Clark & Paivio, 1991; Mayer, 2014; Mayer & Massa, 2003; Wittrock, 1989).
Cognitive style can be conceptualised in many dimensions (e.g., wholist-analyst, intuition- analyst, object-imagery and object-spatial), however, the visualiser-verbaliser dimension is most relevant to differences in text (predominately, verbal) and video (visual/verbal).
Visualisers preferentially attend to visual information and learn better when given visual instruction (Tsianos, Germanakos, Lekkas, Mourlas, & Samaras, 2009; Kraemer,
Rosenberg, & Thompson-Schill, 2009). In comparison, verbalisers preferentially attend to verbal content and learn better when given verbal instruction. Given this, I might expect that the extent to which someone prefers verbal content may result in more attention to and comprehension of moral content presented in video format (i.e., with visual cues). Greater attention to and comprehension of the moral transgression may then lead to more arousal and harsher moral judgments, than when those who prefer visual information are presented with
Chapter 7: Testing the Effect of Medium on Judgment 182
moral vignettes (i.e., verbal cues). I expect no relationship between the extent to which a participant prefers verbal information and medium, as video has both verbal and visual cues available.
H9: Cognitive style will moderate the channel effect on moral judgement, such that: those that prefer visual information will be more aroused and have harsher judgements when presented a moral video vs. a moral text, while there will be no effect of preference for verbal information on moral judgement, across the two media.
Within the text condition, some participants may have the ability to overcome the absence of social and contextual cues by simulating the social context with their imagination.
Individuals that enjoy reading fiction books (compared to those who prefer non-fiction) are more able to simulate rich social experiences using their imagination and develop better empathetic skills (Mar, Oatley, Hirsh, dela Paz, & Peterson, 2006). In comparison, those who speak English as a second language (referred to as ‘L2’) are less able to simulate social experiences from text compared to native English speakers (referred to as ‘L1’), and consequently experience less activation in brain regions associated with empathy (Hsu,
Jacobs, & Conrad, 2015). This difference may be partly because those who are not native
English speakers generally have lower verbal comprehension than Native speakers.
Therefore, I hypothesize that:
H10: Within the text condition: those with high imaginative ability will experience more empathy than those with low imaginative ability.
L1 or L2 English speaking and verbal ability will each moderate the channel effect on empathy, such that:
H11: L2 speakers will experience less empathy in the text condition than L1 speakers in the text condition, but this difference will disappear in the video condition where the visual cues will compensate for limitations associated with verbal ability.
Chapter 7: Testing the Effect of Medium on Judgment 183
H12: Individuals with low verbal ability will experience less empathy in the text condition than individuals with high verbal ability, but this difference will disappear in the video condition where the visual cues may compensate for limitations associated with verbal ability.
Chapter 7: Testing the Effect of Medium on Judgment 184
Transcription of Moral Videos to Moral Texts Two research assistants transcribed each moral video (N = 69) into a text vignette (of approximately 15 words), with the basic form "subject verb object + context". The 'subject' indicated the perpetrator of the moral transgression, the verb(s) referred to the moral action(s), and the object was the target of the moral action. The object could be another person, an animal, an inanimate object or the subject him or herself (e.g. when the moral action self-directed, such as in self-harm). Basic contextual information was also included, typically indicating location, (e.g., 'in a school, in a car'). Information about social roles was also included when the social roles were clearly apparent in the video. Such as, if it were clear that the relationship between the subject and object was mother-child.
As a result, vignettes took this basic structure:
A reporter [social role/subject] yells and throws his shoe [verb(s)] at George Bush
[social role/object] during a meeting [social context].
Transcriptions were reviewed and modified if they (a) did not match the above structure; (b) did not directly described the morally-central aspects of the video and/or (c) contained extraneous information, such as mitigating conditions. Twelve vignettes were altered (8%). This framework was used to ensure that transcriptions were as close as possible in syntactic structure and content to commonly used vignettes in moral psychology research
(e.g. Clifford et al., 2015; Cannon, Schnall, & White, 2011; Graham et al., 2011).
Vignettes were then rated for how well each represented the associated moral video.
This was to ensure that moral content was equivalent in the text and video stimuli. MTurk participants rated how well the transcriptions represented their associated video.
Chapter 7: Testing the Effect of Medium on Judgment 185
Method
Participants
109 Mturk participants (Mage = 32.9, SDage = 9.2, 50.5% female) rated the descriptions of the moral videos. No other demographic data were collected.
Procedure Participants were presented with four randomly-selected videos; each video was presented with the two matching transcriptions (transcription order was counterbalanced).
Participants were asked, “How well does this description represent the video above?”
Responses were collected on a five-point Likert scale: 1 (not at all), 2 (slightly), 3
(somewhat), 4 (well), 5 (extremely well). Participants were asked to provide feedback on what the vignette was missing from the transcription. Each transcription was rated at least 10 times.
Next, participants were also presented with two randomly selected videos, each presented with two random, mismatching transcriptions (transcriptions did not match the video). This was to ensure that participants were not rating any given transcription (even an inappropriate or mismatched transcription) as highly representative of a video.
Results
Inter-Rater Reliability Inter-rater reliability was calculated according to the procedures in (Rothbart & Park,
1986). Following this method, a Pearson bivariate correlation was calculated for each participant's ratings and the average ratings by all other participants. Raters that had poor agreement with the other raters (defined as > 1.5 standard deviation from the mean) were considered outliers and were removed from the analyses. Eight raters were removed from the main analysis.
Chapter 7: Testing the Effect of Medium on Judgment 186
Transcription Representativeness Average ratings of representativeness were calculated for each description.
Transcriptions that are rated, on average, between 4.0 - 5.0 represented the moral video ‘well’ or ‘extremely well’ and so were included in the main study. If two transcriptions for the same video met inclusion criteria, the transcription that received the highest average rating was selected. Where two transcriptions for the same video had equivalent means, the transcription with the smallest standard deviation was selected.
Of the 138 descriptions, 38 (27%) did not meet the inclusion criteria as they were rated as less than 4.0 (on average). There were 8 videos (11%) for which neither transcription met the inclusion criteria (Table 35). These transcriptions were modified according to feedback provided by participants and re-evaluated in a second wave of transcription validation. Iterations of modification and re-evaluation continued until the vignettes were rated, on average, greater than 4.0. After the second wave of text transcription12, 67 (97% of the MAAFS) text transcriptions were rated as acceptably representative of MAAFS videos.
After a third iteration13, all 69 videos were acceptably reliable; the reliability ratings and version number are described in (see Table 35 for the iterations of modifications and Error!
Reference source not found. for the final transcriptions).
All mismatched transcriptions had extremely low representativeness (M= 1.2, SD =
0.16) and all but two mismatched transcriptions had an average representativeness rating lower than 2.00. This suggests that the high representative ratings for the transcriptions matched to videos were valid.
12 During the second wave of transcription rating, 12 Mturk workers participated (25% female, M = 36.7, SD = 11.63), following the same procedure as the first round.
13 During the second wave of transcription rating, 8 Mturk workers participated (37% female, M = 34.5, SD = 10.41), following the same procedure as the first round.
Chapter 7: Testing the Effect of Medium on Judgment 187
Table 35. Waves of Modification and Representativeness Ratings for Vignettes With < 4.0 Representativeness Ratings
Wave 1 Wave 2 Wave 3
Original Description M SD Modified Transcription M SD Modified M SD Transcription Three men leave a 3.71 1.38 A woman wears a fat 4.7 0.48 date because the girl suit to meet guys from a was fatter than her dating application. The pictures suggest men leave the date when A woman wears a fat 3.93 1.14 she is fatter than her suit to meet guys from pictures implied. a dating application A woman changes 3.88 0.89 A woman engages in 4.5 0.53 other people's votes vote tampering by when counting the placing some yes ballot results cards into the no ballot A woman places the 3.94 1.18 card pile. yes ballot cards into the no ballot card pile during a voting process A man cheats during a 2.7 1.64 A man in a pool 3.5 1.58 A man harasses 4.6 0.70 game of snooker competition in a bar another guy playing cheats and retakes a shot pool and places a bet A man backs out of a 3.48 1.56 when he loses a bet. against him. When $200 bet when he he loses he tries to loses a pool get out of paying his competition in a bar debt.
A man blames all 2.92 1.55 A lawyer tells a girl 4.1 0.74 criminal activity on while in court that her his girlfriend to reduce boyfriend blamed all his sentence while in criminal activity on her court to reduce his sentence A boy admits he 3.23 1.24 would protect himself and give up his girlfriend during a court trial A young man 3.88 1.17 Young men repeatedly 4.5 0.53 repeatedly swears at swear at two off-duty two policemen in the policemen in the street. street A man swears at a 3.88 1.11 police officer in the street
Chapter 7: Testing the Effect of Medium on Judgment 188
Online content is 3.62 1.33 The Chinese 4.4 0.84 censored by the government censor what government in China the Chinese public see The Chinese 3.85 0.99 online and influence the government censor the public's beliefs. media from the Chinese public A mother is verbally 3.5 1.17 A reality TV show 3.9 1.20 A reality TV show 4.4 1.27 and physically abused captures a mother being captures a teenage by her teenage son in verbally and physically son yelling and their home abused by her teenage slamming a door on A young boy tries to 3.67 1.23 son in their home his mother in their slam the door on a home woman trying to talk to him A young man steals a 3.71 1.21 A young man steals a 4.3 0.68 bike and rides off with bike and rides off with it it in a carpark. A man steals a bike 3.94 0.90 and rides it in the street
Version Minimum Maximum Number Video Description Rating Rating M SD 1 A teacher hits a student with a stick during class 3 5 4.22 0.83 A woman wears a fat suit to meet guys from a dating application. The men leave the date when 2 she is fatter than her pictures implied. 4 5 4.70 0.48 A group of children taunt an old woman on the 1 bus for being fat 3 5 4.53 0.64 A group of schoolchildren tease a boy for being 1 overweight in the schoolyard 4 5 4.93 0.27 1 Someone throws shoes at their puppy 3 5 4.46 0.78 A woman is forced to wear skimpy clothing chosen by her boyfriend despite feeling 1 uncomfortable 2 5 4.30 0.88 A teenage girl is forced to marry an old man 1 against her will 3 5 4.71 0.59 A woman catches her boyfriend having sex with 1 another woman in their bedroom 2 5 4.07 1.03 A basketball player disrespects his coach before 1 failing to hit him 2 5 4.15 0.90 Two employees at work speak negatively about 1 their co-workers 4 5 4.62 0.51 A wealthy businessman steals money from a 1 homeless man 1 5 4.28 0.96 A qualified candidate is not given a job as she is a 1 woman 3 5 4.53 0.70
Chapter 7: Testing the Effect of Medium on Judgment 189
A poorly qualified woman with bad references is hired instead of a qualified man with a foreign 1 accent 3 5 4.56 0.62 A woman engages in vote tampering by placing 2 some yes ballot cards into the no ballot card pile. 4 5 4.50 0.53 Two offices containing boxes of votes 1 determining an election are purposely set on fire 3 5 4.38 0.77 A woman lies and then steals money from a blind 1 man 2 5 4.20 0.86 A man lies about having a disability to receive a 1 pension 1 5 4.25 1.24 A woman purposely damages tinned food at the 1 supermarket to receive a discount 3 5 4.53 0.64 A man harasses another guy playing pool and places a bet against him. When he loses he tries to 3 get out of paying his debt. 3 5 4.60 0.70 A basketball player yells at his coaches during a 1 game 1 5 4.30 1.25 A soccer player fakes an injury after not receiving 1 a free kick 1 5 4.59 1.00 An athlete runs around the hurdles instead of 1 jumping them during a race 3 5 4.25 0.71 A girl goes to the bathroom and injects herself 1 with drugs 3 5 4.50 0.71 A woman smashes her boss' computer after 1 getting in trouble 4 5 4.86 0.36 A toddler gives the finger and then swears at his 1 mother 3 5 4.39 0.70 Two women are separated during a fight in the 1 street 3 5 4.36 0.67 Someone throws a shoe at President George Bush 1 during a press conference 3 5 4.50 0.85 A lawyer tells a girl while in court that her boyfriend blamed all criminal activity on her to 2 reduce his sentence 3 5 4.10 0.74 Members of the KKK perform a ritual in the 1 forest 3 5 4.55 0.69 A man cuts to the front of a line of people waiting 1 to buy Broadway tickets 2 5 4.45 0.93 A man cheats during Who Wants to Be a 1 Millionaire 2 5 4.17 1.03 Parents abandon their child to highlight the 2 cruelty of abandoning pets 1 5 4.00 1.25 A disabled man is bullied about his appearance on social media by famous basketballer Shaquel 1 O'Neille 2 5 4.58 1.00 A poacher hunts and kills an endangered 1 rhinoceros 3 5 4.14 0.86
Chapter 7: Testing the Effect of Medium on Judgment 190
Farm animals are starving to death because of 1 negligent owners 4 5 4.50 0.52 A man betrays his family by legally stealing all of 1 their money 4 5 4.58 0.51 A father interviewed on a talk-show says he 1 wishes he never adopted his out of control son 2 5 4.15 0.93 A young woman sues her parents for refusing to 1 pay her college tuition 3 5 4.44 0.73 A bride cheats on the groom with the best man on 1 their wedding day 3 5 4.76 0.54 A woman pranking her boyfriend discovers that 1 he cheated on her 3 5 4.30 0.68 A student is filmed copying off his peer during a 1 test 4 5 4.78 0.43 A teacher is verbally abused and physically 1 intimidated by her students during class 3 5 4.65 0.70 1 Children disrespect their deaf parents at home 4 5 4.72 0.46 A man films himself drinking a whole bottle of 1 spirits in a public library 2 5 4.30 1.06 Young men repeatedly swear at two off-duty 2 policemen in the street. 4 5 4.50 0.53 A mother cruelly punishes her young son for 1 getting into trouble at school 3 5 4.44 0.73 Lance Armstrong admits to drug use during an 1 interview with Oprah 3 5 4.47 0.64 After declaring their close friendship, a young woman abandons her other band members during 1 an audition 4 5 4.60 0.51 The Chinese government censor what the Chinese public see online and influence the public's 2 beliefs. 3 5 4.40 0.84 1 A man snorts cocaine on public transport 2 5 4.42 1.08 A woman is interviewed after objects from her 1 child's grave are continually stolen 3 5 4.60 0.63 A woman quickly removes all the flowers from a 1 grave before running off 2 5 4.23 1.01 A woman is interviewed by Dr Phil about lying to 1 the police about her husband 4 5 4.59 0.51 A reporter kicks migrant families fleeing from 1 police 2 5 4.58 0.90 A teenage boy fails to comply with police officers 1 after he is found breaking the law 3 5 4.50 0.71 A pregnant woman is punched in the stomach by 1 her boyfriend on the street 4 5 4.64 0.51 A young man disrespects the authority of a Judge 1 whilst in court 3 5 4.53 0.74 A reality TV show captures a teenage son yelling 3 and slamming a door on his mother in their home 1 5 4.40 1.27
Chapter 7: Testing the Effect of Medium on Judgment 191
A student verbally disrespects her teacher during 1 class 2 5 4.28 0.96 A student talks back to his basketball coach 1 before attempting to hit him 3 5 4.57 0.65 A slave-owner displays his slaves to potential 1 buyers 3 5 4.69 0.63 Policeman use batons and excessive physical 1 force to control people in the street 2 5 4.19 0.80 1 A security guard is attacked by a young man 2 5 4.56 0.96 1 A police officer punches a woman 3 5 4.32 0.78 A young man steals a bike and rides off with it in 2 a car park. 3 5 4.30 0.68 A man disrespects his wheelchair-ridden 1 grandmother in public 3 5 4.29 0.77 A young boy is hit by a larger group of teenage 1 bullies 4 5 4.33 0.50 A soldier shoots defenceless prisoners from his 1 balcony 2 5 4.29 0.91 Policemen use sticks to try to control a mob of 1 protesters 3 5 4.23 0.60 A group of men break in and steal computers 1 from an Apple store 4 5 4.64 0.51 1 A man has sex with a donkey outdoors 3 5 4.57 0.76 1 Dogs are eaten in Vietnamese street markets 3 5 4.29 0.73
Testing the Effect of Presentation Medium Next, the complete set of text transcriptions and video stimuli were used to explore
the effect of presentation medium on moral judgement.
Method
Design Presentation medium was manipulated within-subjects: participants were assigned to
view 10 videos and read 10 text vignettes. Stimuli were pseudo-randomly generated from the
pool of 69, that is, stimuli were randomly generated except to ensure that participants were
not shown the same moral content in both video and text.
Chapter 7: Testing the Effect of Medium on Judgment 192
Participants 203 Mechanical Turk workers (92 female) were presented with a random sample of the 10 text vignettes and 10 moral videos from the MAAFS.
It is very difficult to calculate the required sample size for a mixed-effects model
(Bates, Mächler, Bolker, & Walker, 2014). Instead, sample size was estimated with reference to the more basic statistical analyses that could be used to analyse these data. Thus, according to a two-sided, within-subjects t-test, 198 participants would achieve 80% power, assuming a small effect size (d = 0.2).
The average age of the sample was 35.9 (SD = 10.25). The sample was highly educated; 85.7% of participants had at least some tertiary education. The sample was balanced for political ideology; 26.6% conservatives, 22.2% moderates, 28.6% liberals,
11.3% libertarians, and 10.4% were apolitical or identified as another ideology. Finally, 63% of the sample was atheist and the average religiosity was low in the theists (M = 1.64, SD =
.48). The sample was largely native English speakers (N=197 or 98%).
Procedure and Materials The major foci of this study were measures of moral judgement, that is, wrongness, moral foundation categorisation, and blame. These items are typically measured with a single item. I also included a range of other single-item measures for exploratory purpose. While there is a reliability/brevity trade-off for the exploratory items, it’s justified given the primary focus of the study was the judgement variables. Study measures are summarised in Table 36 and descriptive statistics for all measures are displayed in Table 37.
Participants were randomly presented with either 10 moral texts and 10 moral videos.
For each moral stimulus, the participant first reported any technical problems (e.g., video did not display), then rated wrongness, moral foundation relevance, and arousal (these measures are the same as described in Study 4).
Chapter 7: Testing the Effect of Medium on Judgment 193
Table 36. Summary of the Measures Used to Rate the Video and Text Stimuli
Measured Question-Wording Response Scale Source Variable Emotional To what extent do you feel the 1 (Not at all) - 5 (very much) New Empathy emotions that the victim feels? Cognitive To what extent do you see things 1 (Not at all) - 5 (very much) Empathy from the victim’s perspective? Humanness People can vary in how human-like Kteily et al. they seem. Some people seem highly (2015) evolved whereas others seem no different than lower animals. Using 1 2 3 4 5 the image below, indicate using the sliders how evolved you consider [the perpetrator | the victim] Intentionality Did the perpetrator act intentionally? 1(Strong no) - 7 (Strong yes) Mele and Cushman (2007) Cause To what extent is the perpetrator the 0 (Not at all the cause) - 100 Hilton et al. cause of the outcome (Completely the cause) (2005); Lagnado and Channon (2008); McClure et al. (2007) Blame How much blame does the 1 (None at all), 4 (some), 7 (Cushman, perpetrator deserve? (very much) 2008) Cognitive In a learning situation, sometimes 1 (Strongly more verbal than Mayer and Style information is presented verbally visual) - 7 (strongly more Massa (e.g., with printed or spoken words) visual than verbal) (2003) and sometimes information is presented visually (e.g., with labelled illustrations, graph, or narrated animations). Please select the answer that represents your learning preference: Imaginative Fantasy subscale of the IRI Multiple Items Davis Ability (1980) Verbal Shipley Institute of Living Scale – Multiple Items Zachary Ability Verbal subscale and Shipley (1986) English Is your first language English? Yes/No New Nativity Note. Measures also included in previous studies are excluded from the table, e.g., wrongness, moral foundation relevance, and arousal. For details on those measures, refer to table (cannot be included until chapters are combined).
Chapter 7: Testing the Effect of Medium on Judgment 194
Blame Next, participants rated how much the perpetrator should be blamed.
Humanness Next, participants rated the victim and perpetrator in each stimulus for humanness. A shorter measure of humanness was used in this study, compared to Study 1 and Study 2. The
Ascent of Man scale is the most widely validated (and perhaps only) single-item measure for humanness (Kteily, Bruneau, Waytz, & Cotterill, 2015). The scale depicts the evolution of man from ape to human form. This item is sensitive to changes in human uniqueness, given that when people are denied human uniqueness qualities they are likened to animals (e.g., apes) (Haslam et al., 2005). Human nature is not assessed here as there is no succinct measure available and in Study 1 and Study 2 presentation medium affects human uniqueness but not human nature.
Cause and Intentionality Participants then rated cause and intentionality of the moral perpetrator. Cause and intentionality are often manipulated and rarely rated; the measure was taken from the few studies that have measured these constructs have used variants on this wording, although a slight change in wording is necessary for this context (Hilton, McClure, & Slugoski, 2005;
Lagnado & Channon, 2008; McClure, Hilton, & Sutton, 2007; Mele & Cushman, 2007).
Empathy State empathy was measured in terms of cognitive (i.e., perspective taking) and emotional empathy. Participants rated how much cognitive and emotional empathy with the victim they experienced when reading/viewing the moral stimulus. This study used new, ad hoc, state cognitive and emotional empathy measures. Most measures of empathy are at the trait level and not the state (e.g. Interpersonal Reactivity Index, Davis, 1980, and Toronto
Empathy Scale, Spreng, McKinnon, Mar, & Levine, 2009). The few scales that measure state
Chapter 7: Testing the Effect of Medium on Judgment 195
empathy are too long to be used in this study (State Empathy Scale, Shen, 2010), or are not self-report measures (Multi-Faceted Empathy Scale, Dziobek et al., 2008, Reading in the
Mind’s Eye, Baron-Cohen, Wheelwright, Hill, Raste, & Plumb, 2001, Pictorial Empathy
Test. Koirikivi, 2014).
Individual Differences Finally, participants completed a battery of brief individual difference measures, including cognitive style, verbal ability, empathy, and demographics.
First, imaginative ability was measured using the fantasy subscale of the Interpersonal
Reactivity Index (IRI, Davis, 1980). Participants completed the fantasy subscale using a five- point Likert scale range from 1 (does not describe me well) to 5 (describes me well). The scale included seven items, the scores for each item were averaged to give an overall score for each participant ( = 0.87).
The selection of a verbal ability measure was limited to those measures that were not proprietary. As this study is run online, many of the most commonly used verbal ability measures could not be used in this study because of copyright limitations (e.g., the WAIS-III,
Wechsler, 2008). Other alternative verbal measures were too long for this study (e.g., Nation
& Beglar, 2007). As a result, an older measure, the vocabulary subtest of the Shipley Living
Institute Scale (SLIS) was used (Nation & Beglar, 2007). The Shipley Living Scale is a brief scale (10 minutes) for testing intellectual functioning that is designed for computer administration. Despite its age, the SLIS has good psychometric properties, including good re-test reliability (median 0.79) and high correlation with the WAIS (0.85) (Naglirei &
Graham, 2003; Zachary & Shipley, 1986). Participants completed the vocabulary subtest only. The vocabulary test contained 40 multi-choice items of increasing difficulty on which the participant selected a synonym for each stimulus word. Participant responses were
Chapter 7: Testing the Effect of Medium on Judgment 196
dummy coded according to if the correct (1) or incorrect (0) synonym was selected.
Participant’s scores were then summed to give a result out of 40 ( = 0.87).
Next, participants completed a single item to determine their cognitive style.
Participants completed the single item, verbal-visual learning style measure (Mayer & Massa,
2003). This involved participants rating the degree to which they are more visual or verbal learners on a 7-point scale. This single item measure is equally effective as traditional multi- item questionnaires (i.e., Verbaliser–Visualiser Questionnaire and Santa Barbara Learning
Style Questionnaire) at measuring cognitive style and is correlated strongly with these questionnaires (Mayer & Massa, 2003).
Participants also disclosed whether their native language was English (yes or no).
Religiosity was measured in two ways: first, participants were asked whether they identified with any religion or denomination with a binary scale (yes or no). Next, participants rated their religiosity. Participants were asked “To what extent do you identify yourself as a religious person?” and responded with a 5-point Likert scale, 1 (not at all) to 5 (very much so). Age and gender were described using open-ended responses. Finally, participants identified their political orientation using three items. Participants were asked “To what extent would you describe yourself on each of the following items… liberal issues/economic issues/in general” and responded using a fully labelled 7-point Likert scale, 1 (strongly liberal) to 7 (strongly conservative).
Chapter 7: Testing the Effect of Medium on Judgment 197
Table 37. Descriptive Statistics for Dependent Measures Variable Mean SD Skew Kurtosis Wrongness 3.68 1.18 -0.68 -0.33
Arousal 3.17 1.22 -0.18 -0.91
Blame 4.46 1.93 -0.31 -1.03
Causal Responsibility 84.31 22.73 -1.75 2.72
Cognitive Empathy (state) 3.45 1.31 -0.42 -0.96
Emotional Empathy (state) 3.27 1.35 -0.27 -1.11
Intention 6.12 1.41 -1.76 2.46
Humanness of Victim 4.57 0.93 -2.36 5.03
Humanness of Perpetrator 3.7 1.38 -0.69 -0.84
Verbal IQ 34.66 5.47 -1.51 2.97
Cognitive Empathy (trait) 3.55 0.62 -0.32 -0.03 Emotional Empathy (trait) 3.72 0.83 -0.43 -0.20 Fantasy (trait) 3.28 0.89 -0.33 -0.18
Results
Analysis Strategy A mixed-effects model was fit for each of the dependent measures with random effects for participant and stimulus. As was the case for previous studies, these mixed-effects models account for variation associated with the participant and the stimulus by specifying these variables as random effects. Results for the mixed models are presented in Table 38.
Correlations and Descriptive Statistics First, an initial exploration of the data is presented in a correlation matrix in Figure
22. Unsurprisingly, the moral judgement variables (e.g., wrongness, intention, blame) strongly positively co-vary. Likewise, arousal and empathy (both cognitive and emotional)
Chapter 7: Testing the Effect of Medium on Judgment 198
positively correlate with the measures of moral judgement, consistent with theorising.
Humanness of the perpetrator was strongly and negatively related to moral judgement and arousal/empathy, suggesting that the more human the perpetrator is perceived, the less they are perceived as culpable, wrong, or acting intentionally (or vice versa). For victims of transgressions, humanness perceptions were positively related to empathy (cognitive and emotional), and unusually, intention.
There are some surprising relationships with the individual difference measures and judgement variables. For example, the extent to which someone was a visualiser (high scores on the measure), positively related judgements of intention and humanness of the victim. It’s not clear why the extent to which someone prefers visual information would be associated with perceiving a victim as more human or judging behaviours as more intentional. Less surprising, the extent to which someone was a verbaliser (low scores) negatively related to verbal ability (i.e., verbalisers and high verbal ability scores tend to co-occur). Verbal ability was negatively correlated with imaginative ability, emotional empathy (but not cognitive), and arousal, but positively correlated with intention and humanness judgements of the victim.
Finally, imaginative ability was positively correlated with emotional and cognitive empathy, arousal, and blame.
Chapter 7: Testing the Effect of Medium on Judgment 199
Figure 22. Correlational matrix of dependent measures and presentation medium. Note. *p <
0.01 ** p < 0.05, *** p < 0.01, any cell that does not contain an asterix is a non-significant correlation. Colour of the cell refers to the directionality and magnitude of the relationship.
Chapter 7: Testing the Effect of Presentation Medium
Table 38. Results for the Mixed-Effects Models
Perp. Vic. Wrongness Arousal Emotional Empathy Cognitive Empathy Intent Blame Human Human
Hypothesis Tested H4 H9 H2 H9 H3 H10 H11 H12 H3 H10 H11 H12 H6 H6 H7 H8 Fixed Effects
Medium Condition -.07 [- -.09 [-.37, 0.08 [-.002, -.38 [- 0.05 -.228 [-.88, .03 [-.06, -.03 [- -0.02 [- .07 [- -.02 [- -.49 [- -.15 [- -0.14 -0.03 0.04 (Text = 0, Video = 1) .15, .01]# .20] .16]# .33, .25] [-.05, .43] .12] .62 - .11, .08] .56, .82] .12, .08] 1.09, .23, -.08] [-.20, -.06] [-.13, [-.10, .16] .12] .65] .07] .007]
Cognitive Style .02 [-.04, .00002 .08] [-.08,
.07] Imaginative Ability .21 [-.09, .48] .14 [- .13 - .42]
Native English Speaker .34 [-.50, .26 [- (1 = Native, 2 = Non-Native) 1.15] .12, .08]
Verbal IQ -.02 [- -.01 [- .03, .03, .01]# .01] 2-Way Interactions
Condition*Cognitive Style .004 [.05, 0.02 [- .06] .03, .08]
Condition*Imaginative .09 [-.12, .31] -.03 [- Ability .27, .18]
Condition*Native Language .03 [-.58, - -.19 [-
Speaker .57] .77, .40] Condition*Verbal IQ .002 .01 [- [-.02, .003 - .02] .03]
Note: The table presents the 95% confidence intervals for unstandardized regression coefficients; confidence intervals that do not include zero are in boldface and the cells are shaded grey. # Refers to significant (or near-significant) t-values but confidence intervals that contain zero. The hypotheses tested row refers to the values in the column below and refer to which hypothesis is tested by each model.
Chapter 7: Testing the Effect of Presentation Medium 201
Moral Foundation Categorisation (H1)
First, I tested whether there was equal variance of moral foundation categorisations across the text and video conditions. Data were aggregated to the group level and a χ2 test assessed the distribution of moral foundation categorisations in each presentation medium.
The difference in distributions of moral foundation categorisations between conditions approached significance: χ2 (7) = 12.9, p = 0.07. Thus, there is marginal support for H1, that there is a difference in the variation of the categorisation of a moral action when the action is presented as a video, compared to when it is presented as a text vignette (H1).
Given that the difference in distributions of moral foundations was nearing significance, I generated a confusion matrix as a way of further exploring the variability in how the moral videos and text were categorised (Figure 23 and Figure 24). A confusion matrix is typically used as a way of visualising the accuracy of a classification procedure. The rows of the matrix represent how the data should be classified according to some ground
‘truth’. In this case, the ground truth is the modal moral content category describing why the content of the video was morally wrong, derived from Study 4. The columns represent how the data were actually classified by participants in this study (i.e., the average percentage of times each moral content category was selected).
To create the confusion matrix the videos and text were first categorised by the moral foundation that participants most frequently selected as representative of the video’s moral content. This categorisation is represented by the row labels. Within each of these categories
(e.g., within the first row, representing videos whose modal moral content category was care), we then calculated the average proportion of each content category such that each row sums to 1. For example, for videos (Figure 23) with a modal content category of harm
(summarised in row 1), harm was selected 61% of the time, and liberty 11% of the time. To look at "fairness" videos, you simply apply the same logic to row 2 (and so-on for each other
Chapter 7: Testing the Effect of Presentation Medium 202
category). The diagonal running from the top left to the bottom right represents how often the participants agree with the categorisation into foundations from Study 3. Off-diagonal reflects how much disagreement there is in the categorisation into moral foundations.
A descriptive comparison of the two confusion matrices reveals, that there is less consensus regarding the classification of purity and liberty transgressions when moral content is presented in video compared to text. Specifically, liberty violations presented in video were categorised as liberty transgressions only 42% of the time, compared to 62% of the time in the text condition. Likewise, purity violations were categorised as purity transgressions only
29% of the time when presented in video, compared to 43% of the time in text. In particular, purity transgressions when presented in video format tend to be categorised (20%) as not wrong or wrong but not for any of the reasons described by the moral foundations (19%).
These results are consistent with H1: for some kinds of moral violations (i.e., purity and liberty transgressions) there is more variation in categorisations of the violation when the behaviour is presented via video compared to text.
Chapter 7: Testing the Effect of Presentation Medium 203
Figure 23. Confusion matrix where moral videos are categorised according to the most frequently selected moral foundation and the proportion with which each alternative foundation was selected is shown.
Figure 24. Confusion matrix where moral texts are categorised according to the most frequently selected moral foundation and the proportion with which each alternative foundation was selected is shown.
Chapter 7: Testing the Effect of Presentation Medium 204
Arousal and Empathy (H2 and H3) Next, mixed-models were fit to test the main effect of presentation medium on (1) arousal, (2) emotional empathy, (3) cognitive empathy. First, the effect of condition on arousal was marginally significant (t(201)=2.00). The marginal significance tests may be because the effect was so small (ß = 0.081) that there was- insufficient power to detect the effect. Post-hoc power calculations show that the study only achieved 21% power to detect an effect of this size and a sample of 1191 is necessary to achieve 80% power. Thus, overall, there is mixed support for the hypothesis that videos are more arousing than the equivalent text (H2).
Mixed-models were then fit where presentation medium predicted either cognitive or emotional empathy. Results reveal that presentation medium had no main effect on state emotional or cognitive empathy (H3). Therefore, participants experienced equivalent emotional and cognitive empathy when reading and viewing moral transgressions.
Wrongness Judgements (H4) Similarly, a mixed-model assessed the effect of presentation medium on the severity of wrongness judgements. The model revealed that there was a near-significant difference in wrongness judgements (t(201)= -1.72). This result suggests that there may be a very small effect of presentation medium (ß = -0.070) on wrongness judgements that this study had insufficient power to detect. Specifically, wrongness judgements of videos were marginally less harsh than when participants judged the equivalent moral content in text form, the contrary direction to H4. A post-hoc calculation for statistical power of a bivariate linear regression shows that the study only achieved 17% power to detect an effect of this size. A sample of 1578 would be needed to achieve 80% power to detect an effect of this size.
Chapter 7: Testing the Effect of Presentation Medium 205
To assess if arousal mediated any of the effect of condition on wrongness, a mediation model was fitted using the same method specified in Chapter 314. There was some support for the claim that presenting moral content via video (compared to text) lead to more arousal, and in turn, harsher wrongness judgements. There was a very small (=0.062), indirect effect of presentation medium on wrongness, such that there was more arousal in the video condition, which in turn had a small positive effect on wrongness. However, there was a larger direct effect in the opposite direction (=-0.12), whereby moral content presented via video lead to less harsh wrongness judgements.
There were also no significant direct effects of presentation medium on cognitive or emotional empathy (Table 38), and so there was no support for mediation of the effect of presentation medium on wrongness via empathy.
Table 39. Estimated Mediation Effects of Arousal on Medium and Wrongness
Indirect Effect of Medium on Wrongness 0.062 [0.012, 0.11], p=0.02
(Via Arousal)
Direct Effect of Medium on Wrongness -0.12 [-0.19, -0.05], p<0.001
Total Effect -0.06 [-0.15, 0.01], p=0.16
Proportion Mediated -0.92 [-8.9, 11.83], p=0.18
Note. Estimates were produced with 1,000 quasi-Bayesian Monte Carlo simulations
Humanness Perception (H6) Humanness ratings of the perpetrator and the victim across presentation medium were also assessed with a mixed effects model. Unexpectedly, participants rated both the perpetrator(s) and the victim(s) as less human when presented with a video than when they
14 The ‘causal mediation’ package in R (described here: Tingley, Yamamoto, Hirose, Keele, & Imai, 2014) estimates the direct effects of medium on wrongness, indirect effects via arousal, the total effect on wrongness, and the proportion of the total effect that is the indirect effect from the mixed effects models specified above. This process is similar to mediation estimated with linear regression (e.g., Hayes, 2017), except this method also allows the specification of a random effect.
Chapter 7: Testing the Effect of Presentation Medium 206
read a text vignette. This is the opposite directionality to the hypothesis (H6) that text would filter out non-verbal cues to humanness, obstructing perception of humanness in text when compared to video transgressions.
To further explore this unexpected result, I examined whether changes to humanness perception had downstream effects on moral judgement (the main dependent measure) by fitting mediation models. Results indicate that hyperhumanisation of the victim and perpetrator in the text condition lead to different effects on wrongness judgements. For perceptions of the perpetrator, text leads to more humanness that, in turn, results in less harsh judgements of wrongness. However, there is still a larger, direct effect of condition on wrongness in the opposite condition. For humanness perceptions of the victim, text leads to more humanness and this, in turn, results in more harsh judgements of wrongness. Further, this indirect effect completely accounts for the direct effect of medium on wrongness.
Table 40. Estimated Mediation Effects of Humanness on Medium and Wrongness
Indirect Effect of Medium on Wrongness -0.013 [-0.02, 0.00], p=0.002
(Via Humanness of the Victim)
Direct Effect of Medium on Wrongness -0.072 [-0.16, 0.02], p=0.12
Total Effect -0.086 [-0.18, 0.01], p=0.08
Proportion Mediated 0.13 [-0.74, 0.92], p=0. 08
Indirect Effect of Medium on Wrongness 0.047 [0.02, 0.08], p=0.004
(Via Humanness of the Perpetrator)
Direct Effect of Medium on Wrongness -0.10 [-0.19, -0.01], p=0.03
Total Effect -0.056 [-0.15, 0.04], p=0.26
Proportion Mediated -0.59 [-7.27, 7.76], p=0.27
Note. Estimates were produced with 1,000 quasi-Bayesian Monte Carlo simulations
Chapter 7: Testing the Effect of Presentation Medium 207
Blameworthiness and Intention (H7 and H8) Next, the effects of presentation medium on intention (H7) and blame (H8) were assessed. There were no significant effects of presentation medium on either intention nor blame, thus judgements of intention and blame were equivalent in video and text conditions.
Individual Differences Finally, I assessed the potential moderating effect of individual differences on the effects of presentation medium on moral judgement.
Cognitive Style (H9) Two mixed-models were fit with presentation medium and cognitive style as interacting fixed effects, one model predicting wrongness judgements and one predicting arousal. Contrary to H8, cognitive style did not moderate medium effects on moral judgement or arousal. Specifically, there were no main effects of cognitive style on either wrongness judgements or arousal, nor were there any interactions between cognitive style and condition when predicting either dependent measure.
Imaginative Ability (H10) Next, two mixed-models were fit with presentation medium and imaginative ability as fixed and interacting effects, predicting either cognitive or emotional empathy. H10 was not supported: there was no interaction between imaginative ability and condition on either cognitive or emotional empathy. However, there was a main effect of imaginative ability on empathy that was nearing significance (t(201) = 1.473). Similar to the effects of presentation medium on arousal and wrongness judgements, there may be a very small effect of imaginative ability on empathy that was not able to be detected here. Overall, individuals with low imaginative ability experienced equivalent levels of empathy in both the text and video conditions as individuals with high imaginative ability.
Chapter 7: Testing the Effect of Presentation Medium 208
Native English Speakers (H11) As above, mixed-models were fit predicting either cognitive or emotional empathy, native English speaking (dummy coded: 1 = native English speaker, 2 = non-native English speaker) and presentation medium were fixed (and interacting) effects. There was no main or interaction effect of native English speaking on either form of empathy. Thus, there were no differences in the cognitive or emotional empathy experienced by native English speakers compared to non-native speakers, contrary to the H11. Unfortunately, the test of this hypothesis was severely limited by the sample. Very few non-native English speakers were recruited (95% of the sample were native English speakers) and so this result is not a reliable test of this hypothesis.
Verbal IQ (H12) To test whether verbal ability moderated presentation medium effects on empathy, mixed-models were fit with presentation medium and verbal IQ as fixed (and interacting) effects, predicting either emotional or cognitive empathy. There was a marginally significant main effect of verbal IQ on emotional empathy but not cognitive empathy (t(201) = -1.96).
Again, the inconsistent significant tests are likely to relate to the very small effect size and a lack of statistical power in this study for very small effects. This result suggests that those with higher verbal IQ may experience slightly less emotional empathy. There were no significant interactions with condition and so there was no support for the hypothesized moderation (H12).
Chapter 7: Testing the Effect of Presentation Medium 209
Discussion
This study systematically explored the effect of presentation medium of moral judgement and emotion. Overall, there is evidence that presentation medium affects a number of morally relevant outcomes (i.e., wrongness, arousal, moral foundation categorisation and humanness), although the size of these effects was smaller than expected
(in most cases) and the directionality of some effects was contrary to expectations.
Specifically, when content was presented via text (compared to video) judgements were harsher, but (contrary to expectations) the content was less arousing, and there was greater consensus on why a moral transgression was wrong (i.e., moral foundation categorisation).
Contrary to expectations, when content was presented via text (compared to video) both perpetrators and victims were also judged as more human. There was no evidence to suggest that the presentation medium of the moral content affected perceptions of causality, blame, nor evoked empathy (cognitive or emotional). Finally, there was no evidence to suggest that presentation medium interacts with individual differences (i.e., verbal IQ, native language, trait empathy) to affect wrongness judgements or the experience of arousal and empathy.
Taken together, presentation medium affects aspects of moral psychology but in limited ways.
Medium Effects on Wrongness Judgements There was a very small, marginally significant effect of presentation medium on wrongness judgements, such that when participants judged the same moral content as less wrong when presented in video, compared to text. Contrary to H4, these results suggest that the additional social and contextual cues available in video (but not in text) can lead to less harsh wrongness judgements. This surprising result might suggest that the non-verbal cues present in the video actually convey extra information that can lessen wrongness judgements.
Chapter 7: Testing the Effect of Presentation Medium 210
There are multiple possible interpretations as to how this additional information lessens wrongness judgements.
One possibility is that presentation medium affects the level at which the moral transgression is mentally construed and, in turn, moral judgement. According to construal- level theory, the same objects and events may be represented (construed) at multiple levels of abstraction (Liberman & Trope, 2008; Trope & Liberman, 2010). High-level construals are abstract in nature and have fewer concrete contextual details (e.g., being a good person). In comparison, low-level construals include concrete features of the situation and attributes
(e.g., donating money). Amit, Algom, and Trope (2009) show that text as a presentation medium facilitates higher level, abstract construals, while images facilitate lower construals.
For example, during a sorting task, participants sorted items into larger and more abstract categories when the items were presented in text, compared to when those same items were presented as images. Therefore, by extension, the text stimuli in this study may have been construed at lower levels and the video at high levels. According to Eyal et al. (2008), when an event is also construed at a high level (i.e., when presented in text) people are more likely draw on their moral principles (as moral values are also abstract), compared to when the same event is construed at a low level. As a result, events construed at a high level (i.e., moral transgressions presented in text) are judged more morally wrong than those construed at a low level (i.e., moral transgressions presented in video) (Eyal et al., 2008). Therefore, in the current study, text stimuli may be construed more abstractly, making moral values more salient and thus resulting in more harsh judgements, than when judging the equivalent content as a video.
Alternatively, the additional social and contextual cues in the video condition may communicate mitigating circumstances that reduce the harshness of the judgement. For example, the text vignette “A basketball player yells at his coaches during a game” was rated,
Chapter 7: Testing the Effect of Presentation Medium 211
on average, as 2.22 for wrongness, while the video was rated only 1.78. Non-verbal cues present in the video (e.g., calm facial expression and body language) may imply that there was no harm inflicted by the player’s actions; thereby reducing perceptions of wrongness in the video condition. In the text condition, these non-verbal cues of the coach’s emotional experience are filtered out and thus the participant may assume that the coach is negatively affected (e.g., embarrassed), leading them to judge the transgression as more wrong than the same transgression in video.
Although an effort was made to equate the content of the text and video conditions
(i.e., the transcription process), including contextual information, there may be less salience of contextual information conveyed by text compared to video. According to construal level theory, when events are construed at a high level (i.e., text) the relevance of moderating contextual information is discounted (Eyal et al., 2008). Therefore, even if the text and videos were equivalent in contextual information, participants may have attended to the contextual
(and mitigating) information in the video condition more than in the text condition. This mitigating information, in turn, may have lead to less harsh judgements of wrongness.
One further possibility is that medium effects on wrongness are underpinned by differences in the extent to which people attend to different kinds of moral considerations.
Namely, Amit and Greene (2012) argue that images (compared to text) make deontological values more salient and utilitarian values less salient. In particular, the authors argue that people mentalise the harm caused by a moral transgression more when presented with visual stimuli compared to text. Therefore, in this study, the effect of presentation medium on wrongness may be driven by increased mentalisation of harm when the moral event is presented via video compared to text. Future research should examine whether mentalisation of harm varies by presentation medium and if the type of moral judgement differs by
Chapter 7: Testing the Effect of Presentation Medium 212
presentation medium. That is, whether there is more deontological judgement (compared to utilitarian) when the moral event is presented via video/image compared to text.
Relatedly, this study finds preliminary evidence that medium not only affects the severity of the moral judgement but also the type of judgement. Specifically, there was less consensus on the categorisation of liberty and purity violations into moral foundations when the moral transgression was presented in video compared to text. Consistent with theorising, the presence of individuating social or contextual cues in the video condition (assumedly) leads to more variation in the perception of the moral event and consequent categorisation into moral foundations. When the moral event was represented in text only, participants relied on widely held stereotypes to inform their judgement and so there was greater agreement in their judgement of the liberty and purity violations.
Overall, the result of medium on wrongness judgements has implications for the field of moral psychology. If the richness of the presentation medium affects the harshness of wrongness judgements, this suggests that the moral psychology literature may be misrepresenting wrongness effects. Specifically, if text stimuli facilitate (slightly) harsher judgements of wrongness and more than 90% of stimuli in moral psychology studies use text stimuli, then the vast majority of studies may be overestimating the size of wrongness effects.
If video stimuli more closely represent ‘real world’ moral events, then we might expect that some findings related to wrongness judgements may not replicate with more ecologically valid stimuli.
Medium Effects on Moral Emotions There was also a marginally significant, small effect of medium on arousal, such that video was more arousing than text, but there was no effect of presentation medium on (state) cognitive or emotional empathy. Overall, these results suggest that the richness of the medium have a small effect on affect but not on empathy. This contrasts with emotion
Chapter 7: Testing the Effect of Presentation Medium 213
induction literature that has shown that medium affects both the type and intensity of state emotion. As noted previously, non-text emotion induction stimuli (e.g., video) tend to elicit more intense emotional responses than text stimuli for discrete emotions such as anger and sadness (Ferrer et al., 2015; Gerrards‐Hesse et al., 1994; Westerman et al., 2014). This study assessed undifferentiated arousal only, and so one possibility is that presentation medium affects only certain discrete emotions (e.g., anger and sadness). If this was the case, the lack of granularity in the measurement of arousal may obscure any effect of medium on discrete emotion(s). Unfortunately, this study was not able to assess discrete emotions as the number of questions per stimulus would become excessively long and cause participant fatigue.
Future research should consider whether the small effect of medium on arousal is driven by changes to discrete emotions.
This effect of presentation medium on arousal only had a very small, downstream effect on wrongness. Specifically, when content was presented via video, there was slightly increased arousal which led to a slight increase in the harshness of wrongness judgements.
However, because the effect of presentation medium on arousal was so small, the indirect effect of medium on wrongness (via arousal) was much smaller than the negative and direct effect of presentation medium on wrongness. Taking together these results with the above results of presentation medium on wrongness, there appears to be multiple, competing effects of medium on wrongness. There is a larger, direct effect of presentation medium on wrongness, and a small indirect effect via arousal, where the richness of the video medium led to slightly more arousal than the equivalent moral content presented in text, this increased arousal lead to slightly harsher judgements.
Given that there was an effect of medium on arousal and wrongness, it’s surprising that there was no effect of presentation medium on empathy, cognitive or emotional. This is unexpected given that there are frequent links in the moral psychology literature between
Chapter 7: Testing the Effect of Presentation Medium 214
arousal, empathy, and moral judgement (Greene et al., 2001; Rozin et al., 1999; Schnall et al.,
2008). These results might imply that while non-verbal cues that typically convey emotion and facilitate empathy (e.g., voice tone or facial expression) are filtered out in text compared to video, the more explicit nature of text compensates for the absence of these cues. For example, the text vignette “A woman is forced to wear skimpy clothing chosen by her boyfriend despite feeling uncomfortable” explicitly states the feelings of the moral patient
(the woman), thereby facilitating emotion recognition and, in turn, empathy. Overall, these results suggest that text (compared to video) does not impede the experience of emotional or cognitive empathy. Importantly, this implies that previous research on empathy in the moral domain may not have misspecified effects as a result of the over-reliance on text stimuli.
It should be noted, however, that this study used an ad hoc measure of state empathy, which may limit the validity of these results. The results of this study can’t verify that this ad hoc measure was a valid and reliable measure of empathy and this potential lack of validity could contribute to the absence of an effect. Participants also completed a state measure of empathy (not reported here), which was only weakly correlated with the theoretically associated measures of trait empathy (remotional = 0.23 [0.19, 0.28], rcognitive = 0.09 [0.04,0.14]).
This suggests that this state measure has poor convergent validity and thus may not be an accurate measure of state empathy. Further efforts are required to develop a robust and brief measure of state cognitive and emotional empathy, as this is currently lacking in the literature. Once a measure has been developed, additional research should confirm this negative finding.
Medium Effects on Person Perception
Humanness Perception Contrary to hypotheses, there was greater humanness attributed to both the victim and the perpetrator in the text condition than in the video condition. While inconsistent with
Chapter 7: Testing the Effect of Presentation Medium 215
hypotheses, these results are partially consistent with theorising from Stream 1. Specifically, it may be that when non-verbal cues important to person perception are filtered out in text stimuli, participants compensate for the absence of information by exaggerating available cues to humanness. Thus, resulting in exaggerated perceptions of humanness in text compared to video. Following this logic, however, we would expect that perceptions of humanness would become polarised, that is both positively (i.e., hyperhumanisation) and negatively exaggerated (i.e., dehumanisation) in text compared to video and not just positively shifted. Specifically, we would expect that when the cues in text convey very low humanness, such as cues associated with the moral perpetrators, the relative importance of those cues will be exaggerated and lead to more extreme dehumanisation compared to those same cues presented via video. When those cues in text convey high humanness, such as cues associated with moral victims, those cues should be exaggerated and result in more extreme hyperhumanisation compared to the same cues in video. Instead, the results show a positive shift in perceptions of humanness for both moral perpetrators and moral victims. Overall, it’s unclear what is driving this unusual effect.
Exploratory analyses revealed that hyperhumanisation mediates the effect of condition on judgements of wrongness. Specifically, for perceptions of the victim’s humanness, hyperhumanisation in the text condition led to more harsh judgements of wrongness. This result suggests changes to perception of the victim as a result of the presentation medium account for the effect of medium on wrongness. Assumedly, the more human the participant perceives the victim, the more that the victim is perceived as a moral patient that requires protection from moral transgressions (Bastian et al., 2011). Thus, hyperhumanness perceptions of the victim lead to more harsh judgements of wrongness.
There is a different pattern of effects for humanness perceptions of the perpetrator.
Hyperhumanisation of the perpetrator in the text condition led to a less harsh judgement of
Chapter 7: Testing the Effect of Presentation Medium 216
wrongness. This result conflicts with the previously established positive relationship between the perception of HU traits and attributions of culpability (Bastian et al., 2011). According to the findings by (Bastian et al., 2011), we would expect that more the perpetrator is perceived to have HU, the greater the attributions of moral agency (i.e., capacity to be responsible for immoral behaviour), and thus the more wrong the action. One potential explanation is that the direction of effects could be reversed. That is, condition may lead to more harsh judgements of wrongness, that in turn, leads participants to dehumanise a perpetrator and humanise a victim. Given that these analyses are exploratory, rather than a-priori, future research is required to confirm the effects and examine directionality.
Blame and Intention Although presentation medium affected perceptions of humanness, surprisingly, medium did not have a significant effect on qualities tied to humanness: intention and blame.
Specifically, there were no significant effects of presentation medium on judgements of intention or blame. These results suggest that the medium with which information about cause and intent are conveyed does not affect judgements. Importantly, these results suggest that the overreliance on text stimuli in the field of moral psychology has not impacted cause or intent measurements.
Individual Difference Effects on Presentation Medium Across all measured individual differences, there was no support for any hypothesized interactions between individual differences and presentation medium on moral judgment.
Different cognitive styles (visual versus verbal), verbal ability, imaginative ability, nor native language (English or not) interacted with presentation medium to affect wrongness judgements or empathy/arousal. Overall, this study found no evidence that presentation medium effects on moral judgement are impacted by any of the measured individual differences.
Chapter 7: Testing the Effect of Presentation Medium 217
In part, the absence of significant effects may be the result of homogeneity in the sample. That is, the sample had limited variability on many of the individual difference measures. For example, the distribution of cognitive styles was negatively skewed, such that sample was disproportionately visualisers (64% scored > 4 on the scale) and far fewer verbalisers (14% scored <4 on the scale, refer to Table 37). As such, this study captured very limited data on how verbalisers respond to moral content presented in text versus video. The lack of variability on the cognitive style measure inflates the potential for a type II error, which may contribute to the absence of an effect.
A similar limitation applies to the assessment of the effects of (1) English as a native language and (2) verbal ability on presentation medium and moral judgment. First, there were so few participants that did not have English as their first language (2% of the sample) that this study was not able to robustly test H11. Second, there was a ceiling effect for the verbal ability scale such that there were very few participants on the lower half of the verbal ability scale (M=34.67 where the max = 40, skew = -1.49, see Table 37). This study may have failed to sample any participants with below average verbal ability. Indeed, demographic research on Mechanical Turk workers reveals that the population tends to higher than average education (Levay, Freese, & Druckman, 2016). As a result, there is limited data to quantify whether those low on verbal ability or those that do not natively speak English differed in their ability simulate the moral transgression using their imagination and experience of empathy, compared to those high on verbal ability or native English speakers. Future research with a more balanced sample is required to robustly test these hypotheses.
Overall, there is no evidence to suggest that presentation medium affects mentalisation. That is, those high on imaginative ability in the text condition experienced equivalent levels of empathy to those low on imaginative ability. This implies that those high
Chapter 7: Testing the Effect of Presentation Medium 218
on imaginative ability did not compensate for the absence of non-verbal cues in the text condition and ‘fill in the gaps’ by mentally simulating the moral transgression.
Although the results were limited by sampling biases, results for the verbal ability and cognitive style measures can be interpreted in a similar way. That is, those who are low in verbal ability or experienced the same level of empathy in the text condition as those who are high on verbal ability and made equivalent judgments of wrongness. Likewise, visualisers experienced the same level of arousal and made equivalent judgements of wrongness as verbalisers, in the text condition. This suggests that both verbalisers and visualisers/those low on verbal IQ and high on verbal IQ equal attend to and comprehended the content, irrespective of the medium. This negative result could relate to demand characteristics of the study, that is, participants may have been more conscientious and attentive to the stimuli because they were paid to participate in a study – obstructing differences between presentation media.
Limitations and Future Directions Most obviously, a central limitation of this study was the sample size. The effects identified by this study were unexpectedly small and, as a result, several effects only reached marginal significance. One possibility is that interactions, mediations, and other main effects were not detected due to insufficient statistical power. While one could argue that the very small effects that (potentially) were not detected in this study do not have practical significance for the field, the downstream consequences of over-relying on text stimuli on the moral psychology field still aren’t clear. Thus, very small effects may have practical significance for the field. Future research should replicate this study with a larger sample size.
Second, this study used single item (state) measures, which has inherent limitations.
In this case, the use of single item measures was justified by precedent (i.e., previous moral
Chapter 7: Testing the Effect of Presentation Medium 219
studies typically use single item measures) and given that the aim of the study was to explore the effect of medium on moral judgement, affect, and perception outcomes. This study has served to identify potential outcomes that are impacted by medium, specifically, arousal, wrongness judgements, moral foundation categorisation, and humanness perceptions. Future research should further investigate these outcomes by utilising multi-item measures to improve accuracy and reliability.
In addition, a fertile area for future research is the role of discrete emotions on medium effects. Study 4 provided some initial evidence that discrete emotions may be impacted by medium (see Figure 21), however this was not within scope for Study 5. Given that arousal appears to differ when presented with text stimuli compared to video, this result may extend to discrete emotions. One possibility is that medium effects on discrete emotions could account for the effect of medium on wrongness judgements and differences in moral foundation categorisations across media.
Conclusion
This study presents an exploration of how the medium by which a moral stimulus is presented affects judgement. Results show that presentation medium affects a number of morally relevant outcomes: wrongness, arousal, moral foundation categorisation and humanness. The study reveals evidence that the non-verbal social and contextual cues available in the video condition, but filtered out in the text condition, conveyed mitigating circumstances that undermined wrongness judgments that led to less consensus when categorising the transgression into a moral foundation. The richness of the video condition also led to slightly more arousal, which had a small effect on the wrongness of the judgement.
Unexpectedly, when the moral stimulus was presented in text lead to hyperhumanness perceptions or positively exaggerated perceptions of humanness for both the victim and the perpetrator. Taken together, these results show that presentation medium of stimuli is an
Chapter 7: Testing the Effect of Presentation Medium 220
important factor to consider when designing a moral psychology study; relying on text stimuli alone can result in a pattern of effects for wrongness, moral foundation categorisation, arousal and humanness perceptions that do not necessarily generalise to more ecologically valid media. However, these results suggest that using text stimuli to study attributions of blame or intent and feelings of empathy may not result in misspecification of effects. Chapter 8: General Discussion 221
Chapter 8: General Discussion
Across two streams, this thesis has illustrated the importance of the medium by which we communicate or present a stimulus in shaping morally relevant perception, judgment and behaviour.
The first stream of this thesis examined the role of medium in moral outcomes in the context of CMC and utilised the CFO as a theoretical framework to understand behaviour in the context of CMC. Overall, there was limited support for this theoretical approach; however, both studies in this stream identify ways in which medium affects positive and negative moral behaviours and moral perception. Study 1 demonstrated that listening to the voice of someone who disagrees with your political opinion can lead to more aggression and paradoxically, humanization, compared to reading an equivalent text passage. Study 2 shows that, under certain circumstances, reading a dating profile can lead to more self-disclosure compared to watching an equivalent video. Study 2 also found that in the text condition some aspect of person perception was exaggerated.
Given the limited support for CFO or other major CMC theories, as well as unexpected and paradoxical effects of medium on moral outcomes, the second stream of this thesis took a more basic approach to explore the effect of presentation medium on moral outcomes. First, a systematic review of the moral psychology literature revealed that the vast majority of research in the field relied on text stimuli. The limited number of comparisons between studies that have used text and non-text presentation media provided some additional evidence that presentation medium affects moral outcomes. Importantly, however, the results of the review highlighted a risk that the field of moral psychology may have systematically biased empirical outcomes and theorising by relying on text stimuli near- exclusively. After developing the moral video stimulus set and the matching text stimuli
(Study 4), Study 5 found that judging moral transgressions in a cue-poor medium (i.e., text) Chapter 8: General Discussion 222 results in some aspects of moral judgement and affect (i.e., wrongness, moral foundation categorisations, arousal, and humanness) did not generalise to a more ecologically valid medium (i.e., video).
Implications for Moral Perception
Across both streams of the thesis, there is evidence that presentation medium affects perceptions of humanness – although the mechanism by which these effects arise remain unclear. Study 1 found that text (compared to voice) can lead dehumanization, thereby leading to reduced human presence (the quality of being a human moral agent or patient) and dehumanisation of a target. The second stream of this thesis also found that humaneness perceptions are impacted by medium, however, in the opposite direction. Contrary to expectations, Study 5 found that text (compared to video) lead to exaggerated perceptions of humanness for both a moral perpetrator and a moral victim. Presumably, in the text condition participants may have compensated for the filtering out of non-verbal cues relevant to person perception by exaggerating by perceptions of humanness.
Although communication medium can, under some circumstances, impact humanness perceptions in both positive and negative directions, it’s not clear whether these changes have meaningful downstream consequences. In Study 1, the increased dehumanisation in the text condition had only a very small (positive) effect on aggression, which was masked by a larger direct effect of medium in the opposite directionality. However, in Study 5, exploratory analyses found that hyperhumanisation of the perpetrator (in the text condition) led to less harsh judgments of wrongness, while hyperhumanisation of the victim (in the text condition) led to more harsh judgements of wrongness. Yet, the directionality of this results is unclear. It’s equally possible that medium impacts wrongness, which in turn, changes humanness perception. Thus, across four studies of this thesis there is limited evidence to show that medium driven changes to humanness perception has meaningful consequences for Chapter 8: General Discussion 223 behaviour or judgement. Indeed, while Schroeder and Epley (2016) demonstrate the dehumanising effect of text on perceptions, the authors do not present any research to illustrate downstream consequences on perception or behaviour.
The Role of Expectations
A promising avenue for future research revealed by this thesis is the role of expectations on medium effects. Stream one of this thesis found preliminary evidence that pre-interactional expectations can impact medium effects. For example, in Study 1, one interpretation of the results is that when an actor holds negative expectations for how their partner will behave in a given situation, richer media will exacerbate their negativity and lead to aggressive behaviour. Study 2 reveals that pre-interactional expectations were one of the few significant predictors of self-disclosing behaviour.
Although not explored in Stream 2, participant's expectations about medium effects could contribute to medium effects on moral judgement. That is, people might hold expectations about how reduced cues in text stimuli (versus viewing a video or a FtF event) affect their impression formation and moral judgement. These expectations might, in turn, lead them to change their judgement to account for the presumed effect of reduced cues. For example, if an individual expects that text stimuli tend to filter out mitigating circumstances which would have otherwise reduced the harshness of their judgement, the individual may judge the text stimulus less harshly to compensate. Consistent with this possibility, individual’s expectations about the effect of reduced cues in text has been shown to affect other morally relevant outcomes: disinhibition and self-disclosure (Schouten et al., 2007).
That is, the extent to which an actor expected that the reduced cues in text allow them to control their self-presentation positively predicted disinhibition and online self-disclosure.
Although the authors did not assess moral judgement, these results illustrate that people generally held beliefs about how medium can affect behaviour and perception which, in turn, Chapter 8: General Discussion 224 had consequences on morally relevant outcomes. Therefore, we might expect that people also hold beliefs about medium effects on moral judgement and associated processes.
The Social Implications of Cue-Poor Media
Importantly, the results of this thesis suggest that the filtering out of non-verbal cues in text can have positive consequences for perception and behaviour. Stream 1 of this thesis found that text (compared to a cue-rich medium) exerted a positive influence on moral behaviour: text lead to less aggression (Study 1) and more self-disclosure (Study 2), compared to either image or video. In Stream 2 of this thesis, the text condition led to harsher moral judgements and positively exaggerated perceptions of humanness compared to the cue-rich medium (video). As previously noted, the abstract nature of text may facilitate a higher construal mindset which, in turn, makes moral principles more salient than content presented in cue-rich media. Taking together these results, the increased salience of moral values in text may contribute to reduced anti-social behaviour (Study 1), increased pro-social behaviours (Study 2), and hasher wrongness judgements (Study 5). This theorising stands in contrast to much of the rhetoric in the CMC literature and general population, which points to anti-social consequences for text-based communication compared to FtF interaction
(Hmielowski et al., 2014; Jane, 2015; Lapidot-Lefler & Barak, 2012; Sproull & Kiesler,
1991; Suler, 2004). These results instead highlight that cue-poor media, such as CMC or stimuli presented via text, can facilitate intimacy related behaviour like self-disclosure, reduce aggression in specific contexts, or harsher moral judgements.
The idea that cue-poor media can have positive social consequences echoes the theorising of the hyperpersonalism/social information processing theories (Walther, 1996).
According to these theories, cue-poor communication media, such as text, don't obstruct the communication of relational information because people instead compensate for the absence of non-verbal cues through their use of verbal cues. For example, non-verbal cues typically Chapter 8: General Discussion 225 convey emotion in FtF interaction (e.g., facial expression), Walther (1996) argues that people instead explicitly state their emotions (i.e., verbally) in CMC. Further, social information processing theory argues that communicating via text allows communicators to manage their impression more positively than interacting via cue-rich media. These theories imply that text-based communication can have positive consequences for relationship development and interaction compared to richer communication media, such as video or FtF interaction.
Although reciprocal interaction was not the focus of this thesis, the results of this thesis also converge on the idea that presenting a target via text (compared to voice or video) often has equivalent and even more positive impacts on behaviour and impression formation.
Implications of the Over-Reliance on Text
An implication from this is that text, in many cases, does not obstruct the communication of morally relevant social-contextual cues. Specifically, both text and video or images were equally able to convey information about blameworthiness (Study 5), cause
(Study 5), moral status (Study 1 and 5), trust (Study 2), attractiveness (Study 2), and intention
(Study 5). Overall, I find evidence that text stimuli and video/voice stimuli are equally suitable media for moral psychology researchers interested in studying the above constructs.
Moreover, these results imply that previous research on these topics is unlikely to have been mischaracterised by the overreliance of text stimuli in previous moral psychology studies.
However, the overreliance of text in moral psychology may have affected the measurement and theorising about judgements of wrongness, moral foundations, humanness, and the role of arousal. The pattern of effects for these variables is different for text versus video and/or voice. As a result, it’s not clear that research on these variables that uses only text stimuli is generalizable to stimuli presented in other media, or, more importantly, real moral events. Although the effects, overall, are small and thus the practical impact on the field and the measurement of these variables is uncertain. Chapter 8: General Discussion 226
Future Directions
All studies in this thesis address the role of medium on psychological outcomes, the first part of this thesis (Study 1 and Study 2) address a very constrained form of text-based communication. In comparison, the second set of studies assess the role of medium more broadly and in a more basic manner. As Stream 1 and Stream 2 of this thesis have markedly different manipulations of medium, it should be noted that conclusions drawn across all studies are limited.
This thesis served as an initial exploration into the role of medium on aspects of moral judgement, perception, and behaviour. Future researchers should utilise the stimulus set presented here to (1) continue to clarify medium effects on moral outcomes, including exploring potential mechanisms; (2) and to move beyond the limits of text-based stimuli and increase the ecological validity of moral psychology research.
In particular, this thesis has identified evidence that medium impacts aspects of moral judgement, affect, and perception (e.g., humanness perception). However, it’s unclear what the downstream consequences of these effects are, if any. For example, medium impacted humanness in Study 1 and Study 5, but in both studies, I was unable to identify any consequences to moral judgement or moral behaviour. Future research could clarify the downstream consequences for the field of moral psychology.
Conclusion
Across five studies, this thesis has explored the role of presentation medium in moral psychology. The exact mechanism by which medium affects moral outcomes remains unclear, that is, there was limited support for CMC theories in accounting for medium effects of a positive or negative moral behaviour. However, Stream 2 of this thesis identifies that the medium used to present a stimulus can impact aspects of moral judgement. Contrary to dominant rhetoric in the CMC literature and general population, text as a presentation or Chapter 8: General Discussion 227 communication medium does not necessarily lead to more negative outcomes. Instead, I find evidence that text can have positive implications for moral behaviour: facilitating more moral judgment, less antisocial behaviour, and more positive moral behaviours. Overall, this thesis has identified the importance of presentation medium in moral psychology and identified a fertile ground for future research. Future research in the cross section of medium and moral psychology will have impact on both the understanding of moral behaviours in an increasingly technology saturated world, but also reveals the importance for accurate measurement and theorising in moral psychology research Chapter 8: General Discussion 228
References
Abrams, D., & Hogg, M. A. (2006). Social identifications: A social psychology of intergroup relations and group processes: Routledge. Abrams, Z. I. (2003). Flaming in CMC: Prometheus' fire or Inferno's? CALICO Journal, 245- 260. Adolphs, R. (2002). Recognizing emotion from facial expressions: psychological and neurological mechanisms. Behavioral and Cognitive Neuroscience Reviews, 1(1), 21- 62. Adolphs, R., Nummenmaa, L., Todorov, A., & Haxby, J. V. (2016). Data-driven approaches in the investigation of social perception. Phil. Trans. R. Soc. B, 371(1693), 20150367. Agerström, J., & Björklund, F. (2009). Moral concerns are greater for temporally distant events and are moderated by value strength. Social Cognition, 27(2), 261-282. Aiken, M., & Waller, B. (2000). Flaming among first-time group support system users. Information & Management, 37(2), 95-100. Alonzo, M., & Aiken, M. (2004). Flaming in electronic communication. Decision Support Systems, 36(3), 205-213. Amit, E., Algom, D., & Trope, Y. (2009). Distance-dependent processing of pictures and words. Journal of Experimental Psychology: General, 138(3), 400. Amit, E., & Greene, J. D. (2012). You see, the ends don’t justify the means: Visual imagery and moral judgment. Psychological Science, 23(8), 861-868. Antheunis, M. L., Valkenburg, P. M., & Peter, J. (2007). Computer-mediated communication and interpersonal attraction: An experimental test of two explanatory hypotheses. CyberPsychology & Behavior, 10(6), 831-836. Ayduk, Ö., & Kross, E. (2010). From a distance: implications of spontaneous self-distancing for adaptive self-reflection. Journal of Personality and Social Psychology, 98(5), 809. Baayen, R. H., Davidson, D. J., & Bates, D. M. (2008). Mixed-effects modeling with crossed random effects for subjects and items. Journal of Memory and anguage, 59(4), 390- 412. Bahnemann, M., Dziobek, I., Prehn, K., Wolf, I., & Heekeren, H. R. (2009). Sociotopy in the temporoparietal cortex: common versus distinct processes. Social cognitive and affective neuroscience, 5(1), 48-58. Bain, P., Park, J., Kwok, C., & Haslam, N. (2009). Attributing human uniqueness and human nature to cultural groups: Distinct forms of subtle dehumanization. Group Processes & Intergroup Relations, 12(6), 789-805. Bandura, A. (1999). Moral disengagement in the perpetration of inhumanities. Personality and Social Psychology Review, 3(3), 193-209. Barak, A., & Gluck-Ofri, O. (2007). Degree and reciprocity of self-disclosure in online forums. CyberPsychology & Behavior, 10(3), 407-417. Barberá, P., Jost, J. T., Nagler, J., Tucker, J. A., & Bonneau, R. (2015). Tweeting from left to right: Is online political communication more than an echo chamber? Psychological Science, 26(10), 1531-1542. Bastian, B., Laham, S. M., Wilson, S., Haslam, N., & Koval, P. (2011). Blaming, praising, and protecting our humanity: The implications of everyday dehumanization for judgments of moral status. British Journal of Social Psychology, 50(3), 469-483. Bates, D., Mächler, M., Bolker, B., & Walker, S. (2014). Fitting linear mixed-effects models using lme4. arXiv preprint arXiv:1406.5823.
Chapter 8: General Discussion 229
Batson, C. D., Kennedy, C. L., Nord, L. A., Stocks, E., Fleming, D. Y. A., Marzette, C. M., Lishner, D. A., Hayes, R. E., Kolchinsky, L. M., & Zerger, T. (2007). Anger at unfairness: Is it moral outrage? European Journal of Social Psychology, 37(6), 1272- 1285. Berg, J., Dickhaut, J., & McCabe, K. (1995). Trust, reciprocity, and social history. Games and Economic Behavior, 10(1), 122-142. Berger, J., & Milkman, K. L. (2012). What makes online content viral? Journal of Marketing Research, 49(2), 192-205. Blair, R. J. R. (1995). A cognitive developmental approach to morality: Investigating the psychopath. Cognition, 57(1), 1-29. Blair, R. J. R. (2005). Applying a cognitive neuroscience perspective to the disorder of psychopathy. Development and Psychopathology, 17(3), 865-891. Blanken, I., van de Ven, N., & Zeelenberg, M. (2015). A meta-analytic review of moral licensing. Personality and Social Psychology Bulletin, 41(4), 540-558. Boccia, M., Dacquino, C., Piccardi, L., Cordellieri, P., Guariglia, C., Ferlazzo, F., . . . Giannini, A. M. (2017). Neural foundation of human moral reasoning: an ALE meta- analysis about the role of personal perspective. Brain Imaging and Behavior, 11(1), 278-292. Brambilla, M., & Riva, P. (2017). Predicting pleasure at others’ misfortune: Morality trumps sociability and competence in driving deservingness and schadenfreude. Motivation and Emotion, 41(2), 243-253. Brosig, J., Weimann, J., & Ockenfels, A. (2003). The effect of communication media on cooperation. German Economic Review, 4(2), 217-241. Brown-Iannuzzi, J. L., McKee, S., & Gervais, W. M. (2018). Atheist horns and religious halos: Mental representations of atheists and theists. Journal of Experimental Psychology: General, 147(2), 292. Brunet, P. M., & Schmidt, L. A. (2008). Are shy adults really bolder online? It depends on the context. CyberPsychology & Behavior, 11(6), 707-709. Buon, M., Jacob, P., Loissel, E., & Dupoux, E. (2013). A non-mentalistic cause-based heuristic in human social evaluations. Cognition, 126(2), 149-155. Burgoon, J. K., Guerrero, L. K., & Manusov, V. (2011). Nonverbal signals. The SAGE handbook of interpersonal communication. London: SAGE. Burgoon, J. K., & Le Poire, B. A. (1999). Nonverbal cues and interpersonal judgments: Participant and observer perceptions of intimacy, dominance, composure, and formality. Communications Monographs, 66(2), 105-124. Cannon, P. R., Schnall, S., & White, M. (2011). Transgressions and expressions Affective facial muscle activity predicts moral judgments. Social Psychological and Personality Science, 2(3), 325-331. Caruso, E. M., & Gino, F. (2011). Blind ethics: Closing one’s eyes polarizes moral judgments and discourages dishonest behavior. Cognition, 118(2), 280-285. Castellá, V. O., Abad, A. Z., Alonso, F. P., & Silla, J. P. (2000). The influence of familiarity among group members, group atmosphere and assertiveness on uninhibited behavior through three different communication media. Computers in Human Behavior, 16(2), 141-159. Chaiken, S., & Eagly, A. H. (1976). Communication modality as a determinant of message persuasiveness and message comprehensibility. Journal of Personality and Social Psychology, 34(4), 605. Chang, C.-W., & Heo, J. (2014). Visiting theories that predict college students’ self- disclosure on Facebook. Computers in Human Behavior, 30, 79-86. Chapter 8: General Discussion 230
Chapman, H. A., & Anderson, A. K. (2013). Things rank and gross in nature: a review and synthesis of moral disgust. Psychological Bulletin, 139(2), 300. Child, D. (2006). The essentials of factor analysis: A&C Black. Cho, D., & Kwon, K. H. (2015). The impacts of identity verification and disclosure of social cues on flaming in online user comments. Computers in Human Behavior, 51, 363- 372. Chui, R. (2014). A Multi-faceted Approach to Anonymity Online: Examining the Relations between Anonymity and Antisocial Behaviour. Journal For Virtual Worlds Research, 7(2). Cikara, M., Farnsworth, R. A., Harris, L. T., & Fiske, S. T. (2010). On the wrong side of the trolley track: Neural correlates of relative social valuation. Social cognitive and Affective Neuroscience, 5(4), 404–413. Clifford, S., Iyengar, V., Cabeza, R., & Sinnott-Armstrong, W. (2015). Moral foundations vignettes: A standardized stimulus database of scenarios based on moral foundations theory. Behavior Research Methods, 47(4), 1178-1198. Coleman, L., Paternite, C., & Sherman, R. (1999). A reexamination of deindividuation in synchronous computer-mediated communication. Computers in Human Behavior, 15(1), 51-65. Collins, N. L., & Miller, L. C. (1994). Self-disclosure and liking: a meta-analytic review. Psychological Bulletin, 116(3), 457. Condon, P., & DeSteno, D. (2011). Compassion for one reduces punishment for another. Journal of Experimental Social Psychology, 47(3), 698-701. Corcoran, K. J. (1988). The relationship of interpersonal trust to self-disclosure when confidentiality is assured. The Journal of Psychology, 122(2), 193-195. Craig, E., Igiel, M., Wright, K., Cunningham, C., & Ploeger, N. (2007). Will you be my friend”: Computer-mediated relational development on Facebook.com. Paper presented at the annual meeting of International Communication Association, San Francisco, CA. Craig, E., & Wright, K. B. (2012). Computer-mediated relational development and maintenance on Facebook. Communication Research Reports, 29(2), 119-129. Culnan, M. J., & Markus, M. L. (1987). Information technologies. Cushman, F., Young, L., & Hauser, M. (2006). The role of conscious reasoning and intuition in moral judgment: Testing three principles of harm. Psychological Science, 17(12), 1082-1089. De Greef, P., & Ijsselsteijn, W. A. (2001). Social presence in a home tele-application. CyberPsychology & Behavior, 4(2), 307-315. Decety, J., & Jackson, P. L. (2004). The functional architecture of human empathy. Behavioral and Cognitive Neuroscience Reviews, 3(2), 71-100. Decety, J., & Porges, E. C. (2011). Imagining being the agent of actions that carry different moral consequences: an fMRI study. Neuropsychologia, 49(11), 2994-3001. Derks, D., Fischer, A. H., & Bos, A. E. (2008). The role of emotion in computer-mediated communication: A review. Computers in Human Behavior, 24(3), 766-785. Diener, E. (1977). Deindividuation: Causes and consequences. Social Behavior & Personality: An International Journal, 5(1). Dogruel, L., Joeckel, S., & Bowman, N. D. (2013). Elderly people and morality in virtual worlds: A cross-cultural analysis of elderly people’s morality in interactive media. New Media & Society, 15(2), 276-293. Drolet, M., Schubotz, R. I., & Fischer, J. (2012). Authenticity affects the recognition of emotions in speech: behavioral and fMRI evidence. Cognitive, Affective, & Behavioral Neuroscience, 12(1), 140-150. Chapter 8: General Discussion 231
Drouin, M., Miller, D., Wehle, S. M., & Hernandez, E. (2016). Why do people lie online?“Because everyone lies on the internet”. Computers in Human Behavior, 64, 134-142. Dubrovsky, V. J., Kiesler, S., & Sethna, B. N. (1991). The equalization phenomenon: Status effects in computer-mediated and face-to-face decision-making groups. Human- Computer Interaction, 6(2), 119-146. Ekman, P. (2004). Emotional and conversational nonverbal signals. In Language, knowledge, and representation (pp. 39-50): Springer. Ellard, K. K., Farchione, T. J., & Barlow, D. H. (2012). Relative effectiveness of emotion induction procedures and the role of personal relevance in a clinical sample: a comparison of film, images, and music. Journal of Psychopathology and Behavioral Assessment, 34(2), 232-243. Ellison, N., Heino, R., & Gibbs, J. (2006). Managing impressions online: Self-presentation processes in the online dating environment. Journal of Computer-Mediated Communication, 11(2), 415-441. Epley, N., & Kruger, J. (2005). When what you type isn’t what they read: The perseverance of stereotypes and expectancies over e-mail. Journal of Experimental Social Psychology, 41(4), 414-422. Escobedo, J. R., & Adolphs, R. (2010). Becoming a better person: temporal remoteness biases autobiographical memories for moral events. Emotion, 10(4), 511. Eyal, T., & Liberman, N. (2012). Morality and psychological distance: A construal level theory perspective. The social psychology of morality: exploring the causes of good and evil, 185-202. Eyal, T., Liberman, N., & Trope, Y. (2008). Judging near and distant virtue and vice. Journal of Experimental Social Psychology, 44(4), 1204-1209. Eyal, T., Sagristano, M. D., Trope, Y., Liberman, N., & Chaiken, S. (2009). When values matter: Expressing values in behavioral intentions for the near vs. distant future. Journal of Experimental Social Psychology, 45(1), 35-43. FeldmanHall, O., Mobbs, D., Evans, D., Hiscox, L., Navrady, L., & Dalgleish, T. (2012). What we say and what we do: the relationship between real and hypothetical moral choices. Cognition, 123(3), 434-441. Ferrer, R. A., Grenen, E. G., & Taber, J. M. (2015). Effectiveness of internet-based affect induction procedures: A systematic review and meta-analysis. In: American Psychological Association. Fletcher, T. D., & Major, D. A. (2006). The effects of communication modality on performance and self‐ratings of teamwork components. Journal of Computer‐ Mediated Communication, 11(2), 557-576. Gao, T., Newman, G. E., & Scholl, B. J. (2009). The psychophysics of chasing: A case study in the perception of animacy. Cognitive Psychology, 59(2), 154-179. Gerrards‐Hesse, A., Spies, K., & Hesse, F. W. (1994). Experimental inductions of emotional states and their effectiveness: A review. British Journal of Psychology, 85(1), 55-78. Gesn, P. R., & Ickes, W. (1999). The development of meaning contexts for empathic accuracy: Channel and sequence effects. Journal of Personality and Social Psychology, 77(4), 746. Gong, H., & Medin, D. L. (2012). Construal levels and moral judgment: Some complications. Judgment and Decision Making, 7(5), 628. Gosling, S. D., Rentfrow, P. J., & Swann Jr, W. B. (2003). A very brief measure of the Big- Five personality domains. Journal of Research in Personality, 37(6), 504-528. Graham, J., Haidt, J., Koleva, S., Motyl, M., Iyer, R., Wojcik, S. P., & Ditto, P. H. (2012). Moral foundations theory: The pragmatic validity of moral pluralism. Chapter 8: General Discussion 232
Graham, J., Haidt, J., & Nosek, B. A. (2009). Liberals and conservatives rely on different sets of moral foundations. Journal of Personality and Social Psychology, 96(5), 1029. Graham, J., Nosek, B. A., Haidt, J., Iyer, R., Koleva, S., & Ditto, P. H. (2011). Mapping the moral domain. Journal of Personality and Social Psychology, 101(2), 366. Gray, H. M., Gray, K., & Wegner, D. M. (2007). Dimensions of mind perception. Science, 315(5812), 619-619. Gray, K., Young, L., & Waytz, A. (2012). Mind perception is the essence of morality. Psychological Inquiry, 23(2), 101-124. Greene, J. D., Cushman, F. A., Stewart, L. E., Lowenberg, K., Nystrom, L. E., & Cohen, J. D. (2009). Pushing moral buttons: The interaction between personal force and intention in moral judgment. Cognition, 111(3), 364-371. Greene, J. D., Nystrom, L. E., Engell, A. D., Darley, J. M., & Cohen, J. D. (2004). The neural bases of cognitive conflict and control in moral judgment. Neuron, 44(2), 389-400. Greene, J. D., Sommerville, R. B., Nystrom, L. E., Darley, J. M., & Cohen, J. D. (2001). An fMRI investigation of emotional engagement in moral judgment. Science, 293(5537), 2105-2108. Grezes, J., Fonlupt, P., Bertenthal, B., Delon-Martin, C., Segebarth, C., & Decety, J. (2001). Does perception of biological motion rely on specific brain regions? Neuroimage, 13(5), 775-785. Gui, D.-Y., Gan, T., & Liu, C. (2016). Neural evidence for moral intuition and the temporal dynamics of interactions between emotional processes and moral cognition. Social Neuroscience, 11(4), 380-394. Haidt, J. (2001). The emotional dog and its rational tail: a social intuitionist approach to moral judgment. Psychological Review, 108(4), 814. Haidt, J. (2007). The new synthesis in moral psychology. Science, 316(5827), 998-1002. Hall, D. L., Cohen, A. B., Meyer, K. K., Varley, A. H., & Brewer, G. A. (2015). Costly signaling increases trust, even across religious affiliations. Psychological Science, 26(9), 1368-1376. Hall, J. A., & Schmid Mast, M. (2007). Sources of accuracy in the empathic accuracy paradigm. Emotion, 7(2), 438. Hammick, J. K., & Lee, M. J. (2014). Do shy people feel less communication apprehension online? The effects of virtual reality on the relationship between personality characteristics and communication outcomes. Computers in Human Behavior, 33, 302-310. Hancock, J. T., & Dunham, P. J. (2001). Impression formation in computer-mediated communication revisited an analysis of the breadth and intensity of impressions. Communication Research, 28(3), 325-347. Hancock, P. A., Billings, D. R., Schaefer, K. E., Chen, J. Y., De Visser, E. J., & Parasuraman, R. (2011). A meta-analysis of factors affecting trust in human-robot interaction. Human Factors, 53(5), 517-527. Harenski, C. L., Antonenko, O., Shane, M. S., & Kiehl, K. A. (2008). Gender differences in neural mechanisms underlying moral sensitivity. Social Cognitive and Affective euroscience, 3(4), 313-321. Haslam, N., Bain, P., Douge, L., Lee, M., & Bastian, B. (2005). More human than you: Attributing humanness to self and others. Journal of Personality and Social Psychology, 89(6), 937. Hayes, A. F. (2012). PROCESS: A versatile computational tool for observed variable mediation, moderation, and conditional process modeling. In: University of Kansas, KS. Chapter 8: General Discussion 233
Hilton, D., McClure, J., & Slugoski, B. (2005). Counterfactuals, conditionals and causality: A social psychological perspective. In David R. Mandel, Denis J. Hilton & Patrizia Catellani (eds.), The psychology of counterfactual thinking. Routledge. pp. 44-60. Hmielowski, J. D., Hutchens, M. J., & Cicchirillo, V. J. (2014). Living in an age of online incivility: Examining the conditional indirect effects of online discussion on political flaming. Information, Communication & Society, 17(10), 1196-1211. Hoffman, G., Birnbaum, G. E., Vanunu, K., Sass, O., & Reis, H. T. (2014). Robot responsiveness to human disclosure affects social impression and appeal. Paper presented at the Proceedings of the 2014 ACM/IEEE international conference on Human-robot interaction. Hofmann, W., Wisneski, D. C., Brandt, M. J., & Skitka, L. J. (2014). Morality in everyday life. Science, 345(6202), 1340-1343. Holgado–Tello, F. P., Chacón–Moscoso, S., Barbero–García, I., & Vila–Abad, E. (2010). Polychoric versus Pearson correlations in exploratory and confirmatory factor analysis of ordinal variables. Quality & Quantity, 44(1), 153. Horberg, E., Kraus, M. W., & Keltner, D. (2013). Pride displays communicate self-interest and support for meritocracy. Journal of Personality and Social Psychology, 105(1), 24. Hutchens, M. J., Cicchirillo, V. J., & Hmielowski, J. D. (2015). How could you think that?!?!: Understanding intentions to engage in political flaming. New Media & Society, 17(8), 1201-1219. Iliev, R. I., Sachdeva, S., & Medin, D. L. (2012). Moral kinematics: The role of physical factors in moral judgments. Memory & Cognition, 40(8), 1387-1401. Jackson, L. E., Gaertner, L., & Batson, C. D. (2016). Can Affect Disengagement Produce Moral Standard Violation? Self and Identity, 15(1), 19-31. Jane, E. A. (2015). Flaming? What flaming? The pitfalls and potentials of researching online hostility. Ethics and Information Technology, 17(1), 65-87. Jensen, C., Farnham, S. D., Drucker, S. M., & Kollock, P. (2000). The effect of communication modality on cooperation in online environments. Paper presented at the Proceedings of the SIGCHI conference on Human Factors in Computing Systems. Jiang, L. C., Bazarova, N. N., & Hancock, J. T. (2011). The disclosure–intimacy link in computer‐mediated communication: An attributional extension of the hyperpersonal model. Human Communication Research, 37(1), 58-77. Joinson, A. N. (2001). Self‐disclosure in computer‐mediated communication: The role of self‐awareness and visual anonymity. European Journal of Social Psychology, 31(2), 177-192. Joinson, A. N., & Paine, C. B. (2007). Self-disclosure, privacy and the Internet. In A. Johnson, K. McKenna, T. Postmes (eds), Oxford handbook of Internet psychology, 237-252. Joinson, A. N., Reips, U.-D., Buchanan, T., & Schofield, C. B. P. (2010). Privacy, trust, and self-disclosure online. Human–Computer Interaction, 25(1), 1-24. Joinson, A. N., Woodley, A., & Reips, U.-D. (2007). Personalization, authentication and self- disclosure in self-administered Internet surveys. Computers in Human Behavior, 23(1), 275-285. Judd, C. M., Westfall, J., & Kenny, D. A. (2012). Treating stimuli as a random factor in social psychology: A new and comprehensive solution to a pervasive but largely ignored problem. Journal of Personality and Social Psychology, 103(1), 54. Kayany, J. M. (1998). Contexts of uninhibited online behavior: Flaming in social newsgroups on Usenet. Journal of the American Society for Information Science, 49(12), 1135- 1141. Chapter 8: General Discussion 234
Keltner, D. (1996). Evidence for the distinctness of embarrassment, shame, and guilt: A study of recalled antecedents and facial expressions of emotion. Cognition & Emotion, 10(2), 155-172. Kiesler, S., Siegel, J., & McGuire, T. W. (1984). Social psychological aspects of computer- mediated communication. American Psychologist, 39(10), 1123. Kiesler, S., Zubrow, D., Moses, A. M., & Geller, V. (1985). Affect in computer-meditated communication: An experiment in synchronous terminal-to-terminal discussion. Human-Computer Interaction, 1(1), 77-104. Knyazev, G. G., Savostyanov, A. N., Bocharov, A. V., Dorosheva, E. A., Tamozhnikov, S. S., & Saprigyn, A. E. (2016). Oscillatory correlates of moral decision-making: Effect of personality. Social Neuroscience, 11(3), 233-248. Kock, N. (2005). Media richness or media naturalness? The evolution of our biological communication apparatus and its influence on our behavior toward e-communication tools. IEEE Transactions on Professional Communication, 48(2), 117-130. Koda, T., & Maes, P. (1996). Agents with faces: The effect of personification. Paper presented at the Robot and Human Communication, 1996., 5th IEEE International Workshop on. Kohlberg, L. (1966). A cognitive-developmental analysis of children’s sex-role concepts and attitudes. In E. R. Macoby (ed), The development of sex differences (pp. 82 – 173). Standford University Press. Koo, T. K., & Li, M. Y. (2016). A guideline of selecting and reporting intraclass correlation coefficients for reliability research. Journal of Chiropractic Medicine, 15(2), 155-163. Kress, G. (2009). Multimodality: A social semiotic approach to contemporary communication: Routledge. Kross, E., Ayduk, O., & Mischel, W. (2005). When asking “why” does not hurt distinguishing rumination from reflective processing of negative emotions. Psychological Science, 16(9), 709-715. Kteily, N., Bruneau, E., Waytz, A., & Cotterill, S. (2015). The ascent of man: Theoretical and empirical evidence for blatant dehumanization. Journal of Personality and Social Psychology, 109(5), 901. Kurzban, R., DeScioli, P., & O'Brien, E. (2007). Audience effects on moralistic punishment. Evolution and Human Behavior, 28(2), 75-84. Lagnado, D. A., & Channon, S. (2008). Judgments of cause and blame: The effects of intentionality and foreseeability. Cognition, 108(3), 754-770. Laineste, L. (2013). Funny or aggressive? Failed humour in Internet comments. Folklore, 53, 29-46. Lapidot-Lefler, N., & Barak, A. (2012). Effects of anonymity, invisibility, and lack of eye- contact on toxic online disinhibition. Computers in Human Behavior, 28(2), 434-443. Lea, M., O'Shea, T., Fung, P., & Spears, R. (1992). 'Flaming' in computer-mediated communication: Observations, explanations, implications. Harvester Wheatsheaf. Lea, M., Spears, R., & de Groot, D. (2001). Knowing me, knowing you: Anonymity effects on social identity processes within groups. Personality and Social Psychology Bulletin, 27(5), 526-537. Lee, J. J., & Gino, F. (2015). Poker-faced morality: Concealing emotions leads to utilitarian decision making. Organizational Behavior and Human Decision Processes, 126, 49- 64. Lench, H. C., Flores, S. A., & Bench, S. W. (2011). Discrete emotions predict changes in cognition, judgment, experience, behavior, and physiology: a meta-analysis of experimental emotion elicitations. Psychological Bulletin, 137(5), 834. Chapter 8: General Discussion 235
Lenhart, A. (2015). Teens, Technology and Friendships. Retrieved from http://www.pewinternet.org/2015/08/06/teens-technology-and-friendships/%3E Leue, A., & Beauducel, A. (2015). Effects of injustice sensitivity and sex on the P3 amplitude during deception. Biological Psychology, 109, 29-36. Levay, K. E., Freese, J., & Druckman, J. N. (2016). The demographic and political composition of Mechanical Turk samples. Sage Open, 6(1), 2158244016636433. Levine, D. (2000). Virtual attraction: What rocks your boat. CyberPsychology & Behavior, 3(4), 565-573. Liao, W., Bazarova, N. N., & Yuan, Y. C. (2018). Unpacking Medium Effects on Social Psychological Processes in Computer-mediated Communication Using the Social Relations Model. Journal of Computer-Mediated Communication, 23(2), 90-106. Lotto, L., Manfrinati, A., & Sarlo, M. (2014). A new set of moral dilemmas: Norms for moral acceptability, decision times, and emotional salience. Journal of Behavioral Decision Making, 27(1), 57-65. Luguri, J. B., Napier, J. L., & Dovidio, J. F. (2012). Reconstruing intolerance: Abstract thinking reduces conservatives’ prejudice against nonnormative groups. Psychological Science, 23(7), 756-763. Martin, A. J., & Debus, R. L. (1999). Alternative factor structure for the revised self- consciousness scale. Journal of Personality Assessment, 72(2), 266-281. McClure, J., Hilton, D. J., & Sutton, R. M. (2007). Judgments of voluntary and physical causes in causal chains: Probabilistic and social functionalist criteria for attributions. European Journal of Social Psychology, 37(5), 879-901. McCroskey, J. C., & McCain, T. A. (1974). The measurement of interpersonal attraction. Mele, A. R., & Cushman, F. (2007). Intentional action, folk judgments, and stories: Sorting things out. Midwest Studies in Philosophy, 31(1), 184-201. Metzger, M. J. (2004). Privacy, trust, and disclosure: Exploring barriers to electronic commerce. Journal of Computer-Mediated Communication, 9(4), JCMC942. Miller, L. C., Berg, J. H., & Archer, R. L. (1983). Openers: Individuals who elicit intimate self-disclosure. Journal of Personality and Social Psychology, 44(6), 1234. Molaei, H. (2014). The prospect of civility in Indonesians' online polarized political discussions. Asian Journal of Communication, 24(5), 490-504. Molenberghs, P., Gapp, J., Wang, B., Louis, W. R., & Decety, J. (2014). Increased moral sensitivity for outgroup perpetrators harming ingroup members. Cerebral Cortex, 26(1), 225-33. Molenberghs, P., Ogilvie, C., Louis, W. R., Decety, J., Bagnall, J., & Bain, P. G. (2015). The neural correlates of justified and unjustified killing: an fMRI study. Social Cognitive and Affective Neuroscience, 10(10), 1397-1404. Monin, B., Pizarro, D. A., & Beer, J. S. (2007). Deciding versus reacting: Conceptions of moral judgment and the reason-affect debate. Review of General Psychology, 11(2), 99. Moor, P. J., Heuvelman, A., & Verleur, R. (2010). Flaming on youtube. Computers in Human Behavior, 26(6), 1536-1546. Morio, H., & Buchholz, C. (2009). How anonymous are you online? Examining online social behaviors from a cross-cultural perspective. Ai & Society, 23(2), 297-307. Mulder, L. B., Verboon, P., & De Cremer, D. (2009). Sanctions and moral judgments: The moderating effect of sanction severity and trust in authorities. European Journal of Social Psychology, 39(2), 255-269. Mungeam, F. (2011). Commenting on the news: How the degree of anonymity affects flaming online. Journal of Information Technology and Politics, 13(1), 37-51. Chapter 8: General Discussion 236
Naglirei, J., & Graham, J. (Ed). (2003). Handbook of psychology, assessment psychology. Wiley. Napier, J. L., & Luguri, J. B. (2013). Moral mind-sets: Abstract thinking increases a preference for “individualizing” over “binding” moral foundations. Social Psychological and Personality Science, 4(6), 754-759. Nation, I., & Beglar, D. (2007). A vocabulary size test. The Language Teacher, 31(7), 9-13. Navarrete, C. D., McDonald, M. M., Mott, M. L., & Asher, B. (2012). Virtual morality: Emotion and action in a simulated three-dimensional “trolley problem”. Emotion, 12(2), 364. Ng, E. W., & Detenber, B. H. (2005). The impact of synchronicity and civility in online political discussions on perceptions and intentions to participate. Journal of Computer-Mediated Communication, 10(3), JCMC1033. Nguyen, M., Bin, Y. S., & Campbell, A. (2012). Comparing online and offline self- disclosure: A systematic review. Cyberpsychology, Behavior, and Social Networking, 15(2), 103-111. Niedenthal, P. M. (2007). Embodying emotion. Science, 316(5827), 1002-1005. Niedenthal, P. M., Barsalou, L. W., Winkielman, P., Krauth-Gruber, S., & Ric, F. (2005). Embodiment in attitudes, social perception, and emotion. Personality and Social Psychology Review, 9(3), 184-211. Nowak, K. L. (2004). The influence of anthropomorphism and agency on social judgment in virtual environments. Journal of Computer-Mediated Communication, 9(2), JCMC925. Nowak, K. L., & Biocca, F. (2003). The effect of the agency and anthropomorphism on users' sense of telepresence, copresence, and social presence in virtual environments. Presence: Teleoperators & Virtual Environments, 12(5), 481-494. Nowak, K. L., & Rauh, C. (2005). The influence of the avatar on online perceptions of anthropomorphism, androgyny, credibility, homophily, and attraction. Journal of Computer‐Mediated Communication, 11(1), 153-178. Otto, P. E., & Bolle, F. (2015). Exploiting one’s power with a guilty conscience: An experimental investigation of self-serving biases. Journal of Economic Psychology, 51, 79-89. Patil, I., Cogoni, C., Zangrando, N., Chittaro, L., & Silani, G. (2014). Affective basis of judgment-behavior discrepancy in virtual experiences of moral dilemmas. Social neuroscience, 9(1), 94-107. Pfeil, U., & Zaphiris, P. (2007). Patterns of empathy in online communication. Paper presented at the SIGCHI Conference on Human factors in Computing Systems, San Jose, California. Pizarro, D. (2000). Nothing more than feelings? The role of emotions in moral judgment. Journal for the Theory of Social Behaviour, 30(4), 355-375. Pizarro, D. A., & Bloom, P. (2003). The intelligence of the moral intuitions: A comment on Haidt (2001). Psychological Review, 110(1), 193-196. Pizarro, D. A., Uhlmann, E., & Bloom, P. (2003). Causal deviance and the attribution of moral responsibility. Journal of Experimental Social Psychology, 39(6), 653-660. Postmes, T., Spears, R., & Lea, M. (1998). Breaching or building social boundaries? SIDE- effects of computer-mediated communication. Communication Research, 25(6), 689- 715. Proeve, M. J., & Howells, K. (2006). Effects of remorse and shame and criminal justice experience on judgements about a sex offender. Psychology, Crime & Law, 12(2), 145-161. Chapter 8: General Discussion 237
Ramirez, A., Walther, J. B., Burgoon, J. K., & Sunnafrank, M. (2002). Information‐seeking strategies, uncertainty, and computer‐mediated communication. Human Communication Research, 28(2), 213-228. Raney, A. A. (2002). Moral judgment as a predictor of enjoyment of crime drama. Media Psychology, 4(4), 305-322. Reicher, S. D., Spears, R., & Postmes, T. (1995). A social identity model of deindividuation phenomena. European Review of Social Psychology, 6(1), 161-198. Reid, D. J., & Reid, F. J. (2005). Online focus groups. International Journal of Market Research, 47(2), 131. Reinig, B. A., & Mejias, R. J. (2004). The effects of national culture and anonymity on flaming and criticalness in GSS-supported discussions. Small Group Research, 35(6), 698-723. Richardson, M. J., Marsh, K. L., & Schmidt, R. (2005). Effects of visual and verbal interaction on unintentional interpersonal coordination. Journal of Experimental Psychology: Human Perception and Performance, 31(1), 62. Rietveld, T., & van Hout, R. (2007). Analysis of variance for repeated measures designs with word materials as a nested random or fixed factor. Behavior Research Methods, 39(4), 735-747. Riva, G., Mantovani, F., Capideville, C. S., Preziosa, A., Morganti, F., Villani, D., . . . Alcañiz, M. (2007). Affective interactions using virtual reality: the link between presence and emotions. CyberPsychology & Behavior, 10(1), 45-56. Rothschild, Z. K., Landau, M. J., Keefer, L. A., & Sullivan, D. (2015). Another’s punishment cleanses the self: Evidence for a moral cleansing function of punishing transgressors. Motivation and Emotion, 39(5), 722-741. Rottman, J., Kelemen, D., & Young, L. (2014). Tainting the soul: Purity concerns predict moral judgments of suicide. Cognition, 130(2), 217-226. Royzman, E., Atanasov, P., Landy, J. F., Parks, A., & Gepty, A. (2014). CAD or MAD? Anger (not disgust) as the predominant response to pathogen-free violations of the divinity code. Emotion, 14(5), 892. Rozin, P., Lowery, L., Imada, S., & Haidt, J. (1999). The CAD triad hypothesis: a mapping between three moral emotions (contempt, anger, disgust) and three moral codes (community, autonomy, divinity). Journal of Personality and Social Psychology, 76(4), 574. Saarela, M. V., Hlushchuk, Y., Williams, A. C. d. C., Schürmann, M., Kalso, E., & Hari, R. (2007). The compassionate brain: humans detect intensity of pain from another's face. Cerebral Cortex, 17(1), 230-237. Sachdeva, S., Iliev, R., Ekhtiari, H., & Dehghani, M. (2015). The Role of Self-Sacrifice in Moral Dilemmas. PloS One, 10(6), e0127409. Santana, A. D. (2014). Virtuous or vitriolic: The effect of anonymity on civility in online newspaper reader comment boards. Journalism Practice, 8(1), 18-33. Sasaki, M., & Ohbuchi, K.-i. (1999). Conflict processes on computer-mediated communication. Tohoku Psychologica Folia, 58, 50-55. Schaefer, A., Nils, F., Sanchez, X., & Philippot, P. (2010). Assessing the effectiveness of a large database of emotion-eliciting films: A new tool for emotion researchers. Cognition and Emotion, 24(7), 1153-1172. Schnall, S., Haidt, J., Clore, G. L., & Jordan, A. H. (2008). Disgust as embodied moral judgment. Personality and Social Psychology Bulletin, 34(8), 1096-1109. Scholl, B. J., & Tremoulet, P. D. (2000). Perceptual causality and animacy. Trends in cognitive sciences, 4(8), 299-309. Chapter 8: General Discussion 238
Schouten, A. P., Valkenburg, P. M., & Peter, J. (2007). Precursors and underlying processes of adolescents' online self-disclosure: Developing and testing an “Internet-attribute- perception” model. Media Psychology, 10(2), 292-315. Schroeder, J., & Epley, N. (2015). The Sound of Intellect Speech Reveals a Thoughtful Mind, Increasing a Job Candidate’s Appeal. Psychological Science, 26(6), 877-891. Schroeder, J., & Epley, N. (2016a). Mistaking minds and machines: How speech affects dehumanization and anthropomorphism. Journal of Experimental Psychology: General, 145(11), 1427. Scollon, C. N., & King, L. A. (2004). Is the good life the easy life? Social Indicators Research, 68(2), 127-162. Septianto, F., Septianto, F., Soegianto, B., & Soegianto, B. (2017). Being moral and doing good to others: Re-examining the role of emotion, judgment, and identity on prosocial behavior. Marketing Intelligence & Planning, 35(2), 180-191. Sheldon, P. (2009). " I'll poke you. You'll poke me!" Self-disclosure, social attraction, predictability and trust as important predictors of Facebook relationships. Cyberpsychology: Journal of Psychosocial Research on Cyberspace, 3(2). Short, J., Williams, E., & Christie, B. (1976). The social psychology of telecommunications. Shweder, R. A., Mahapatra, M., & Miller, J. G. (1987). Culture and moral development. In J. Kagan & S. Lamb (Eds.), The emergence of morality in young children (pp. 1-83). University of Chicago Press. Simon-Thomas, E. R., Keltner, D. J., Sauter, D., Sinicropi-Yao, L., & Abramson, A. (2009). The voice conveys specific emotions: evidence from vocal burst displays. Emotion, 9(6), 838. Simpson, A., & Laham, S. M. (2015). Individual differences in relational construal are associated with variability in moral judgment. Personality and Individual Differences, 74, 49-54. Spears, R. (2017). Social identity model of deindividuation effects. In P. Rössler (Ed), The International Encyclopedia of Media Effects (pp. 1-9). Wiley-Blackwell. Sproull, L., & Kiesler, S. (1986). Reducing social context cues: Electronic mail in organizational communication. Management Science, 32(11), 1492-1512. Sproull, L., & Kiesler, S. (1991). New ways of working in the networked organization. Cambridge, MA. Stanger, N., Kavussanu, M., Willoughby, A., & Ring, C. (2012). Psychophysiological responses to sport-specific affective pictures: A study of morality and emotion in athletes. Psychology of Sport and Exercise, 13(6), 840-848. Steel, J. L. (1991). Interpersonal correlates of trust and self-disclosure. Psychological Reports, 68(3), 1319-1320. Stritzke, W. G., Nguyen, A., & Durkin, K. (2004). Shyness and computer-mediated communication: A self-presentational theory perspective. Media Psychology, 6(1), 1- 22. Suler, J. (2004). The online disinhibition effect. CyberPsychology & Behavior, 7(3), 321-326. Sunnafrank, M. (1986). Predicted outcome value during initial interactions: A reformulation of uncertainty reduction theory. Human Communication Research, 13(1), 3-33. Swaab, R. I., Galinsky, A. D., Medvec, V., & Diermeier, D. A. (2012). The communication orientation model: Explaining the diverse effects of sight, sound, and synchronicity on negotiation and group decision-making outcomes. Personality and Social Psychology Review, 16(1), 25-53. Swaab, R. I., & Swaab, D. F. (2009). Sex differences in the effects of visual contact and eye contact in negotiations. Journal of Experimental Social Psychology, 45(1), 129-136. Chapter 8: General Discussion 239
Taddei, S., & Contena, B. (2013). Privacy, trust and control: Which relationships with online self-disclosure? Computers in Human Behavior, 29(3), 821-826. Tait, S., & Jeske, D. (2015). Hello stranger! Trust and self-disclosure effects on online information sharing. International Journal of Cyber Behavior, Psychology and Learning, 5(1), 42-55. Tanis, M., & Postmes, T. (2003). Social cues and impression formation in CMC. Journal of Communication, 53(4), 676-693. Tidwell, L. C., & Walther, J. B. (2002). Computer‐mediated communication effects on disclosure, impressions, and interpersonal evaluations: Getting to know one another a bit at a time. Human Communication Research, 28(3), 317-348. Tingley, D., Yamamoto, T., Hirose, K., Keele, L., & Imai, K. (2014). Mediation: R package for causal mediation analysis. Journal of Statistical Software, 59(5). Tong, S. T., & Walther, J. B. (2015). The confirmation and disconfirmation of expectancies in computer-mediated communication. Communication Research, 42(2), 186-212. Tottenham, N., Tanaka, J. W., Leon, A. C., McCarry, T., Nurse, M., Hare, T. A., . . . Nelson, C. (2009). The NimStim set of facial expressions: judgments from untrained research participants. Psychiatry Research, 168(3), 242-249. Tourangeau, R., Couper, M. P., & Steiger, D. M. (2003). Humanizing self-administered surveys: experiments on social presence in web and IVR surveys. Computers in Human Behavior, 19(1), 1-24. Valenzuela, S., Park, N., & Kee, K. F. (2009). Is there social capital in a social network site?: Facebook use and college students' life satisfaction, trust, and participation. Journal of Computer-Mediated Communication, 14(4), 875-901. Van de Vyver, J., & Abrams, D. (2015). Testing the prosocial effectiveness of the prototypical moral emotions: Elevation increases benevolent behaviors and outrage increases justice behaviors. Journal of Experimental Social Psychology, 58, 23-33. Verma, R., Nitin, N., & Srivastava, A. (2016). On Behavioural Responses and Different Shades of Flaming in Social Media and Computer Mediated Communication. International Journal of Human Capital and Information Technology Professionals (IJHCITP), 7(4), 33-49. Vess, M., Rogers, R., Routledge, C., & Hicks, J. A. (2016). When being far away is good: Exploring how mortality salience, regulatory mode, and goal progress affect judgments of meaning in life. European Journal of Social Psychology. Walther, J. B. (1993). Impression development in computer‐mediated interaction. Western Journal of Communication, 57(4), 381-398. Walther, J. B. (1996). Computer-mediated communication: Impersonal, interpersonal, and hyperpersonal interaction. Communication Research, 23(1), 3-43. Walther, J. B. (2007). Selective self-presentation in computer-mediated communication: Hyperpersonal dimensions of technology, language, and cognition. Computers in Human Behavior, 23(5), 2538-2557. Walther, J. B., Hoter, E., Ganayem, A., & Shonfeld, M. (2015). Computer-mediated communication and the reduction of prejudice: A controlled longitudinal field experiment among Jews and Arabs in Israel. Computers in Human Behavior, 52, 550- 558. Walther, J. B., Jang, J.-W., Kashian, N. C., Falin, J., Shin, S. Y., & Paul, A. (2013). Intimacy in computer-mediated communication: The underestimation of affective influence leads to the illusion of liking online. Paper presented at the Annual Meeting of the International Communication Association, London, England. Walther, J. B., & Tong, S. T. (2014). Inventing partners in computer-mediated communication: how CMC sustains self-fulfilling prophecies and relational Chapter 8: General Discussion 240
attributions. Paper presented at the International Conference on Social Computing and Social Media. Wang, T., Mo, L., Mo, C., Tan, L. H., Cant, J. S., Zhong, L., & Cupchik, G. (2014). Is moral beauty different from facial beauty? Evidence from an fMRI study. Social Cognitive and Affective Neuroscience, 10(6), 814-823. Wenburg, J. R., & Wilmot, W. V. (1973). The personal communication process. New York: John Wiley and Sons. Westerman, D., Spence, P. R., & Van Der Heide, B. (2014). Social media as information source: Recency of updates and credibility of information. Journal of Computer‐ Mediated Communication, 19(2), 171-183. Wexelblat, A. (1997). Don’t make that face: A report on anthropomorphizing an interface. Retrieved from http://wex.www.media.mit.edu/people/wex/anthro-expt- paper/Anthro-r.html. Wheatley, T., & Haidt, J. (2005). Hypnotic disgust makes moral judgments more severe. Psychological Science, 16(10), 780-784. Wheeless, L. R. (1978). A follow‐up study of the relationships among trust, disclosure, and interpersonal solidarity. Human Communication Research, 4(2), 143-157. Wheeless, L. R., & Grotz, J. (1977). The measurement of trust and its relationship to self- disclosure. Human Communication Research, 3(3), 250-257. Whitton, A. E., Henry, J. D., Rendell, P. G., & Grisham, J. R. (2014). Disgust, but not anger provocation, enhances levator labii superioris activity during exposure to moral transgressions. Biological Psychology, 96, 48-56. Whitty, M. T., & Carr, A. N. (2006). New rules in the workplace: Applying object-relations theory to explain problem Internet and email behaviour in the workplace. Computers in Human Behavior, 22(2), 235-250. Wiegmann, A., & Waldmann, M. R. (2014). Transfer effects between moral dilemmas: A causal model theory. Cognition, 131(1), 28-43. Williams, L. E., & Bargh, J. A. (2008). Keeping one's distance: The influence of spatial distance cues on affect and evaluation. Psychological Science, 19(3), 302-308. Wisneski, D. C., & Skitka, L. J. (2017). Moralization through moral shock: Exploring emotional antecedents to moral conviction. Personality and Social Psychology Bulletin, 43(2), 139-150. Yilmaz, O., Harma, M., Bahçekapili, H. G., & Cesur, S. (2016). Validation of the moral foundations questionnaire in Turkey and its relation to cultural schemas of individualism and collectivism. Personality and Individual Differences, 99, 149-154. Yoder, K. J., & Decety, J. (2014a). The good, the bad, and the just: Justice sensitivity predicts neural response during moral evaluation of actions performed by others. Journal of Neuroscience, 34(12), 4161-4166. Yoder, K. J., & Decety, J. (2014b). Spatiotemporal neural dynamics of moral judgment: a high-density ERP study. Neuropsychologia, 60, 39-45. Young, L., Koenigs, M., Kruepke, M., & Newman, J. P. (2012). Psychopathy increases perceived moral permissibility of accidents. Journal of Abnormal Psychology, 121(3), 659. Zachary, R. A., & Shipley, W. C. (1986). Shipley institute of living scale: Revised manual. WPS, Western Psychological Services. Zaki, J., Bolger, N., & Ochsner, K. (2009). Unpacking the informational bases of empathic accuracy. Emotion, 9(4), 478. Žeželj, I. L., & Jokić, B. R. (2014). Replication of experiments evaluating impact of psychological distance on moral judgment. Social Psychology. Chapter 8: General Discussion 241
Zhao, X., Cusimano, C., & Malle, B. F. (2016). Do people spontaneously take a robot's visual perspective? Paper presented at the Human-Robot Interaction (HRI), 2016 11th ACM/IEEE International Conference on. Supplementary Materials 242
Supplementary Materials
Appendix A
Table 41. Studies with Non-Text Moral Stimuli
Citation Study Moral Stimulus Description Stimulus Dependent Variable Outcome
Medium
Bahnemann et al. (2009) 1 Time-series still images of a harm Dynamic image Moral judgement (Is the Identify neurological violation, participants judged whether behaviour wrong? – binary) structures related to theory of the action was a violation or not mind Batson et al. (2007) 1 Economic game where lottery tickets are FtF Fairness judgement Personal anger and empathic assigned, and empathy is manipulated by Moral behaviour (offering to anger have distinct effects on a confederate communicating via note help) fairness judgements. Only with the participant empathic anger relates to fairness judgements. Brambilla and Riva 2 Picture of a target and a description of Image Moral character (Likert (2017) their characteristics scale) Buon et al. (2013) 1 Three animated clips in which harm to a Video Moral character An agent’s harmless intention victim (non-human object) was a (categorisation as good or is given less weight than her coincidence, accidental, or purposeful bad) harmful causal role, inducing 2 Three animated clips in which harm to a Video participants to judge harshly victim (non-human object) was a the accidental agent coincidence, accidental, or purposeful Caruso and Gino (2011) 1 Audio description of moral behaviours Sound Wrongness judgement Polarised moral judgement 3 Audio description of a moral behaviour Sound (Likert) and more ethical behaviour in Moral behaviour the audio condition compared 4 Audio description of a moral behaviour Sound to the text condition. Medium effect mediated by mental simulation. Supplementary Materials 243
Cikara, Farnsworth, 1 Images of a trolley problem – either 5 Sound Acceptability of moral action Stereotypes about out-group Harris, and Fiske (2010) people were pictured (people saved by (Likert) members were used as a diverting the trolley) or 1 person was trade-off save in-group pictured (person killed by diverting members in trolley dilemmas trolley) Condon and DeSteno 1 Confederate lies about the number of FtF Punishment Inducing compassion lead to (2011) math problems they complete to get less punishment of a third more financial reward from the (unrelated) party experimenter. The participant punishes the confederate with hot sauce Decety and Porges 1 Animated visual sequences comprised of Dynamic image Neural activation (2011) photographs depicting two human actors, whose torsos and limbs were visible but not their face. All stimuli showed two individuals interacting, where one individual was either intentionally harming the other or alleviating their pain. Dogruel, Joeckel, and 1 Participants selected avatars and Virtual reality Number of moral violations Found support for moral Bowman (2013) interacted in a virtual environment foundations theory in a virtual similar to that of World of Warcraft. and interactive world There were 6 possible moral scenarios (based on MFT), where participants could choose to intervene to stop a perpetrator or not Escobedo and Adolphs 1 Participants recalled moral memories, Memory Rating of own memory: Memories of immoral (2010) the memories were also prepared as helping/hurting someone, behaviours are perceived as written transcripts doing the right/wrong thing, more distant than memories feelings of personal moral of moral behaviours strength/weakness
Gui, Gan, and Liu 1 Moral images from the IAPS were rated Image Wrongness judgment Evidence for moral intuition (2016) (Likert) that precedes affective processes Harenski, Antonenko, 1 IAPS moral images, participants judged Image Wrongness judgment There are gender differences Shane, and Kiehl (2008) wrongness (Likert) in neurological activation during moral judgement Horberg, Kraus, and 1 8 images of people either posing with a Image Non-verbal cues are used to Keltner (2013) pride or joy display, participants judged infer moral character Supplementary Materials 244
the targets on their moral values Moral character (support for (meritocracy and egalitarianism) egalitarianism and 3 12 images of people either posing with a Image meritocracy) pride or joy display, participants judged the targets on their moral values (meritocracy and egalitarianism) 4 Video of people describing their Video strengths. The targets were rated on their character and values (meritocracy, egalitarianism, self-interest, pride/joy) Iliev et al. (2012) 1 Animated clips where non-human actors Video Wrongness judgement Physical factors (motion and aggressed against one another (binary – actor X is worse contact) influence moral 2 Animated clips where non-human actors Video than actor Y) judgements aggressed against one another Jackson et al. (2016) 1 One shot dictator game played on a FtF Moral decision (trolley computer using text for both the game dilemma) play and the instructions. Participants allocated tickets to a bogus other and rated how moral their behaviour was Knyazev et al. (2016) 1 Drawing of moral dilemmas paired with Image Sensitivity to moral issues moral vignette descriptions and the ability to grasp the nuances of moral situation are essential for understanding the implications of utilitarian choices in personal and impersonal condition
Kurzban, DeScioli, and 1 A trust game (Berg, Dickhaut, & FtF Punishment of a perpetrator Punishment of a third party O'Brien (2007) McCabe, 1995); Instructions were was greater when there was printed text and game choices were hand an audience privy to the written notes punishment/immoral act 2 A sequential Prisoner's Dilemma; FtF Instructions were printed text and game choices were hand written notes Lee and Gino (2015) 2 Audio presentation of moral dilemmas Sound Moral decision Emotion suppression facilitates utilitarian decisions Supplementary Materials 245
3 Four-minute video clip from the movie Video Vertical Limit where a moral dilemma is pictured Leue and Beauducel 1 Images of faces, participants were Image Moral behaviour – lie or be There are gender differences (2015) instructed to behave deceptively or truthful in the processing of moral truthfully content Molenberghs, Gapp, 1 First-person perspective video game Dynamic image Moral character and Neuroimaging correlates Wang, Louis, and clips in which a soldier, civilian or wrongness judgements related to whether moral Decety (2014) nobody would get shot. Participants actors are in-group or out- decided whom to shoot and were group members instructed to imagine themselves as the shooter. Molenberghs et al. 1 A time-series of static visual images of Dynamic image Guilt about immoral Neuroimaging correlates (2015) people harming each other or acting behaviour related to justified/unjustified peacefully killing Mulder, Verboon, and 1 Economic game where participants allot FtF Wrongness of their Sanctions effect how harsh De Cremer (2009) raffle tickets to behaviour in the game moral judgements are and this others. Instructions/behaviours were effect is moderated by trust in presented with text. authority Navarrete, McDonald, 1 Virtual reality trolley problem including Virtual reality Moral decision (trolley Utilitarian action was more Mott, and Asher (2012) an action and omission condition to de- dilemma) likely than deontological confound action action, in a trolley problem. Paradoxically, utilitarian action was associated with more arousal and arousal was negatively predictive of utilitarian decisions Otto and Bolle (2015) 1 Three economic games FtF Players rated fairness A model-based estimation of social concerns identifies self- serving biases
Raney (2002) 1 Clips from the movie Rob Roy, featuring Video Integrates entertainment either a sexual or non-sexual moral theory and moral judgement transgression during movie/television scenes Supplementary Materials 246
Royzman, Atanasov, 3 Moral vignettes and images of facial Image Facial expressions were the Evidence for anger, not Landy, Parks, and Gepty expressions (Tottenham et al., 2009) response scale to determine disgust, in response to (2014) disgust/anger in response to pathogen-free divinity a moral transgression violations Rothschild, Landau, 2 News article about a hit and run crash, Image Own moral character, Punishment of a third party’s Keefer, and Sullivan with a photograph of the perpetrator perpetrator’s cleanliness, transgression can cleanse the (2015) prosocial behavioural self of guilt for one’s own intentions transgression Sachdeva, Iliev, 3 Drawing of a trolley problem (either in Image Moral decision People are willing to self- Ekhtiari, and Dehghani 3rd or 1st person) and a verbal sacrifice in moral dilemmas (2015) description. Septianto, Septianto, 1 Pictures of faces, paired with written Image, sound Soegianto, and moral behaviours, paired with audio Soegianto (2017) presentation of the moral behaviour Scollon and King (2004) 1 Hand-written career survey Image Moral character (goodness) Living a life with meaning 2 Hand-written career survey Image and effort are associated with higher ratings of moral 3 Hand-written career survey Image character Stanger, Kavussanu, 1 Images of sports people acting morally Image Moral disengagement and Link emotion and moral Willoughby, and Ring toward one another physiological responses to judgement in the context of (2012) images were taken as arousal sporting behaviour Van de Vyver and 1 Three videos: (1) BBC report where Video Moral outrage and pro-social Moral elevation and moral Abrams (2015) money is stolen, (2) man is blinded but behavioural intentions outrage both increased pro- forgives the perpetrator, (3) non-moral social behaviour (or control, then a moral vignette to assess behavioural intentions) but prosocial intentions each emotion was related to 2 As above Video As above different behaviours. 3 Video as above and an economic game Video Moral outrage and pro-social where punishment and compensation behaviour was measured by moral behaviour (allocation of points). Economic game used text for decisions and instructions. Wang et al. (2014) 1 Drawn images of morally relevant Image Moral beauty Moral beauty is associated scenes with a distinct neurological pathway when compared to aesthetic beauty Supplementary Materials 247
2 As above Image Whitton et al. (2014) 1 Autobiographical memory recall of a Image, memory Facial expression was Trait disgust, but not trait moral transgression that made them recall measured anger, correlated with levator angry, participants then viewed moral labii responses to moral images (IAPS) themes
Wiegmann and 1 Trolley problem with text and a Image Moral choice Provides an account of moral Waldmann (2014) matching graphical illustration transfer effects 2 Trolley problem with text and a Image matching graphical illustration 3 Trolley problem with text and a Image matching graphical illustration 5 Trolley problem with text and a Image matching graphical illustration 6 Trolley problem with text and a Image matching graphical illustration 3 Participants were told by the FtF experimenter about moral behaviours of past participants
Wisneski and Skitka 1 People were presented with moral Image Attitude on moral issues Moral conviction about (2017) images and words either at or below abortion increased (compared conscious awareness. After (and with control) only for independently) rated their attitudes on participants exposed to moral issues. abortion-related images at 2 People were presented with moral Image speeds slow enough to allow images and words. After (and conscious awareness. independently) rated their attitudes on moral issues. Yoder and Decety 1 Videos were recorded in a previous Dynamic image Wrongness judgement – is Identify spatio-temporal (2014b) study where confederates were behaving the behaviour good, bad, or pathways for moral immorally. Still frames were extracted neutral? judgements from the videos Blame judgements Yoder and Decety 1 A series of visual stimuli taken from Dynamic image Wrongness judgement and Individual differences in (2014a) short clips of moral behaviour, the faces motive evaluation of actor justice sensitivity impact of the actors were obscured so that the neural computations that Supplementary Materials 248
participants could not see their support psychological expressions. processes involved in moral judgment and mental-state reasoning
Note. Although the aim of this systematic review was to assess the medium of moral stinulus where the stimulus is the direct object of moral judgment, it is worth nothing that several studies that use non-text moral stimuli but the non-text moral stimuli were not the target of moral judgement. If presentation medium affects moral judgement, then, in certain situations, presentation medium may also affect moral judgement indirectly – such as by inducing affect before a moral judgement. Although these studies were not included in the final analyses, they have been noted in this table with an *.
Supplementary Materials 249
Appendix B
Table B1. Descriptive statistics for MAAFS for moral and affective variables Categorised Video Video Moral Uniqueness Seen Number Description Foundation Scores Wrongness Frequency Arousal Funny Punishment Before Clarity Weirdness Basketballers disrespect their 1 coach Authority (4) 43 3.54 3.12 2.53 1.32 3.04 1.09 6.26 2.07
A kid sues her parents to pay for 2 her education. Authority (4) -34 3.32 2.00 2.92 2.30 2.92 2.03 6.19 3.70
A young man 3 swears at police Authority (4) 2 3.07 2.07 3.50 1.00 1.63 1.03 6.63 2.21 Students disrespecting 4 teacher Authority (4) 4 4.12 2.03 3.41 1.00 3.09 1.21 5.91 2.77 An employee destroys her boss' 5 laptop and office Authority (4) -12 2.82 1.68 2.94 1.91 1.68 1.00 6.82 4.11 Child swears at 6 their guardians Authority (4) 18 3.03 1.44 2.97 2.09 2.12 1.06 4.47 3.13 Children disrespect deaf 7 mother Authority (4) -9 3.30 2.41 2.81 2.14 2.16 1.05 6.51 2.98 A basketball player yells at his 8 coaches Authority (4) 3 3.84 2.12 3.14 1.65 3.12 1.12 6.42 2.37 A man disrespects the judge when he 9 is on trial Authority (4) 45 1.60 1.62 2.49 2.04 1.02 1.58 6.76 2.37 Child is abandoned on the side of the 10 road Care (1) 59 2.00 1.25 2.14 2.00 1.19 1.00 6.19 3.24 A man shoots at 11 people Care (1) 25 3.18 3.14 2.73 1.06 1.92 1.22 6.04 3.75 A police officer 12 assaults a woman Care (1) 22 3.31 3.22 2.66 1.36 2.47 1.02 5.57 2.74 Someone throws a shoe at President 13 George Bush Care (1) 11 2.94 2.79 2.39 1.64 2.42 1.00 5.82 4.08 Someone purposefully trips 14 fleeing refugees Care (1) 29 3.46 2.71 3.11 1.04 2.46 1.07 5.93 3.33 A hunter kills an 15 endangered rhino. Care (1) 16 4.09 1.88 2.97 2.47 3.71 1.21 6.21 2.95 A disfigured man was bullied on Instagram by the 16 athlete Shaq Care (1) 3 1.79 1.52 2.66 1.10 1.17 1.07 6.28 1.95 People are starving 17 animals to death Care (1) 48 2.08 2.92 2.00 1.31 1.58 1.00 5.47 2.80 Protesters are 18 beaten by police Care (1) 9 3.77 2.29 2.69 1.37 2.96 1.00 6.40 2.44 Women gossip at 19 work Care (1) -16 4.33 3.31 2.77 1.31 3.03 1.08 6.67 1.86 A man yells insults 20 at his grandma Care (1) 16 3.58 2.89 2.67 1.04 3.35 1.71 6.78 3.25 A bigger boy bullies a smaller 21 boy Care (1) 2 2.90 2.82 2.38 2.59 1.61 1.25 6.70 2.29 A man punches a pregnant woman in 22 the stomach Care (1) 60 1.62 2.79 1.44 1.15 1.03 1.38 6.50 3.00 Kids bully another kid for being 23 overweight Care (1) 7 4.53 2.73 3.87 1.03 3.87 3.63 6.53 2.00 Guys wont date a girl because she’s 24 overweight Care (1) -16 2.68 2.68 2.37 2.13 1.81 3.39 6.79 2.78 Someone throws a 25 shoe at a dog Care (1) 91 2.14 2.64 1.81 1.19 1.23 2.81 6.78 2.50 Two girls fight 26 each other Care (1) 94 2.54 2.54 2.41 1.51 1.89 1.03 4.97 2.95 People make fun of an overweight 27 woman Care (1) 23 2.81 2.34 2.19 2.06 1.41 3.41 6.97 2.41 Teacher hits a student with a 28 ruler Care (1) 71 3.85 2.15 3.09 1.88 3.35 2.38 6.24 3.28 Police in riot gear forcefully deal 29 with protesters Care (1) -7 3.93 2.02 3.42 1.04 3.20 1.09 6.69 1.93 A teenager disrespects his 30 mother Care (1) 3 4.25 2.48 3.33 1.77 3.21 1.06 6.71 2.74 A man is disrespectful toward his 31 adoptive parents Care (1) -14 3.67 2.64 3.12 1.36 3.15 1.03 5.24 2.57 A mother yells at 32 child Care (1) 71 3.80 1.97 2.77 1.43 3.29 1.46 6.60 3.62 Vote are rigged 33 during an election Fairness (2) 85 3.75 1.61 3.02 1.07 2.97 1.05 6.36 3.42 People rob an Apple computer 34 store Fairness (2) 0 3.98 1.62 3.09 1.23 3.30 1.11 6.06 2.42 Someone is not hired for a job because of their 35 ethnicity Fairness (2) 77 2.94 3.03 2.72 1.38 2.16 1.03 6.34 2.49 Student cheats in 36 test Fairness (2) 77 2.88 2.59 2.45 1.45 2.00 1.06 5.39 1.77 A man steals a 37 bike Fairness (2) -5 4.37 2.17 4.25 1.06 4.20 1.11 6.77 2.62 A boy in a hurdle race cheats and runs around the 38 hurdles. Fairness (2) 79 2.74 2.12 2.21 1.79 1.68 1.00 5.38 3.50
Supplementary Materials 250
A man cuts a line so that he can get tickets before other 39 people Fairness (2) 78 3.40 2.00 2.52 2.12 2.86 1.55 6.41 3.11 A woman intentionally dents cans of food in order to get a discount on the 40 product. Fairness (2) 60 4.23 1.94 3.26 1.00 3.94 1.03 6.20 3.30 A man refuses to hire a woman, because she is a 41 woman Fairness (2) 49 1.83 1.80 2.37 1.13 1.07 1.30 6.80 1.76 Man lies about a disability to get extra welfare 42 payments Fairness (2) 42 4.45 1.76 3.67 1.00 4.30 1.00 6.67 2.98 A man backs out of a bet during a 43 pool competition Fairness (2) 0 4.40 1.77 3.80 1.17 3.63 1.00 6.50 2.28 A woman lied to put her husband in 44 jail Fairness (2) 0 4.57 2.08 3.55 1.06 3.96 1.02 5.98 2.92 A guy fakes an illegal tackle to try 45 and get a free kick Fairness (2) 38 3.34 1.66 2.57 2.03 2.69 1.03 6.34 2.72 Woman lies to blind man about the value of money 46 bills Fairness (2) 0 4.17 2.73 3.60 1.03 3.30 1.00 6.83 3.40 A rich man steals money from a 47 homeless person Fairness (2) 0 3.06 2.35 2.84 1.48 2.33 1.84 6.68 3.23 Lance Armstrong admitting to drug 48 cheating Fairness (2) 64 2.63 1.45 1.89 2.79 1.71 1.37 5.53 2.33 African people are 49 sold into slavery Liberty (6) -6 4.43 2.39 3.82 1.23 4.18 1.07 6.02 3.48 The Chinese government censors the internet for the 50 Chinese citizens Liberty (6) 37 4.53 2.12 3.72 1.47 3.81 1.09 6.81 2.06 A girl is forced to wear what her 51 boyfriend wants Liberty (6) 30 3.93 3.18 3.11 1.04 2.77 2.64 6.86 2.82 A young girl is forced to marry an 52 old man Liberty (6) -37 4.27 2.73 3.67 1.12 3.39 2.70 6.21 3.90 A woman catches 53 man cheating Loyalty (3) -5 3.73 2.42 3.03 1.15 3.09 1.00 6.67 1.82 A guy cheats his family out of their money and 54 property Loyalty (3) -2 3.00 2.09 2.85 2.47 2.53 1.03 6.71 3.21 Guy admits to cheating on 55 girlfriend Loyalty (3) 4 2.00 2.23 2.62 1.23 1.58 1.00 5.54 2.21 Bride kisses best 56 man on wedding Loyalty (3) 13 1.86 1.93 2.04 2.25 1.32 1.11 5.71 2.98 A girl is betrayed by her boyfriend to avoid a criminal 57 sentence Loyalty (3) 24 2.74 2.83 2.52 1.70 2.30 1.04 6.35 2.13 A female group audition on talent show and one girl betrays the others for a chance to proceed in the 58 competition Loyalty (3) 80 2.48 1.36 2.45 1.64 1.18 1.33 6.67 1.98 There is a theft from an infant's 59 grave Moral Other (7) 3.72 2.59 3.22 1.19 3.31 1.03 6.03 3.62 A girl goes to the bathroom and injects herself with 60 drugs Sanctity (5) -17 3.09 1.77 2.60 1.54 1.29 1.00 6.74 3.13 A woman steals 61 flowers from grave Sanctity (5) -38 4.33 2.33 3.82 1.06 3.74 1.06 6.00 3.65 A man takes drugs 62 on a bus Sanctity (5) -23 3.89 2.97 3.11 1.11 2.89 1.08 6.70 3.33
63 A KKK ceremony Sanctity (5) -53 3.06 2.32 2.32 2.28 2.56 1.42 6.30 3.47 Note. Values are averages
Supplementary Materials 251
Table B2. Descriptive statistics for MAAFS for discrete emotions Video Warm- Grossed Physical Number Intereste Joyful Disgusted Fearful Anxious Disdainful Surprised hearted Loving Guilty Moved Satisfied Calm Ashamed out Angry Sad Disgust 1 2.02 1.26 2.19 1.33 1.60 2.23 1.60 1.18 1.18 1.40 1.25 1.18 1.46 1.88 1.58 2.28 1.70 0.61 2 2.76 1.70 2.00 1.51 1.76 1.68 2.62 1.27 1.16 1.16 1.35 1.49 1.22 1.32 1.16 1.68 1.38 0.84 3 2.63 1.00 3.50 1.77 2.17 2.47 1.90 1.00 1.00 1.40 1.57 1.00 1.10 1.63 3.57 2.97 2.33 -0.07 4 2.62 1.03 3.32 2.65 2.76 3.09 2.38 1.00 1.00 1.32 1.35 1.03 1.03 1.97 2.18 3.35 2.56 1.14 5 2.15 1.41 3.32 1.29 1.44 2.15 2.44 1.18 1.15 1.09 1.26 1.15 1.44 1.85 3.32 1.94 1.24 0.00 6 2.00 1.56 2.47 1.47 1.59 2.26 2.53 1.24 1.15 1.12 1.15 1.12 1.29 1.62 2.18 2.24 1.44 0.29 7 2.59 1.46 2.30 1.11 1.27 2.03 2.16 1.14 1.05 1.30 1.16 1.19 1.38 1.49 1.05 2.49 1.35 1.25 8 2.19 1.33 2.74 1.19 1.28 2.67 1.67 1.02 1.00 1.16 1.14 1.00 1.30 1.65 1.56 2.77 1.63 1.18 9 2.98 1.91 3.13 1.53 1.78 1.64 2.69 1.47 1.38 1.38 1.58 1.53 1.76 1.67 3.22 1.47 1.40 -0.09 10 2.25 1.47 1.53 1.36 1.39 1.64 2.11 1.08 1.03 1.06 1.06 1.08 1.69 1.14 1.33 1.36 1.44 0.20 11 2.80 1.35 2.31 1.86 1.96 2.33 1.71 1.31 1.35 1.41 1.39 1.35 1.57 1.71 1.59 2.39 1.96 0.72 12 2.50 1.40 2.79 1.91 2.28 2.53 1.93 1.33 1.31 1.43 1.45 1.33 1.41 1.88 1.98 2.45 1.83 0.81 13 2.12 1.18 1.85 1.24 1.79 1.88 1.45 1.06 1.09 1.21 1.09 1.12 1.21 1.36 1.18 1.97 1.30 0.67 14 2.86 1.25 2.93 1.75 2.14 2.54 2.00 1.25 1.29 1.50 2.21 1.25 1.39 1.57 1.50 2.64 2.54 1.43 15 2.74 1.71 2.24 1.41 1.53 2.26 2.88 1.35 1.18 1.32 1.44 1.50 1.38 1.62 1.32 1.94 1.47 0.92 16 2.79 1.21 3.45 1.76 1.93 1.90 2.45 1.07 1.03 1.21 1.66 1.17 1.17 1.52 3.59 1.69 1.76 -0.14 17 2.36 1.22 1.53 1.03 1.19 1.67 1.53 1.06 1.00 1.17 1.08 1.08 1.53 1.39 1.19 1.61 1.06 0.34 18 2.54 1.38 3.25 2.02 2.23 2.54 2.50 1.23 1.23 1.42 1.33 1.25 1.38 1.98 2.85 2.62 1.83 0.40 19 3.15 1.51 3.26 1.69 2.08 2.77 2.08 1.44 1.49 1.82 2.00 1.59 1.90 1.97 2.23 2.54 2.33 1.03 20 2.73 1.38 2.33 1.36 1.67 2.44 2.13 1.31 1.31 1.42 1.62 1.53 1.60 1.96 1.80 2.36 2.02 0.53 21 2.52 1.92 2.02 1.26 1.36 2.18 2.57 1.41 1.31 1.33 1.33 1.38 1.67 1.85 1.66 1.87 1.67 0.36 22 2.32 1.53 1.62 1.41 1.44 1.65 2.03 1.68 1.65 1.41 1.68 1.82 2.15 1.44 1.41 1.65 1.59 0.21 23 2.73 1.40 3.77 2.20 2.80 3.97 2.50 1.40 1.53 1.83 2.33 1.40 1.57 2.13 2.07 3.73 3.53 1.70 24 2.92 1.92 2.32 2.03 2.39 2.18 2.21 1.45 1.34 1.47 1.50 1.58 1.66 1.82 1.82 2.18 1.84 0.50 25 2.22 1.42 2.11 1.47 1.61 1.64 1.44 1.42 1.42 1.61 1.42 1.47 1.81 1.64 2.14 1.83 1.58 -0.03 26 2.43 1.22 1.68 1.38 1.65 1.54 1.68 1.03 1.03 1.08 1.08 1.08 1.30 1.27 1.11 1.43 1.24 0.57 27 2.94 2.19 2.06 1.31 1.50 1.91 2.22 1.44 1.34 1.56 1.50 1.41 1.88 1.94 1.69 2.16 1.72 0.37 28 2.82 1.79 2.94 2.15 2.38 2.68 2.65 1.56 1.50 1.59 1.88 1.62 1.65 1.88 1.97 2.85 1.97 0.97 29 2.64 1.16 2.71 1.47 1.87 2.24 2.24 1.20 1.11 1.64 1.62 1.11 1.18 2.36 2.31 3.22 2.96 0.40 30 2.50 1.33 3.06 1.54 1.83 2.75 2.29 1.25 1.17 1.35 1.35 1.19 1.40 1.88 1.94 2.58 2.04 1.12 31 2.52 1.21 2.91 2.03 2.36 2.55 1.91 1.06 1.03 1.09 1.18 1.03 1.03 1.88 1.76 2.45 1.70 1.15 32 2.71 1.20 2.03 1.11 1.37 2.17 2.11 1.06 1.06 1.06 1.09 1.06 1.17 1.34 1.17 2.11 1.20 0.86 33 2.79 1.32 3.30 1.60 1.85 3.13 2.55 1.30 1.34 1.60 1.75 1.30 1.58 2.15 2.06 3.25 2.34 1.24 34 2.60 1.47 3.26 1.74 1.91 2.96 2.81 1.45 1.38 1.51 1.62 1.43 1.51 2.11 1.96 2.74 2.19 1.30 35 2.41 1.06 2.41 1.28 1.56 2.09 1.47 1.00 1.00 1.19 1.13 1.03 1.19 1.78 1.63 2.56 1.81 0.78 36 2.49 1.35 1.86 1.20 1.33 1.90 1.90 1.12 1.12 1.18 1.22 1.18 1.45 1.27 1.18 1.92 1.43 0.68 37 3.05 1.31 3.83 2.97 3.18 3.25 2.66 1.28 1.23 1.60 1.83 1.26 1.26 2.03 2.60 3.97 3.00 1.23 38 2.38 1.53 1.68 1.21 1.47 1.88 1.47 1.15 1.06 1.26 1.15 1.18 1.29 1.41 1.29 1.91 1.35 0.39 39 2.45 1.79 1.64 1.14 1.24 1.67 2.05 1.29 1.26 1.12 1.14 1.17 1.43 1.33 1.19 1.64 1.21 0.45 40 2.49 1.03 3.00 1.63 1.63 2.54 1.83 1.00 1.00 1.09 1.20 1.00 1.20 1.40 1.37 2.86 1.91 1.63 41 3.13 1.37 1.93 1.37 1.57 1.37 2.67 1.20 1.23 1.13 1.67 1.13 1.70 1.17 2.13 1.37 1.27 -0.20 42 2.58 1.03 3.45 1.42 1.55 2.70 2.64 1.00 1.00 1.12 1.33 1.00 1.09 1.73 1.42 3.42 1.67 2.03 43 2.80 1.13 3.70 1.33 1.67 3.07 2.47 1.00 1.03 1.23 1.43 1.00 1.07 1.73 1.77 3.43 2.07 1.93 44 3.06 1.47 3.96 2.14 2.55 3.73 2.76 1.49 1.51 1.65 2.14 1.53 1.59 2.14 2.14 3.53 3.10 1.82 45 2.54 1.51 2.31 1.09 1.14 2.17 2.23 1.34 1.09 1.14 1.17 1.23 1.51 1.77 1.46 2.14 1.46 0.85 46 2.87 1.10 3.20 1.30 1.80 2.70 2.23 1.00 1.00 1.53 1.80 1.00 1.20 2.23 1.40 3.03 2.13 1.80 47 2.94 1.81 1.90 1.52 2.10 1.84 1.97 1.52 1.42 1.42 1.97 1.81 1.71 1.77 1.39 2.10 1.77 0.51 48 2.29 1.71 1.76 1.00 1.21 1.24 1.42 1.24 1.08 1.00 1.00 1.13 1.45 1.05 1.66 1.32 1.08 0.10 49 2.75 1.23 3.71 2.29 2.82 3.23 2.52 1.21 1.20 1.54 1.82 1.20 1.20 2.45 2.45 3.61 2.84 1.26 50 2.66 1.44 3.47 2.22 2.41 3.00 2.38 1.25 1.31 1.53 1.91 1.34 1.47 1.59 1.88 3.28 3.03 1.59 51 2.50 1.36 2.71 1.61 2.11 2.36 1.54 1.32 1.32 1.93 2.07 1.32 1.61 1.89 1.61 2.57 2.96 1.10 52 2.73 1.39 3.52 2.27 2.94 3.33 2.27 1.39 1.45 1.79 1.85 1.36 1.42 2.03 2.18 3.61 2.79 1.34 53 2.97 1.33 3.24 1.94 2.33 2.70 2.24 1.30 1.30 1.52 1.70 1.30 1.52 1.94 2.15 3.06 2.42 1.09 54 2.56 1.74 2.71 1.59 1.44 2.24 2.47 1.18 1.18 1.32 1.50 1.24 1.44 2.06 1.62 2.35 1.68 1.09 55 2.96 1.58 1.54 1.35 1.88 1.62 1.73 1.27 1.19 1.12 1.31 1.46 1.38 1.27 1.08 2.00 1.15 0.46 56 2.21 2.11 1.50 1.04 1.14 1.32 1.64 1.57 1.43 1.11 1.11 1.32 1.96 1.11 1.18 1.57 1.18 0.32 57 2.17 1.35 1.91 1.17 1.39 1.65 1.65 1.04 1.04 1.00 1.13 1.04 1.35 1.39 1.43 1.96 1.39 0.48 58 2.55 1.27 2.61 1.64 1.67 2.03 3.00 1.18 1.15 1.12 1.15 1.03 1.45 1.52 2.88 1.33 1.42 -0.27 59 3.03 1.31 3.63 2.22 2.50 3.31 2.06 1.31 1.31 1.91 1.94 1.38 1.53 2.28 2.34 3.50 3.09 1.29 60 2.29 1.43 2.66 1.23 1.20 1.71 2.14 1.29 1.29 1.09 1.31 1.11 1.54 1.51 2.51 1.46 1.26 0.15 61 2.82 1.27 3.58 2.58 3.00 3.06 2.18 1.30 1.36 1.88 2.06 1.33 1.36 2.09 2.97 3.55 3.30 0.61 62 2.70 1.38 3.03 1.62 1.97 2.81 2.24 1.32 1.43 1.89 2.19 1.32 1.41 2.05 1.86 3.08 2.84 1.17 63 2.68 1.72 2.25 1.26 1.43 2.11 2.15 1.32 1.28 1.30 1.30 1.32 1.55 1.72 1.66 2.17 1.70 0.59 Note. Values are averages. Emotions (from left to right) are: interested, joyful, disgusted, fearful, anxious, surprised, warm-hearted, loving, guilty, moved, satisfied, calm, ashamed, grossed-out, angry, sad, physical disgust.
Supplementary Materials 252
Glossary Term Definition Communication medium Communication medium refers to the representation format used to disseminate information. Examples include computer- mediated-communication, voice, radio, film, books (Kress, 2009). Communication modality Communication modality refers to the sensory channel used to encode the information. Examples include written and spoken natural language, audio (Kress, 2009). There is often congruence between medium and mode, e.g., radio (medium) and audio (mode), book (medium) and writing (mode). Social and Contextual Cues Social and contextual cues define the nature of the social situation and actors’ roles. Often these cues are conveyed by the physical environment and verbal/non-verbal behaviour of the actors (Sproull & Kiesler, 1986). These cues can be dynamic or static. Static cues emanate from people's appearance and artefacts such as a clock, a private office, a big desk. Dynamic cues emanate from peoples' verbal or non-verbal behaviour which changes over the course of an interaction-for instance, nodding approval and frowning with displeasure (Sproull & Kiesler, 1986). Non-Verbal Cues Communication through sending and receiving wordless cues, written or spoken. For example, facial expression, voice tone. Verbal Cues Communication through spoken or written language. Computer Mediated Communication Computer-mediated-communication (CMC) refers to any interaction in which two or more electronic devices are used as an intermediate (Walther 1992). Flaming Flaming (also referred to as cyber-incivility, e-bile, cyber-bullying or toxic-disinhibition) is generally defined as hostile, aggressive verbal behaviour that occurs online (Jane, 2015; Lea, O'Shea, Fung, & Spears, 1992; O’Sullivan, & Flanagin, 2003; Suler, 2004).
Minerva Access is the Institutional Repository of The University of Melbourne
Author/s: McCurrie, Caitlin
Title: The role of presentation media in the moral domain
Date: 2018
Persistent Link: http://hdl.handle.net/11343/222543
File Description: The role of presentation media in the moral domain
Terms and Conditions: Terms and Conditions: Copyright in works deposited in Minerva Access is retained by the copyright owner. The work may not be altered without permission from the copyright owner. Readers may only download, print and save electronic copies of whole works for their own personal non-commercial use. Any use that exceeds these limits requires permission from the copyright owner. Attribution is essential when quoting or paraphrasing from these works.