My University Dissertation

Abstract

Is morality identifiable from a single, five minutes-long interaction? Previous research has found that a multitude of physical, social, and personality traits are identifiable from a brief interaction. Thus, it stands to reason that one s moral character could be included as one of these traits. Further, past research into this literature has often fallen short of providing an objective measure with which to compare self and other reports. The principle objective of the present study was to find out if morality can be identified and predicted from a brief interaction, and using an objective measure of morality to compare this to. This objective measure is a task which incentivises cheating. The present study recruited dyads of undergraduate students (N = 94) to engage in a brief interaction lasting five minutes, as well as self-reports, other reports about their interaction partner, and our objective measure of moral character. No significant relationship between self-reported moral character, participants interaction partner s ratings of their moral character, and cheating behaviour in the objective measure of morality. The present study concludes that morality might not be able to be identifiable from a brief interaction, but due to a number of limitations in the study, little strength is given to this conclusion. Limitations of the study are discussed, before discussing directions for future research.

Key words: Thin Slicing, Morality, Moral character, Objective measures.

Introduction

On January 26th 1998 Bill Clinton infamously said one of the most memorable quotes in politics, I did not have sexual relations with that woman , referring to Monica Lewinsky (jw00534, 2012). This, as it turned out, was a complete lie, which had the good fortune of being revealed once Lewinsky received judiciary impunity to give testimony as well as DNA evidence (Starr, 1998). Clinton s lie was received quite poorly, and what he did is quite unambiguously wrong, however, Clinton is in no way an outlier. Anecdotally, near everyone lies on a daily basis, ranging from trivial white lies all the way up to more serious deception such as that of Clinton. Empirically, some studies have found people tell between 1 and 2 lies a day (DePaulo, Kashy, Kirkendol, Wyer, Epstein, 1996), whilst others have found that 60% of people tell at least one lie per 10-minute conversation (Feldman, Forrest, Happ, 2002). Unfortunately, measuring lying behaviour is riddled with problems like social desirability bias and people just not being cognizant of when they are lying. Many people claim to be able to spot a liar in mere seconds, but is this just confirmation bias and overestimation of the accuracy of their judgement, or is the credibility to this? And if there is, can people reliably identify a broader scope of immoral behaviours, such as cheating, in a short period of time as well? The potential use of this has broad implications if detectable. This paper seeks to address the latter question of whether or not it is possible to detect immoral behaviour in others from a single interaction.

In general facets of personality and behaviour, research has shown that people can, with a reasonable degree of accuracy, make judgements about others from a short interaction. These short interactions were dubbed thin slices by Ambady and Rosenthal (1992), and typically refer to interactions with, or exposure to, an individual, lasting less than five minutes. It is important to note that not all studies use the same slices some have used interactions, some have just used audio clips, and some just used images. These all offer up valid methods of measuring thin slicing, however using interactions might open up more identifiable behaviour (more detailed sounds, appearance as a whole, etc) and thus could have the potential to be a stronger measurement.

Examples of previous thin slicing research include Sparks, Burleigh, and Barclay (2016) who showed that individuals could accurately predict someone s decision in a prisoner s dilemma, Kraus and Keltner (2009) who showed individuals could accurately predict a person s socioeconomic status, and Fetchenhauer, Groothuis, and Pradel (2010) who showed individuals could predict how altruistically someone would behave in an interaction. These thin slices offer a potential window into how people really think and behave, and can detect traits and behaviours with more accuracy than chance alone.

Despite the overall increase in research in this field of literature, research into thin slicing is sparse in relation to general negative personality traits, and only tangentially related for immorality. Back, Schmukle, and Egloff (2010) found narcissistic traits perceivable during small interactions. Research by Vazire, Naumann, Rentfrow, and Gosling (2008) found narcissism just as accurately identifiable in snap judgements as the big five personality traits. Friedman, Oltmanns, Gleason, and Turkheimer (2006) found that people rated and reacted to people with pathological personality traits differently to those without pathological personality traits following a 30 second video clip. More in relation to thin slicing, Fowler, Lilienfeld, and Patrick (2009) found that psychopathy was able to be accurately identified following 2, 10, and 20 second audio and video clips. Finally Holtzman (2011) found that dark triad personality traits (machiavellianism, narcissism, and psychopathy) were able to be reliable identified from photo composites of individuals who scored highly on those traits. Most research, while providing suggestive results, does not have the broad empirical support to make large claims confidently. Further, most studies into thin slicing as a research literature only use self and other reports, missing any measure not subject to biases. As well, whilst these traits are seen, typically, as immoral, none of the mentioned studies are explicitly related to morality. There has been no attempt thus far to understand thin slicing in direct relation to immorality. The need for future study to address the lack of overall empirical support, the lack of objective measures, and the lack of work explicitly related to morality is clear.

Whilst some research into thin slicing has adopted objective measures such as the existence of deception (Ambady and Rosenthal, 1992) and company profits (Rule Ambady, 2008), much of the research has had to rely on self and other reports, or a comparison between the two. This is not ideal for a number of reasons. The first of which is that self-report scales are inevitably subject to the way that the participant feels at a particular time, and thus important details may not be remembered or reported. Secondly, participants might either exaggerate or report in a more socially-desirable way to make themselves look better. Both of these potential biases weaken the measure. Further, other-reports are also subject to human expectations and biases, so a comparison between the two might not establish a relationship that is accurate and reflective of the true nature of the individual. There needs to be an objective comparison to help establish better methodological accuracy. This study aims to provide new grounds by measuring morality in relation to thin slicing, whilst simultaneously improving research methodology through the use of an objective measure of morality.

The present study.

This study will be assessing moral character evaluations through a thin slice interaction and their relation to an objective measure of morality. The primary scale this paper will use is an adapted version of the moral character scale from Sabo and Giner-Sorolla (2017), as it reflects the actual individual as a moral being, rather than other scales which typically measure moral reasoning (e.g. Carpendale, 2000). This scale has also been used, to previous success, and with high levels of reliability and validity. The scale was originally used for other-reports and will be adapted for use as a self-report scale alongside the original other-report. For example, the original scale was aimed at a character in a vignette, so the present study will manipulate this so it is aimed at either the participant themselves (as a self-report measure), or the participant s interaction partner (as an other-report measure). As previously mentioned, past research has fell short of having an objective measure of morality. To address this, the present study will use an objective measure which will be a task in which the participant should feel as though they have the option to cheat with impunity to win an Amazon voucher.

Method

Participants

A total of 94 of undergraduate students from the University of Kent participated, all being psychology students participating for course credit (N = 94 Mage = 19.32, SDage = 1.23). Ten participants were male (10.6), 83 were female (88.3%), and 1 participant identified as other (1.1%). The only exclusion criteria for this study was that participants must be at least 18 years of age, and must not have known their respective study partner before the study. Due to the nature of the study requiring interactions with another participant, two participants signed up for each time slot. The study was advertised as A study on relationships and feelings and, whilst not advertised, participants were told in the study information sheet that they would have a chance to win a £75 amazon voucher. All participants gave informed consent prior to the study starting, and full ethical approval was received from the Kent School of Psychology Research Ethics Committee (approval code 201815427989285195). Sample size was determined a priori using a G-power power analysis to obtain 80% power at p < .05 to be 100 individuals, which was then doubled due to the dyadic nature of the study.

Design and Procedure

The design of this study was correlational, specifically comparing participants own self-reported scores of moral character and their study partner s other-report scores of moral character to the extent to which they displayed immoral behaviour at the end of the study in the cheating task, the objective measure.

Procedural overview.

As a general overview of the procedure, participants were invited in pairs to participate in the study. Upon arrival, each participant completed a number of self-report scales. After this, they engaged in a 5 minute interaction with their partner. Then participants completed similar scales to before, but rating their interaction partner instead of themselves. Finally, as our objective test of morality, participants engaged in a maths test, where the more answers they got correct, the more entries into a raffle for a £75 Amazon voucher they got. Upon completion, a fake error message alluded to their answers not being saved, and participants were given the chance to verbally report their score, seemingly without the researcher knowing what they actually scored. This test should have allowed them an opportunity to cheat with seeming impunity. Incentivising cheating in this way is what was used as an objective measure with which to compare self and other reported moral character scales to.

Upon arrival, participants were shown to separate computer cubicles and instructed to follow the instructions on Qualtrics, which contained information about the study, as well as a consent form to both participate and be filmed during the study. Participants were required to create a unique participant ID so their results could be identified if they chose to withdraw their data. For each pair of participants, both needed to consent to being filmed for filming to take place. If one or both did not consent to filming, the study would have still gone ahead, just without the participants being filmed. Once both participants had completed this, and had the chance to ask any questions, the researcher manually progressed the Qualtrics survey into the self-report section of the study.

This section of the survey contained an adapted moral character scale from Sabo and Giner-Sorolla (2017 See Appendix A), alongside a scale measuring the big five personality traits (John, Naumann Soto, 2008 See Appendix B), and the sense of power scale (Anderson, John Keltner, 2012 See Appendix C). This section of the survey had a minimum time that it takes to advance to the next stage, ensuring both participants finished at approximately the same time.

Once both participants had finished the self-report measures, they were given information about the next section of the study and told to wait for the researcher to come and get them. The participants were then led to a small room where they were facing each other over a table. Once seated, they were instructed to have a conversation for five minutes. Prompts were given (e.g. Where would you like to travel? ), however participants were welcomed to speak freely about any topic. If consent was given by both participants, this interaction was filmed by two cameras, one pointed at each participant.

Once the five minutes were up, participants were led back to their respective cubicles to complete the other-report section of the Qualtrics survey, which contained the same scales used in the self-report section (minus the sense of power scale), but this time in reference to their interaction partner. As well as the scales already mentioned, there were a number of small questions relating to task specific judgements (i.e. How accurate was your impression of your partner? from 1 [Disagree strongly] to 9 [Agree strongly]), however these were used for a separate project, and have thus not been included in any analysis.

Following this, each participant individually completed a mental maths test. The test items were from Vohs and Schooler (2008 See Appendix D) and involved a long string of integers which had to be added and subtracted without the use of a pen and paper, or calculator. There were 20 of these to complete in a five-minute time-frame. Before starting this, participants were told, once again, that for each answer they get right, they would earn one ticket into a raffle to win £75 amazon voucher. This was a lie, and in fact all participants who completed the study were entered once into a raffle for the £75 amazon voucher regardless of how well they did in the maths test. This test was deliberately designed to be difficult to complete in five minutes, and past research has found that most participants are able to answer fewer than half of the items in a five-minute window (Gino, Ayal Ariely, 2013).

After the five-minute timer expired, participants were shown a screen where, in large red letters, it showed them how many of their answers were correct and the corresponding number of tickets they received for the raffle. This could not be advanced from for a few seconds, so a key assumption was made that all participants read this score. Once they advanced from this screen, participants were greeted with a fake error message (See Appendix E) which claimed that their answers were not saved. Most participants then came out and found the researcher to let them know what had happened, but after two minutes of no contact from the participant, the researcher checked up on them to see if everything was okay. After acknowledging the error, the researcher then asked participants to verbally report their scores. The expectation was that participants saw this as an opportunity to cheat without consequence as their scores were supposedly not saved. Their scores were in fact saved, and were used to find out the extent to which each participant cheated.

The researcher manually progressed the screen after recording the score, and participants were asked for a number of demographics such as age, year of study, and gender. This was followed by an on-screen debrief which explained the aims of the study, and also that there was no way of knowing personally whether participants cheated as their names are not associated with the data, which hopefully alleviated any feelings of shame the participants might have felt if they cheated or scored poorly on the test. Contact details of the lead researchers were shown at the end of the debrief, as well as a message asking participants to not talk about the deception used in the study to anyone they know studying psychology. They were then thanked for their participation and awarded credits.

Materials

As mentioned before, three self-report scales were used, with two of them then re-used as other-report scales, as well as a maths test. The first scale was the moral character scale from Sabo and Giner-Sorolla (2017) which, as mentioned, has been adapted from its original form for use as a self-report scale. The questions were phrased as Please rate how little or how much you agree with the following statements from 1 (strongly disagree) to 7 (strongly agree), with items such as I am loyal , and I am twisted . In the other-report section of the survey, the statements were rephrased to be directed towards the other participant (i.e. He/she is twisted ).

The second scale used was a measure of the big-five personality traits (John et al., 2008), which asked participants to rate from 1 (disagree strongly) to 5 (agree strongly) how much they see themselves as someone who . Items on the scale included Is full of energy and can be moody . When in reference to the study partner, the original statement was rephrased to I see my interaction partner as someone who , with the same items. This scale was included by others involved with the project, but was not the focus of our study, and only served as a filler item. Thus, subsequent analyses will not include this scale.

The final scale used was the sense of power scale (Anderson et al., 2012) which started In my relationships with others and got participants to rate from 1 (disagree strongly) to 7 (agree strongly) how much they agree with statements such as I can get people to listen to what I say . As with the previous scale, this scale was included by others involved in the project, but was not the focus of the present study, and thus will not be included in subsequent analyses.

Finally, the maths test used (Vohs and Schooler, 2008) involved participants answering a long string of integers that had to be added and subtracted without the use of pencil and paper or a calculator. For example 18 + 13 - 3 + 12 + 17 + 2 - 15 - 2 + 16 - 18 = ? . There were 20 of these to be completed within five minutes.

Results

Firstly, the moral character scale (both the self and other variants) was analysed for homogeneity. An exploratory factor analysis with an unweighted least squares extraction and a pro-max rotation indicated that the scale loads onto two distinct factors. One of the factors exclusively contained items that indicated a positive aspect of personality (i.e. warmth, reliability, etc). The other factor contained items that indicated negative aspects of personality (i.e. being twisted, abnormal, etc). There were no cross-loadings above .32, and this effect was present and nearly identical for the self-report scale and the other-report scale. From here, the scale was split into its constituent factors - moral character (positive traits) and abnormal character (negative traits). Once split, the scale had high levels of reliability (alpha > .7). For sake of simplicity, the variables will be renamed for this paper. For the self-report scales, SelfMoral will refer to the self-reported moral character ratings (positive items), and SelfAbnormal will refer to the self-reported abnormal character ratings (negative items). For the other-reports, OtherMoral will refer to the moral character ratings of the individual given by their interactions partner (positive items), and OtherAbnormal will refer to the abnormal character ratings of the individual given by their interaction partner (negative items).

Table 1 shows the descri ptive statistics for the scales split into its moral and abnormal parts. From it, it is clear that people view themselves and others as more moral than abnormal. It does show that people rate others as less abnormal than themselves as well. The standard deviations show a consistent pattern with the exception of SelfAbnormal, which is slightly higher than the others.

Table 1

Descri ptive statistics for the split moral character scale

Scale

N

Mean Score

Standard Deviation

SelfMoral

94

5.92

.67

SelfAbnormal

94

2.64

.94

OtherMoral

94

5.65

.61

OtherAbnormal

94

1.90

.79

A linear regression was performed to predict OtherMoral based on SelfMoral. A marginally significant regression was found (F(1,92) = 2.96, p = .088) with an R2 of .031. That is to say that there is a marginally significant relationship between an individual`s moral character ratings of themselves and their interaction partner s moral character ratings of that same individual. A second linear regression was performed to predict OtherAbnormal based on SelfAbnormal. A non-significant regression was found (F(1,92) = .13, p = .73) with an R2 of .001. This displayed no significance. Seemingly people had a much harder time predicting how abnormal their interaction partner was, than how moral their partner was.

The primary research focus was to measure whether participants could predict the cheating behaviour of their interaction partner, based upon a thin slice interaction, by means of the moral character scale. Analysis of this proved difficult, however, as only 5 participants cheated in the cheating task. Whilst little can be done of this, a regression was performed to predict cheating instances from OtherAbnormal. This was performed using the 5 cheating participants. A non-significant regression equation was found (F(1,3) = 2.92, p = .19) with an R2 of .49. Thus, abnormal character ratings from the participant s interaction partner did not significantly predict cheating behaviour.

A final regression was performed to predict cheating behaviour based on SelfMoral. Another non-significant regression equation was found (F(1,3) = 1.797, p = .273) with an R2 of .38. Thus, moral character ratings from the participant s interaction partner also did not predict cheating behaviour.

Discussion

The results of this study show that there is little to no significant relationship between self-reported morality, other-reported morality, and cheating behaviour (our objective test of morality). Specifically, participants self-reported Abnormal character (scale name SelfAbnormal) had no significant ability to predict their interaction partner s ratings of their abnormal character (scale name OtherAbnormal). However, participants self-reported moral character (scale name SelfMoral) had a marginally significant ability to predict their interaction partner s ratings of their moral character (scale name OtherMoral). Our primary research aim was to establish whether participants could significantly predict immoral behaviour in their interaction partner, which was measured using the cheating task. Two regressions were performed. The regressions showed that participants ratings of both moral character and abnormal character had no significant ability to predict cheating behaviour in their interaction partner. However, only five people cheated, which makes any result using the cheating task very statistically fragile. Due to the exploratory nature of the study and the lack of previous research into morality in a thin slicing context, there were no predictive hypotheses, and thus nothing to either support or reject. Overall though, the results are not suggestive of there being a significant thin-slicing effect for moral evaluations, nor are they suggestive of participants being able to predict immoral behaviour from a thin-slice interaction.

Theoretical Implications

As mentioned in the introduction, there has been a wealth of evidence for a thin-slicing effect for wide-range of behaviours and traits (e.g. Sparks et al., 2016 Kraus Keltner, 2009). There was also evidence from studies such as Fowler et al. (2009) and Holtzman (2011) that are relevant to morality. However, our results go against the tangentially related work done by such researchers by finding a non-significant result. The main difference between this study and past research into thin-slicing was the objective measure, which will be discussed further for its limitations. An alternative interpretation of the findings can be presented, however, despite the lack of significance. Seemingly from the results, individuals have no real way of discerning an immoral character within an initial meeting with them. This presents an divide between morality and the multitude of traits that have been demonstrated to be identifiable in the thin-slicing studies mentioned above. Morality could be an entirely separate trait to the others that have been studied, and might not manifest in everyday behaviour or appearance. Future research should aim to correct and repeat the present study to establish whether morality is truly separate from other traits identifiable from a thin-slice interaction.

Practical implications

Despite the non-significant results, some practical implications can be speculated upon. One such example is airport security and other security details. Whilst a large part of security involves physical detection of devices, another aspect is assessing suspicious behaviour. Our study seems to suggest that immorality is not detectable from a brief interaction, which could seriously affect how security is viewed. If this study were to receive more empirical support, serious questions about the efficacy of suspicious behaviour detection methods could be raised. This is however, jumping the gun a little bit. The present study only had five people who cheated, so comparisons can hardly be made with any confidence. Further empirical support is needed before any such implications can be made with any surety.

Limitations

It is important to consider why the present study produced non-significant results, to both help understand any methodological problems, and to aid future research in this field as a guide of what to do better. The main problem with the study is that only five people cheated, which made our research question almost unanswerable. Now, it could simply be that psychology students are an exceptionally moral group, who would rarely consider cheating. Given the statistics presented about how often people lie (Feldman, Forrest, Happ, 2002), however, alongside Vohs and Schooler (2008) who found a large number of their participants did cheat on the same test, this is unlikely. Rather, the methodology of the present study should be assessed for limitations.

There are a number of limitations that have arisen from the study that could have led to the low number of cheaters. The first explanation for the lack of cheaters is simply that the cheating task proved too hard. The majority of participants failed to achieve more than 3 correct answers, with a large minority failing to answer any correctly. This might have impacted the results due to participants feeling as though it would be obvious if they cheated. For example, if someone got 2 answers correct, they would likely think the researchers would know that they are lying if they said any reasonably high number. If they decided say they got 15 correct, it would be obvious, and the shame of being a known cheater might have over-rode any potential attempt to cheat. For this task to be effective, participants must feel like they can cheat without it being so obvious. Future research should consider a maths test that is slightly easier, so participants feel as though their cheating is less obvious, and thus feel more free to cheat. This could be accomplished by simply shortening the amount of integers in each individual questions by two or three numbers.

Another explanation for the lack of cheaters is that the fake error message that popped up might have looked too fake to be believable. The timeline of the study did not allow for much time to work on make the error message look real, with the result looking out of place on a modern, up-to-date computer. Whilst this did not prime any participants to guess the aims of the study, it might have made them more cautious, and thus less likely to cheat. Future studies should aim to use programs such as PsychoPy to make a more realistic pop-up, which could relieve any suspicions the participants have.

Another explanation for the lack of cheating is that participants might not have had the time to see it as an option. Anecdotally, the researchers working on this project all noted that many participants seemed flustered when asked to report their score, with a few saying that they did not even consider cheating as an option until after the study concluded. This could be due to participants not paying too much attention to the score they achieved when it was presented, as this was seemingly the end of the study and fatigue might have started to set in. A way of avoiding this could be to have the maths test at the start of the study, which might make participants more attentive to the problems and more aware of the fact that they could cheat. Whilst it is important to create a situation where participants feel as though they can cheat with impunity, it would only work if the participants can even realise that they could cheat.

A final explanation for the lack of cheating is researcher effects. There were 5 researchers working on the project, all students, which might have led participants to feel bad about openly lying to the face of someone in a similar position to them. This could be corrected by having the lead researchers to not be close in age or academic situation to the participants, which might make participants feel less guilty or ashamed of lying.

There are also a number of more general limitations to the study. The first of which is that, due to the dyadic nature of the study, many slots had to be cancelled due to one of the participants not arriving. Whilst this had little effect on the data, it wasted a lot of time that would have been better spent actually researching participants, or writing the project. The only way around this would be to collect videotapes of known cheaters, and use the tapes as the thin-slice. Indeed, this is why the present study recorded the interactions. This is dependent on having participants cheating though, so can only be solved once the above limitations are corrected. Another general limitation is that the sample was overwhelmingly female (88.3%), and were all students at a university in a Western culture. This leads to a very unrepresentative sample, which cannot be applied very far out of academic populations. The only way to avoid this is to sample from multiple countries, socioeconomic backgrounds, genders, and age groups. This is logistically very difficult though, particularly when there is little precedence from past research. This limitation is dependent on more suggestive results and support for the research question.

Future research

Future research should seek to correct the above limitations through the prescribed solutions. This would be useful to establish whether the present study is merely empirically weak and unable to measure the desired effect, or whether the present study is correct in its findings, and that morality cannot be identified from a brief interaction. This is a new area of research, and further studies are needed to push the boundaries of what is currently known.

Conclusion

In closing, the present study has demonstrated that morality is not detectable from a thin-slice interaction, and that objectively immoral behaviours are not predictable from this same interaction. While the results suggest morality may be unidentifiable from a brief interaction, there were a number of methodological issues that led to few participants cheating, such as the cheating task being too difficult, the fake error message not being believable enough, cheating not being an obvious option, and researcher effects. Alongside this, the sample was not representative of a wider population. These methodological issues make conclusions difficult to justify, and thus future research should aim to correct these before making any empirical conclusions with confidence. This study does not necessarily bode well for the future of the study of thin slicing and morality, but serves as an informative warning of what to avoid in future research, and should primarily be used as such.