The Affective, Personality, Cognitive and
Behavioral Correlates of Frequent Lying
Rony Halevy 6249590
Brain & Cognitive Sciences Msc. Program
Cognitive Science Center Amsterdam
University of Amsterdam
Supervised by:
Dr. Bruno Verschuere
University of Amsterdam
Co-assessor:
Dr. Shaul Shalvi
University of Amsterdam
July 2012
Abstract
Lying is claimed to be part of everyday social interaction, and some researchers claim that most people lie (DePaulo, Kirkendol, Kashy, Wyer & Epstein, 1996). However, a recent study found robust individual differences with respect to lying frequency, with most people reporting not lying at all, and a minority reporting very frequent lying (Serota, Levine & Boster, 2010). Little is known about these so called frequent liars – the extreme end of the lying frequency distribution. In the current study, we replicated the skewed lying frequency distribution, and further investigated the different correlates of frequent lying with respect to affect, personality, cognition and behavior. Using one mass survey and two laboratory studies, self reports of lying frequency were found to be related to psychopathic tendencies, a less negative explicit attitude toward deception and a higher chance of cheating in an experimental setting. The current study gives a better understanding of the profile of a frequent liar in our society.
Introduction
While condemned by society (Grubin, 2005) lying is claimed to be a part of everyday social interaction (DePaulo et al., 1996). Not all lies are egocentric, some are pro -social and some even promote the collective good (Fu, Evans, Wang & Kang Lee, 2008). In a study based on a daily-diary methodology, students reported telling on average two lies a day, while older community members reported an average of one lie per day. Lying was not viewed by participants as an unusual event, but rather as an everyday occurrence (DePaulo et al., 1996). The finding of approximately 1.5-2 lies per day was replicated using various methods (Vrij, 2000; Hancock, Thom-Santelli & Ritchie, 2004; George & Robb, 2008) and widely cited (Vrij & Mann, 2007). Some personality traits, such as manipulation, sociability and concern regarding self presentation, were modestly related to more lies (Kashy & DePaulo, 1996). According to the authors lies are also told by people that present high social skills, “liars seem to be able
participants in social life” (Kashy & DePaulo, 1996, p. 1048).
In light of the findings above it seems that most people lie on a daily basis, and that lying is part of most people’s close and casual relationships (DePaulo & Kashy, 1998). This finding was, however, recently challenged. A recent study, using a mass survey of 1000 participants, had shown tremendous individual differences with respect to lying frequency (Serota et al., 2010). In this study, the average amount of lies per day was quite similar to previous studies: 1.65 lies in average per day. The authors argued, however, that the mean is a misleading statistic as the data was heavily skewed. Most people (60%) reported not lying at all, and a small minority reported frequent lying. Half of the lies were told by 5.3% of the sample.
These 5.3% of the sample that report very frequent lying are an extremely interesting group. If indeed half of the lies in our society are told by these individuals, it is interesting to see
what distinguishes them from the rest of the population. There is hardly any empirical research on frequent lying. There are, however, some theories of a condition that may be considered at the extreme end of the lying frequency continuum: Pathological lying. While not recognized as a mental disorder by the DSM, it is widely accepted that some individuals are pathological liars – individuals who repeatedly and perhaps compulsively tell false stories (Poletti, Borelli & Bonuccelli, 2011).
It was claimed that pathological liars often don’t have control over their lies, “an inner
dynamic rather than an external reason is suspected” (Dike, Baranoski & Griffith, 2005, pp.
343). It has also been claimed that this population shows some brain structural abnormalities (Yang, Raine, Lencz, Bihrle, Lacasse & Colletti, 2005), perhaps leading to deficits in executive functions and theory of mind (Poletti et al., 2011). Very different and sometimes opposing predictions have been made in the cognitive, emotional, interpersonal and behavioral domains regarding pathological lying.
In the current study we wish to investigate the different correlates of frequent lying in a non-clinical population. Our hypotheses are based on both theories regarding pathological liars and studies of lying in everyday social interaction. Lying frequency is measured using two separated self report tools:
1. A Dutch version of the questionnaire used by Serota et al. (2010), referred to as the Serota
questionnaire.
2. A lying subscale of the Youth Psychopathic trait Inventory (YPI, Andershed, Kerr, Stattin & Lavender, 2002), referred to as the YPI LIE.
While both instruments use self report to measure lying frequency, they are quite distinct. In the Serota questionnaire, after a short paragraph emphasizing the role lying takes in everyday life, subjects are asked how many times they have lied in the 24 hours preceding the questionnaire. The YPI LIE, on the contrary, is administered in the context of psychopathic traits measurement, and is aimed at investigating general lying habits (i.e. “sometimes I lie for no reason, other than because it’s fun”).
Affect
It was suggested that pathological liars do not have negative emotions associated with lying (Grubin, 2005). We expect frequent liars to show a less negative explicit attitude toward lying. This lack of negative attitude associated with lying can be seen either as a predictor of frequent lying, namely that some people lie more since they do not consider deception to be a negative act, or as a way of justifying an existing behavior by adopting a less negative attitude toward it. Since implicit attitude measurement are claimed to be less controlled and less dependent upon cultural knowledge (Nosek, 2005; Nosek & Hansen, 2008), implicit attitude towards deception will also be measured. It was recently claimed that with respect to deception, implicit measures may be more informative than explicit measures; real cheating in the lab was predicted by a less negative implicit, but not explicit, attitude toward lying (Jung & Lee, 2009).
Cognition
In the Sheffield Lie Test (Spence et al., 2001), subjects are presented with different questions and are requested to lie in response to some questions and tell the truth in response to others. Using this task, it was shown that longer reaction times and higher error rates are associated with lie responses (Spence et al., 2001; Spence et al., 2004; Verschuere, Spruyt, Meijer & Otgaar,
2011). The difference in reaction times and error rates between truth and lie trials is used as an index of the cognitive load of deception, and it was suggested that “...truthful responding may
comprise a relative ‘baseline' in human cognition and communication” (Spence et al., 2004, p.
1755) and that the truth is always the dominant response (Johnson, Barhardt & Zhu, 2005).
Nonetheless, it was recently found that frequent lying makes lying easier in the short run (Verschuere et al., 2011). Subjects were presented with autobiographical questions, and were asked to lie in response to some questions and tell the truth in response to other. When the proportion of lie questions was higher, the cognitive load of lying was reduced. Given this finding, it is interesting to see whether a similar effect can be seen in long-term frequent liars, indicating that the truth is not always the dominant response. We expect frequent liars to show a smaller difference between lying and truth telling in a lab task.
We also explore two other cognitive predictions from the literature. A claim was made that pathological liars show a high verbal intelligence when compared to non-verbal intelligence (Healy & Healy, 1926; Yang et al., 2005; but see King & Ford, 1988), and that pathological liars show diminished moral reasoning abilities, perhaps leading them to a difficulty in distinguishing right from wrong (Healy & Healy, 1926). It is interesting to see if frequent liars that are not pathological, and are able to function in society (at least to the extent of studying in the university) will also show similar patterns.
Personality
Psychopathic traits were claimed to be related to pathological lying (Fullam, McKie & Dolan, 2009; Grubin, 2005). Indeed, lying is one of the 20 defining items in the most often used instrument to diagnose psychopathy (PCL-R; Hare, 1991). In addition, both Machiavellianism
and sociability were found to be correlated with self reports of lying in a daily diary paradigm with normal population (Kashy & DePaulo, 1996). However, while a manipulative personality trait was linked to lower guilt, more pleasure and less difficulty in regard to everyday lies, a sociable personality trait was linked to higher guilt while lying (Gozna, Vrij & Bull, 2001). We expect frequent lying to be associated with psychopathic tendencies. We are curious to see if a correlation will also appear between lying frequency and social potency.
Behaviour
Another interesting line of research focus on investigating deception and ethical behavior using tasks in which participants have a financial incentive to cheat. Using tasks that enable deception, it was shown that creative people can be more dishonest (Gino & Ariely, 2012), that people cheat more when they are not the only ones to profit from the deceptive act (Wiltermuth, 2011), and that self control depletion promotes unethical behavior (Gino, Schweitzer, Mead & Ariely 2011). A rather new theory, based on similar tasks, uses self concept maintenance to account for the decision people make to cheat (and the decision not to cheat). According to this theory, people like to think of themselves as honest but also want to enjoy some of the benefits of dishonest behaviors. The tension is resolved by behaving dishonestly enough to profit, but honestly enough to keep a delusion of integrity (Mazar, Amir & Ariely, 2008).
Using the Dice-Under-Cup paradigm, it was shown that when justifications are available, a higher cheating rate is seen (Shalvi, Dana, Handgraaf & Dreu, 2011), perhaps due to the ability to maintain a positive self-concept when cheating. Subject had to roll a dice inside a cup, report their score and get paid according to what they report. One group only rolled the dice once and reported the score, while another group rolled the dice three times but only reported the first roll.
The group that rolled the dice multiple times reported significantly higher results. The authors suggest that the multiple roll condition gives rise to cheating justifications. These justifications are claimed to enable people to feel honest while deceiving.
If indeed the decision when to deceive is based on the balance between the desire to profit and the wish to keep a positive self concept (Mazar et al., 2008), perhaps this equilibrium can be found in a different point for people who lie frequently. This difference can be due to the fact that frequent liars do not consider lying or cheating to be a negative act. We expect this lack of need in maintaining an honest self image to be manifested in two ways. First, we expect frequent liars to be more prone to cheating in the lab experiment. Second, we expect frequent liars to depend less on self justification in their decision to cheat.
Alternative Explanations for the Serota Distribution
Serota et al. (2010) used a very short survey to demonstrate the skewed lying frequency distribution. Claims have been made regarding the validity of surveys and the extra precautions needed when conducting mass surveys (Merckelbach, Giesbrecht, & Smeets, 2010). Given this argument, it is possible that something other than lying habits explains the results in the Serota questionnaire. Asking people to be honest about their lying habits seems hazardous. It is possible that people scoring high in the Serota questionnaire are malingering, answering randomly or simply deliberately choosing deviate answers. The studies by Serota et al. (2010), however, did not provide data to examine the authenticity and origin of the self reported lying frequency. The first aim of the current study is to replicate the results of Serota et al. (2010) in a platform that will enable more information about the lying frequency distribution, in order to rule out some alternative explanation for the skewed lying frequency distribution.
The Current Study
In the current study, we wish to replicate the lying frequency distribution found by Serota et al. (2010) in a platform that will enable us to collect wider information about these so called
frequent liars. In study 1, a mass survey will be used in which the Serota questionnaire will be
administered together with other personality and cognitive measurements. In this study, we hope to replicate the skewed distribution found by Serota et al. (2010). In addition, we expect to find a correlation between lying frequency and psychopathy and to explore other personality and cognitive correlates of frequent lying. In study 2, subjects will be invited to the lab based on their Serota score. This follow up study will focus on ruling out alternative explanations for high scores in the Serota questionnaire, and investigating the behavioural and affective aspects of frequent lying. We expect to find a correlation between lying habits and both implicit and explicit attitude towards deception, and a correlation between lying frequency and cheating in the lab. Given the exploratory nature of study 1 and 2, in study 3 we will focus on the behavioural aspect and try to replicate some of the findings. We again expect to find correlations between lying frequency, attitude towards deception and cheating rates.
Study 1
The Serota questionnaire was administered as part of a large battery of tasks and questionnaires to all first year psychology students at the University of Amsterdam. This platform enabled testing a large amount of subjects in a relatively short time. The goals of the study were:
1. Replicating the skewed lying frequency distribution found by Serota et al. (2010).
2. Correlating frequent lying with psychopathy and other personality traits and cognitive measures.
For this study, a Dutch version of the Serota questionnaire was used. In addition, the Youth Psychopathic trait Inventory (YPI, Andershed et al., 2002) was used to measure psychopathy and the Sheffield Lie Test (Spence et al., 2001) was used to measure the cognitive cost of deception. Moreover, subscales of different personality measurements were used; the Multidimensional Personality Questionnaire, Brief Form (MPQ BF, Patrick, Curtin & Tellegen, 2002), the Amsterdam Biographical Questionnaire. (Amsterdam Biografische Vragenlijst, ABV, Wilde, 1970), the Adjective Check-List. (ACL, Gough & Heilbrun, 1965) and the Big Five Personality Factors. ( Big 5, Goldberg, 1990). Finally, subjects’ verbal and non verbal intelligence were assessed.
Methods:
Subjects: N= 527 first year psychology students from the university of Amsterdam, 372 females.
Average age was 19.7 years (SD = 2.56). Not all subjects completed all questionnaires. In addition, three subjects were excluded from the Serota questionnaire, one subject was excluded from the YPI questionnaire and two were excluded from the MPQ questionnaire for submitting
the questionnaire too quickly (i.e. the duration was less than 2.5 SD from the mean duration). Hence, a different N is presented for each measurement.
Measurements: The following measurements were used in study 1.
Serota questionnaire. A Dutch translation of the questionnaire used by Serota et al. (2010) was used. After a short explanation, subjects were asked to indicate how many times in the last 24 hours they lied to different people (family, friends, people you know, people you might meet again and complete strangers) using face to face and non-face to face communication. Answers were summed to one total score.
YPI. (Andershed et al., 2002). The YPI is a self reported questionnaire designed to measure traits of psychopathic personality. The YPI consists of 50 items that can be divided into ten subscales: dishonest charm, grandiosity, lying, manipulation, callousness, un-emotionality, remorselessness, impulsiveness, thrill seeking and irresponsibility. Respondents are requested to rate to what degree each of the items apply to them on a 4-point Likert scale (1 = does not apply at all, 2 = does not apply, 3 = applies fairly well, 4 = applies very well). The YPI was found to predict subsequent violence and antisocial behavior (Dolan & Ronnie, 2006). The Dutch version of the YPI was found reliable and valid using a sample of non-referred Dutch and Belgian adolescents (Hillege, Das & Ruiter, 2010; Declercq, Markey, Vandict, & Verhaeghe, 2009) and a community based sample of adults (Uzibelo, Verschuere, van den Bussche & Crombez, 2010). The YPI total score was found to have a Cronbach’s α of 0.72, and the YPI LIE scale was found to have a Cronbach’s α of 0.73 (Hillege et al., 2010).
Sheffield Lie Test. (Spence et al., 2001). This task aims at testing the cognitive load of deception. 20 Autobiographical Yes/No questions were presented to the subjects on the computer screen (i.e. "Are you a student?"). Each question was presented once in yellow and once in blue.
Subjects were requested to answer the questions by pressing one of two keys on a keyboard, answering truthfully when the question is presented in yellow and lying when the question is presented in blue (or vise verse). The different in average latency and error rates between lie and truth trials was calculated and used as an indication for the cognitive cost of deception.
Multidimensional Personality Questionnaire- Brief Form. (MPQ BF, Patrick et al., 2002. See Eigenhuis, Kamphuis & Noordhof, 2012, for the Dutch version). This is a general self report measure of personality, measuring a range of discrete trait dispositions. In the current study, the social potency (e.g. being dominant, persuasive, strong and enjoying visibility) and self control (e.g. the ability to plan ahead, anticipate events and be cautious) scales were used. The social potency scale contains 14 items and was found to have a Cronbach’s α of 0.80. The self control scale contains 13 items and was found to have a Cronbach’s α of 0.74. In addition, a new attempt is now made to use the MPQ in order to evaluate psychopathic tendencies. The Psychopaty measurement based on the MPQ was also used (van Schagen & Verschuere, in press).
Big Five Personality Factors. (Big 5, Goldberg, 1990). The big 5 questionnaire is based on an attempt to describe personality using a structure based on 5 orthogonal factors: neuroticism, openness to experience, extraversion, agreeableness and conscientiousness. In the current study, the conscientiousness (e.g. being careful, responsible, organized and scrupulous, Roccas, Sagiv, Schawartz & Knafo, 2011) and agreeableness (e.g. being compliant, modest and cooperative, Roccas et al., 2011) subscales were used. The agreeableness scale contains 32 clusters, ranging in Cronbach’s α between 0.45 and 0.87. The conscientiousness scale contains 23 clusters, ranging in Cronbach’s α between 0.45 and 0.86 (Goldberg, 1990).
Amsterdam Biographical Questionnaire. (Amsterdam Biografische Vragenlijst, ABV, Wilde, 1970). The ABV is a self report personality measure, widely used in the Netherlands (Nijenhuis,
van der Flier & van Leeuwen, 1997). In the current study, the social conformity (e.g. obliging to social norms) scale was used. This sclae contains 23 items and was found to have a Cronbach’s α of 0.74 (Nijenhuis et al., 1997). In addition, the social desirability subscale (e.g. presenting an answering pattern that will appear favorable by others) of the questionnaire was used.
The Adjective Check-List. (ACL, Gough & Heilbrun, 1965). This measure consists of 300 adjectives and adjectival phrases and could either be completed as a self report measure or as an others-report measure. In the self report version, used here, participants are requested to check the adjectives that describe themselves. In the current study the communality (choosing items typically checked by people), self control (the extent to which self control is imposed and valued), positive adjectives and negative adjectives subscales were used. The favourable and unfavourable scales were found to have a Cronbach’s α of 0.95 and can be considered as measurements of social desirability and social undesirability, respectively. All other scales have an α coefficient of above 0.6 (Dy-Liacco, 2002).
Intelligence. Five intelligence subscales were administered. Two verbal subscales, vocabulary and verbal analogies, and three non verbal subscales, conclusions, series and calculation speed, were administered.
Results
The different measurements’ descriptive statistics and correlation with lying frequency are presented in table 1.
Serota. Lying frequency distribution is presented in figure 1. The lying frequency distribution
was skewed, SK = 4.773, SE = 0.11. An average of 2.04 lies per day was found (SD=3.85). 41% of the subjects told no lies, 51% told 1-5 lies, and 8% told 6 lies or more. 5% of the subjects told 40% of all reported lies.
Table 1.
Descriptive statistics and correlations with lying frequency – Study 1.
Measurement N M (SD) Cronbach’s α1 Pearson Correlation with Serota Pearson Correlation with YPI LIE Serota 497 2.04 (3.85) .67 - .19*** YPI LIE 519 8.01 (2.55) .72 .19*** -YPI 519 88.55 (17.73) .91 .19*** .65***
Sheffield Lie Test Lie effect Reaction times
456 117.22 (165.93) .44 -.03 .03
Sheffield Lie Test Lie effect Errors
456 5.16 (11.09) .23 -.04 -.03
Big 5 agreeableness 509 72.61 (9.43) . 80 -.14** -.29***
Big 5 conscientiousness 509 62.98 (10.99) .76 -.08 -.33***
ABV social desirability 501 27.89 (6.49) .72 -.05 -.21***
ABV social conformity 501 36.34 (8.84) .66 -.09* -.31***
ACL communality 516 42.37 (4.81) .79 -.19*** -.34***
ACL positive 516 259.69 (33.25) .94 -.08 -.11*
ACL negative 516 156.18 (35.45) .96 .12** .32***
ACL positive minus negative
516 103.50 (51.23) - -.13** -.30***
ACL self control 516 5.19 (9.12) . 67 -.04 -.14**
Intelligence 528 115.24 (9.59) - .01* .09*
Verbal minus non-verbal
intelligence 507 -11.81 (7.34) - -.09* .02 MPQ self control 507 6.72 (3.25) .72 -.10 -.30*** MPQ social potency 507 6.84 (3.28) .80 .09 .26*** MPQ psychopathy 507 11.65 (5.19) .73 .16*** .44*** Note * p < 0.05, ** p < 0.01, *** p < 0.0014.
1For ACL, ABV and big 5, Cronbach’s α calculations are based on a similar sampled gathered in 2010. For the
Sheffield lie test, split-half reliability was used instead of Cronbach’s α; lie effects in even trials were correlated to lie effects in odd trials.
(A) Occurrence (B) Number of Lies in the past 24 hours
Figure 1. Serota results distribution- Study 1. 41% report not lying at all, M = 2.04, SD = 3.85.
Sheffield Lie Test. Trials with reaction times faster than 300 ms and slower than 3000 ms were
recoded to 300 ms and 3000 ms, respectively. For each subject, a mean reaction time for truth trials (RT-truth) and lie trials (RT-lie), as well as amount of errors for truth trials and lie trials were calculated. 36 subjects were excluded for having a mean RT or error rate deviating in more than 2.5 SD’s from the mean, or for having less than 10 truth trials and 10 lie trials after exclusion of error trials.
RT and SD for each condition (lie/truth trials) were calculated (in ms). Results were M = 1511.7, SD = 198.37 for truth trials, and M = 1628.93, SD = 213.81 for the lie trials. A paired sample t-test revealed a significant difference in RT between lie and truth trials, t (455) = 15.09, p < 0.001, d = 0.6, indicating that in general, lying trials took more time.
Error rates for lie and truth trials were calculated. Results were M = 15.92%, SD = 9.41, for truth trials; M = 21.08%, SD = 10.99, for lie trials. A paired sample t-test revealed a significant difference in error rate between lie and truth trials, t (455) = 9.95, p < 0.001, d = 0.37, indicating that more errors were made in the lie trials.
Next, for each subject the average RT-truth was subtracted from the average RT-lie, to create the lie effect in reaction times. The same procedure was done with errors rates, creating the lie effect in errors. These variables were correlated with the other tested measures.
Correlations: Due to multiple comparisons, a Bonfferroni adjustment was performed. 19
variables were included in the correlation analysis, 18 were correlated to the Serota score and 18 to the YPI_LIE score, hence alpha was divided by 36, yielding an alpha of 0.0014. Correlations are presented in table 1.
In summary, the skewed lying frequency distribution was replicated. The Serota score was positively correlated with the YPI LIE scale, YPI total score and MPQ measurement of psychopathic tendencies and negatively correlated with the ACL communality scale. The positive correlations that were found between the Serota score and the total YPI score remained significant after excluding all YPI LIE items, r = 0.18, p < 0.001. The YPI LIE was positively correlated with the YPI total score, ACL negative adjectives and MPQ psychopathy measure and social potency, and negatively correlated with the Big 5 agreeableness and conscientiousness, ABV social desirability and social conformity, ACL communality and positive minus negative adjectives and MPQ self control. Controlling for social desirability did not change the correlations reported here.
Discussion
In study 1, using a self report questionnaire for lying frequency, we were able to replicate the skewed lying frequency distribution reported by Serota et al., (2010). The average amount of lies per day was similar to the average reported by DePaulo et al., (1996), but as oppose to the claim made by these authors, namely that everybody lies, it seems as if the majority of lies were told by
a small part of the population. A small group of participants (5% of the sample), here labeled ´frequent liars´, was responsible for 40% of all reported lies.
Importantly, the findings of Serota et al. (2010) were extended by including a variety of other measurements. First of all, lying frequency in the Serota questionnaire was associated with a self report subscale of lying habits (e.g. the YPI LIE scale). It might seem surprising that the correlation between the Serota score and the YPI LIE score (r = 0.19), two tools measuring lying habits, is not higher. However, as mentioned above, the two measurements are quite different, and might measure different kinds of lies. The YPI LIE scale, administered in a context of psychopathic trait measurement, might be more tuned to measuring self centered lies, whereas the Serota questionnaire, which includes a paragraph emphasizing the prevalence of deception in our society, might also measure less obvious pro social lies.
Second, a positive correlation was found between lying frequency according to the Serota questionnaire and general psychopathic tendencies (e.g. the YPI total score and the MPQ psychopathic trait measurement). The correlation between lying frequency according to the Serota questionnaire and the YPI total score remained significant after excluding the item of the YPI LIE scale, indicating that lying frequency is related to psychopathic tendencies in general. The fact that a positive correlation was found between the Serota score and the psychopathic tendencies implies that at least some of the variance in lying frequency might be explained by psychopathic tendencies. This correlation gives support to the finding of Kashy and DePaulo (1996) that people with a tendency to machiavellianism lie more frequently.
Kashy and DePaulo’s (1996) claim regarding the other personality type prone to lying, namely an individual with high social skills, is only partially supported by the current findings.
Social potency was positively correlated to the YPI LIE score but not to the Serota score. One possible explanation for this lack of correlation with the Serota score is that these socially potent individuals choose not to reveal their lying habits in such a straight-forward questionnaire. Another possibility is that socially potent individuals tend to tell more others-oriented and pro-social lies (Feldman, Forrest & Happ, 2002), and these lies are not reported in the Serota questionnaire. A better look at the kind of lies reported in the Serota questionnaire is needed in order to determine if that is indeed the case.
The Serota score was negatively correlated to communality scale of the ACL. This scale represents a normal answering pattern. Deviations from it (e.g. - low scores on this scale) can be considered as a measure of indecency (van Ginkel, Sijtsma, van der Ark, & Vermunt, 2010). This finding could be considered as an indication of a validity issue in the Serota questionnaire; perhaps people scoring high on the Serota questionnaire are not really frequent liars, but show general deviate response patterns. Is study 2 we will focus on alternative explanation for the Serota questionnaire, in an attempt to rule out this explanation. Another possibility is that this correlation is an indication of a real connection between frequent lying and a lack of communality, a real deviation from social norms.
The YPI LIE scale was positively correlated with social undesirability and negatively correlated with social desirability, communality, social conformity, agreeableness and conscientiousness. All of these variables represent a very certain profile of the frequent liar. If we return to the two possible types of frequent liars deduces from Kashy & DePaulo’s findings (1996), namely – manipulative liars and sociable liars, it seems as if the results here mainly support the first type.
No correlation was found between lying habits (in both the Serota questionnaire and the YPI LIE subscale) and lie effect in the Sheffield Lie Test (in both errors and reaction times). Verschuere et al., (2011) showed that in the short-run, frequent lying has an effect on the cognitive cost of deception. Here we found that frequent lying as a personality trait is not related to smaller lie effects. It was shown before that psychopathic traits are not related to reaction times and accuracy in the Sheffield Lie Test (Fullam et al., 2009). Perhaps the short term effect found by Verschuere et al. (2011) is not generalized to long term effects. It is important to note that the lie effects in the Sheffield Lie Test showed low split-half reliability, indicating that perhaps lie effects are not a stable variable for each individual.
Taken together, the results of study 1 provide us some information regarding the personality and cognitive aspects of frequent lying. We found that frequent lying is associated with psychopathic tendencies, but not with a reduced cognitive cost of deception. However, addressing the affective and behavioral aspects of frequent lying was not possible in a mass survey. Moreover, as mentioned above, it is possible that something other than lying habits stands behind the Serota score. Individuals scoring high in this questionnaire could be answering randomly, malingering, or simply trying to give deviate answers. We conducted another study, inviting subjects to the lab according to their Serota score, in an attempt to rule our alternative explanations for the Serota score as well as elaborate our knowledge regarding the different correlates of frequent lying.
Study 2
In study 1 we were able to replicate the lying frequency distribution found by Serota et al. (2010), and to show a link between a high score in this questionnaire and some other variables. In order to further investigate the correlates of frequent lying, in study 2 subjects from the same sample were invited to the lab.
To further investigate the affective correlates of frequent lying, we used both an implicit and explicit measurements of attitude towards truth and deception. The Feeling Thermometer (FT, Jung & Lee, 2009) was used to assess explicit attitude. To assess implicit attitude towards deception, a Dutch version of the Deception IAT (Jung & Lee, 2009, based on the work of Greenwald, Nosek & Banaji, 2003) was used.
To further investigate the cognitive load of deception, we used a modified version of the
Sheffield Lie Test (Spence et al., 2001). In study 1 we found no correlation between lying
frequency and lie effect in the Sheffield Lie Test. It is possible that frequent liars do not show a diminished lie effect in this task, but do benefit less from a manipulation aimed at making the task easier. To investigate this possibility, we used a different version of the task in which half of the trials were longer, enabling a preparation period that might make lying easier. The lie effect was expected to be smaller in the long trials (because subjects have time to prepare the lie). Based on study 1, no effect of lying frequency on the general lie effect was expected. However, we expected the effect of trial length to be smaller in frequent liars, indicating that these subjects do not need the extra time to prepare for lying.
In addition, in an attempt to investigate a possible moral deficit in frequent liars (as suggested for pathological liars by Grubin, 2005), we used the a Dutch version of Defining
Issues Test (DIT-2; Rest, Narvaez, Thoma & Bebeau, 1999; see van Goethem, van Hoof, van Aken, Raaijmakers, Boom & Orobio de Castro, 2011 for the Dutch version) as a measure of moral judgment. This tool is based on Kohlberg’s six phases of moral development theory. It is a widely used measurement for moral development (Xu, Iran-Nejad & Thoma, 2007).
In an attempt to replicate the correlation between lying frequency and psychopathy found in study 1, we administered the YPI again. In addition, another tool aimed at measuring psychopathic tendencies was used: The Psychopathic Personality Inventory Revised (PPI-R, Lilienfeld & Widows, 2005). The PPI-R is a self report measure of psychopathic tendencies. The questionnaire consists of two factors: “fearless dominance” and “antisocial impulsivity”, and the construct validity of the two factors structure was supported (Uzieblo, Verschuere & Crombez, 2007).
To evaluate the behavioral aspect of frequent lying, a modified version of the Dice-Under-Cup (Shalvi et al., 2011) was used. In this task, subjects are requested to roll a dice inside a cup, report their score and get paid according to what they report. It was found in a between subject design that a condition of multiple rolls, in which subjects repeatedly roll the dice but only report the first outcome, is related to higher scores, perhaps due to more justifications for deception (Shalvi et al., 2011). Here, we used a within subjects design of two long blocks. This design enables a classification of honest and dishonest subjects, based on statistics of the reported results. Following Shalvi et al.’s (2011) findings, we expected subjects to show a higher score in the multiple rolls block. However, we predicted this elevation to be smaller within the group of frequent liars, indicating that these individuals are less dependent upon justifications for dishonesty.
In addition, a battery of quality and validity measures was assembled, in order to rule out alternative explanations for the results in the Serota questionnaire (i.e. malingering, over endorsing, deviant responding, random responding). To assess the possibility that subjects scoring high on the Serota questionnaire are malingering, a Dutch version of the Structured Inventory of Malingered Symptoms (SIMS, Smith, 1997) was used. This tool is a self report measure aimed at screening malingering of psychiatric symptoms. The items present bizarre experiences and highly atypical symptoms. This measure was found to be a reliable, sensitive and consistent screening tool for malingering (Merckelbach & Smith, 2003). The Symptom Check-List (SCL 90, Derogatis, Lipman, & Covi, 1973; Derogatis 1975; see Arrindell & Ettema, 1986, for the Dutch version) was used in order to identify patterns of over-endorsement. In addition, validity scales of the PPI-R for deviant responses (DR), virtuous responses (VR) and inconsistency (IR-15 and IR-40), were also used. The DR scale consists of bizarre items and aims at detecting subjects that are responding carelessly, malingering or having difficulty comprehending the items. The VR scale contains positive items, and is aimed at detecting unlikely positive answering patterns. The IR-15 and IR-40 are calculated by comparing similar items that are repeated in the questionnaire, and are hence aimed at detecting subjects which are not consistence with their answers.
Methods
Sampling. Subjects were invited from all the Serota-score range. Low Serota scores were
under-sampled and high Serota scores were over-under-sampled, to ensure variance among the subject that arrive to the lab. From the initial sample (n = 497), we invited 40 subject scoring 0 (about 20% of the 0 sample), 20 subjects scoring 1-2 (about 12% of the 1-2 sample), 60 subject scoring 3-5 (75% of the 3-5 sample) and 43 subjects scoring 6 and above (100% of the 6+ sample).
In the low scores, males were oversampled to ensure an equal amount of males and females throughout the range. 44 subject responded, and 31 were scheduled for the experiment. The experimenter was unaware of the Serota result of a subject throughout scheduling procedure and testing.
An additional sample of 163 Serota questionnaires was gathered from other faculties of the University of Amsterdam. The average of this sample was 2.43. Of this sample, 24 subjects were invited to the lab and 10 were eventually scheduled.
Subjects. 41 students, 20 female. Average age was 20 years (SD = 2.12). Tasks
Extended Serota. An extended version of the Serota questionnaire was administered. After completing the first part (similar to Serota et al., 2010 and study 1), subjects were asked a few questions regarding the last lie they told: who they told the lie to, what was it about, did they get caught, how difficult was it to tell the lie and how emotional was it (on a scale from 0 = not at all to 100 = very difficult/emotional).
Affective Aspect. FT (Jung & Lee, 2009). Subjects were presented with 6 words related to deception and 5 words related to honesty, and were asked to rate the words as pleasant or unpleasant on an 11-points Likert scale (1=unpleasant, 11=pleasant).
Deception IAT (Jung & Lee, 2009). The concept deception was used as a target concept and the
concept honesty was used as its contrast. The affective dimensions were pleasant and unpleasant. 6 words were used in each category. The task was assembled of 7 blocks as presented in table 2. In each block, category labels (Deception/Honesty, Pleasant/Unpleasant) were presented in the top left and top right corners of the screen. Words were presented in the middle of the screen and
participant were asked to classify them as quickly and as accurately as possible using the “P” and “A” keys on the keyboard.
Table 2.
Blocks on Deception IAT
Block number
Block type Left key Right key
1 Practice Honesty words Deception words
2 Practice Pleasant words Unpleasant words
3 Congruent Honesty + Pleasant words Deception + Unpleasant words
4 Congruent Honesty + Pleasant words Deception + Unpleasant words
5 Practice Deception words Honesty words
6 Incongruent Deception + Pleasant words Honesty + Unpleasant words 7 Incongruent Deception + Pleasant words Honesty + Unpleasant words
Cognitive Aspect. DIT2 (Rest et al., 1999). Subjects were presented with three paragraph-long moral dilemmas and were asked to rate 12 statement as important/unimportant on a 5-points Likert scale. Next, subjects were requested to rank the 4 statements they found most important. Two subscales, representing two moral development levels, were calculated; the DIT4 and the DIT-P. The DIT4 reflects the selection of items supporting the established rolls, the legal system and formal organizational structure. The DIT-P score represents the use of post-conventional schemes to guide moral reasoning. This is the highest level of moral reasoning (see Rest, Narvaez, Bebeau & Thoma , 1999, for more elaboration on the different scales).
Sheffield Lie Test (Spence et al., 2001). Subjects were presented with autobiographical questions.
question was answered. Subjects were requested to lie when the “L” cue appear and tell the truth when the “W” cue appears (the “L” and “W” stands for the Dutch words for lie and truth, “Liggen” and “Waar”). The cues were presented either 200 ms or 2000 ms before the question, creating short and long trials. The task consisted of 160 trials: 40 long lie trials, 40 short lie, 40 long truth and 40 short truth trials, presented in a random order.
Interpersonal aspect. YPI. See method section of study 1.
PPI-R. The Psychopathic Personality Inventory Revised (PPI-R, Lilienfeld & Widows, 2005) is
a self report measure of psychopathic tendencies. The questionnaire is constructed of 154 items. Items are answered using a 4-point Likert scale (1= false, 2 = mostly false, 3 = mostly true, 4 = true). The total score was found to have a Cronbach’s α of 0.93 (Lilienfeld & Andrews, 1996). Items can be divided into eight subscales: machiavellian egocentricity, social potency, cold-heartedness, carefree planfulness, fearlessness, blame externalization, impulsive non-conformity and stress immunity. The social potency scale, which was also used in the current study, contains 24 items and was found to have a Cronbach’s α of 0.7 (Lilienfeld & Andrews, 1996).
Behavioral aspect. Dice-Under-Cup (Shalvi et al., 2011). Subjects received a cup with a dice inside it, and were informed that they will be paid according to their reported scores. Subjects had an opportunity to examine the dice, and were aware of the fact that they are the only ones that can see the results. The first block was a single roll block; subjects rolled the dice 30 times and reported every roll on the computer. The second block was a multiple roll block; subjects rolled the dice 90 times but only reported the first out of every three rolls.
Quality and Validity. SIMS. (Smith, 1997). Subjects were presented with 75 symptoms, and were asked to indicate which of the symptoms they experienced.
SCL-90. (Derogatis et al., 1973; Derogatis , 1975; Arrindell & Ettema, 1986). Subjects were
presented with 90 symptoms and were requested to respond to each symptom on a 1-5 Likert scale how much this problem bothered him in the past 4 weeks.
In addition, the DR, VR, IR-15 and IR-40 subscales of the PPI-R were used.
Results
The different measurements’ descriptive statistics and correlations with lying frequency are presented in table 3. One subject reported on the Serota questionnaire that her job is to lie to participants of the Dutch postcode lottery about their chances of winning. This subject reported lying 31 times in the last 24 hours. Given that this is a highly unusual situation, this subject was excluded from further analysis.
A strong, positive correlation was found between the Serota score in the present experiment (study 2) and the Serota score in the prior screening questionnaires (study 1), r = 0.7,
p < 0.001, indicating a good test-retest reliability for a 3-month period. Using the classification
suggested by Feldman, Forrest & Happ (2002), 67% of the last lies told were classified as self-oriented and 18% were classified as others-self-oriented (the rest were inconclusive). Four subjects reported getting caught. No differences were found between the subjects who got caught and the subjects who didn’t get caught with respect to the Serota score, t = 0.542, p = 0.59, the difficulty,
t = 1.43, p = 0.161 or the emotionality, t = 1.44, p = 0.236 of the last lie told. No differences in
any of the variables was detected between people whose last lie was others-oriented and people whose last lie was self oriented.
Table 3.
Descriptive statistics and correlations with lying frequency – Study 2.
Measurement M (SD) Cronbachs α2 Pearson Correlation with Serota Pearson Correlation with YPI Lie
Serota – amount of lies 2.50 (3.86) .6 - .21
Serota – Emotionality of the last lie
17.05 (23.07) / -.08 -.25
Serota – Difficulty of the last lie
28.33 (30.14) / -.31* -.28
YPI LIE 10.58 (2.84) .70 .21
-YPI total score 102.10 (16.63) .93 .31* .85**
PPI-R Social potency 50.48 (7.18) .82 -.01 .24
PPI-R total score 308.60 (32.82) .90 .29 .82**
FT 3.63 (1.89) / -.45** -.40*
Dice under cup total 3.70 (0.33) / .04 .20
IAT - D600 score 1.68 (0.38) / -.13 -.11
Sheffield lie effect -latencies
145.75 (148.17) .41 -.11 -.21
Sheffield lie effect errors
3.06 (3.55) .10 .03 .02
DIT – p score 34.00 (13.9) / -.07 .17
DIT – stage 4 score 28.83 (15.26) / -.19 -.45**
Note. * p < 0.05, ** p < 0.01.
2For the Sheffield lie test, split-half reliability was used instead of Cronbach’s α; lie effects in even trials were
Validity measures. SIMS. Total score was similar to the one found by Merckelbach & Smith (2003) with healthy control subjects. Two subjects presented a total score that is considered malingering according to the norm (=>14). These subjects scored 4 and 1 in the Serota questionnaire.
PPI-R. Two subjects scored higher than norm on the consistency scales, these two had Serota
scores of 0 and 1.
Correlations of the Serota score with the different validity were all lower than 0.23 and insignificant. A cut-off of 6 was used to classify subjects as frequent and infrequent liars (this cut-off is based on the calculation of 2.5 SD from the M in study 1). Independent sample t-tests were used to compare frequent and infrequent subjects with respect to the quality and validity measure. No significant difference was detected for the SIMS, t(38) = 1.33, p = 0.19, SCL-90,
t(38) = 1.37, p = 0.89, PPI-DR, t(38) = 0.22, p = 0.3, PPI-VR, t(38) = 0.38, p = 0.47, PPI-IR15, t(38) = 1.44, p = 0.2 and PPI-IR40, t(38) = 0.3, p = 0.19. In general these findings give no
indication of malingering, over reporting, random responding or deviant responding in the group of frequent liars.
Affective. FT. Subjects generally rated words related to truth as more pleasant than words related to deception, t (38) = 11.86.11, p < 0.001. For each subject, the average for deception words was subtracted from the average for truth words, yielding the FT score.
IAT. Based on the work of Greenwald et al. (2003), four test blocks were used (two compatible
and two incompatible). Latencies longer than 10,000 ms were excluded. For each block, the average latency for correct trial was calculated and all latencies of error trials were replaced by the block mean + 600 ms. For each subject, average latency for each of the four blocks was
calculated. Two differences were then calculated: block 6 minus block 3 and block 7 minus block 4. The differences were divided by the average SD for the two blocks used. Next, an average of these two differences was calculated, yielding the D600 score. A high positive D600 score represent a more positive attitude toward truth related words and/or a negative attitude toward deception related words.
The correlation between D600 and the FT score was not significant, r = 0.23, p = 0.16.
Cognitive. DIT. Two subscales were calculated, DIT4 and DIT P (See Rest, Thoma, Narvaes &
Babeau, 1997, for information regarding scoring method).
Sheffield Lie Test. Similar to study 1, trials with reaction times faster than 300 ms and slower
than 3000 ms were recoded to 300 ms and 3000 ms, respectively. Next, subjects with reaction times or error rates deviating in 2.5 SD or more from the mean for the condition were excluded. four subjects were excluded. A repeated measure ANOVA was used, with 2 (Deception: Lie vs. Truth) * 2 (Length: Short vs. Long) design, for both RT of correct trials and error rates.
For RTs, a main effect of deception was detected, F (1, 35) = 25.83, p < 0.001, as well as a marginally significant interaction of deception*length, F(1, 35) = 3.55, p = 0.068 (see figure 2). Contrasts were used to further investigate the effects. The lie effect was significant for both the short trials, t = 4.74, p < 0.001, and the long trials, t = 3.33, p < 0.005. The difference between short and long lie effects was marginally significant, t = 1.83, p = 0.068. The same analysis was conducted for Errors. A main effect of deception was detected, F (1, 35) = 10.68, p < 0.005, but no main effect for length or block was detected and no interactions.
To investigate the relation between the Serota score and YPI LIE score, and the Sheffield Lie Test, both variables were entered as covariates to a repeated measure ANCOVA. No main
effect was found for the Serota score, F(1, 35) = 0.38, p = 0.54, or the YPI LIE, F(1) = 0.37, p = 0.55, and no interactions were found. The same analysis was conducted with error rates. No main effect was found for Serota, F(1, 35) = 0.05, p = 0.83, or YPI LIE, F(1) = 1.48, p = 0.23. No interactions were detected.
Figure 2. Reaction Times and Errors in the Sheffield Lie Test, Study 2.
Behavioral. The average result on the Dice Under Cup task was 3.7 (SD=0.44) in the single-role
condition and 3.7 (SD=0.33) in the multiple role. The distribution in the single roll condition, χ²= 23.56, p < 0.001, as well as in the multiple roll, χ²= 27.2, p < 0.001, deviated from the expected chance distribution (see figure 3). There was no significant difference between the single role average and the multiple role average, t (39) = 0.07, p = 0.97, and a significant correlation was found between the single role average and multiple role average, r = 0.48, p < 0.005.
To investigate the relation between the Serota score and YPI LIE score and the Dice under cup task, both variables were entered as covariates in two separate Repeated Measure ANCOVA. No main effect was found for the Serota score, F (1) = 3.55, p = 0.94, or for the YPI LIE, F (1) = 7.42, p = 0.69. In addition, no correlation was detected between the total score in the Dice task and the Serota score or YPI score.
Figure 3. Results distributions in the single and multiple roll conditions.
Next, inspired by the work of Greene and Paxton (2009), we used subjects’ success rated to classify them to honest and dishonest3. Given that no difference was found between single and multiple roll blocks, the blocks were used together. Each subject’s total score 4was compared to chance. Taking an alpha of 10%, 12 subjects showed a score deviating from chance. One way ANOVA’s were used to compare the scores of the two groups (honest and dishonest subjects) in the different variables. For variables in which there was a significant difference between males and females, gender was entered as a covariate. A significant difference between honest and dishonest subjects was only found for the SCL-90, F(1,38) = 4.9, p < 0.05. In addition, a non-significant trend of dishonest subjects scoring higher on the Serota questionnaire was detected.
To sum, with respect to the affective aspect, a negative correlation was found between the Serota score and the FT, as well a negative correlation between the YPI LIE scale and the FT score, indicating that frequent liars show a less negative attitude towards deception. With respect to the interpersonal aspect, a positive correlation was found between the Serota score and the
3Since the total score is used, very lucky subjects will also be classified as dishonest.
4An attempt was made to classify subjects as honest and dishonest by comparing each subject’s distribution to
YPI total score, as well as a marginally significant positive correlation between the Serota score and the PPI-R total score. A positive correlation was also found between the YPI LIE and YPI total score, and between the YPI LIE and the PPI-R total score. No correlation were found between lying frequency and social potency. With respect to the behavioral aspect, no correlation was found between lying frequency and cheating rate.
Discussion
A high positive correlation was found between the two measurements of Serota questionnaire, indicating good test-retest reliability. This finding is not trivial. Given that the subjects are requested to answer every time about the lies told in the 24 hours before the questionnaire is administered, answers are likely to vary between occasions. It is plausible that some individuals lie a lot in one day, but not that much in another day. The strong correlation indicates that in spite of that, subjects’ reports regarding lying frequency are quite stable.
No correlation was found between the Serota score and any of the validity measures. Doubt has been raised regarding the usefulness of asking participants to be honest regarding their
lying habits (Kashy and DePaulo, 1996). The fact that no correlation was found between the
Serota questionnaire and malingering, random answering tendency, over endorsement tendency and deviate response tendency is a reason for optimism. Obviously, the question of whether people are honest about their lying habits remains open, yet the fact that none of these measures are correlated to the Serota score gives support to the validity of the tool.
The correlations between the Serota score and the YPI total score, found in study 1, were replicated in the current study. In addition, a marginally significant correlation was found between the Serota score and the PPI-R total score, another self report tool for measuring
psychopathic traits. The YPI LIE subscale was also correlated with both the YPI total score and the PPI-R total score. It seems as if people who lie frequently are more prone to psychopathy. Psychopaths are claimed to show high levels of deception (Hare, 1998), and it is hence interesting that people who score high but within the normal range on a psychopathy questionnaire are also more prone to deception.
However, no correlation was found between social potency and lying frequency. The correlation between social potency and the YPI LIE subscale, found in study 1, was not replicated. This lack of replication can be due to the power difference between the studies. Another possible explanation is the different tools used to measure social potency. In study 1, we used the social potency scale of the MPQ, a general personality measurement. This subscale is constructed by the following traits: enjoys visibility, dominance, likes to be in charge, persuasive and strong (Patrich et al., 2002). In study 2 we used the social potency subscale of the PPI-R, a tool designed to assess psychopathic tendencies. This subscale is focused on the ability to influence others (Edens, Norman, Poythers & Watkins, 2001).
In the discussion of study 1, we suggested that the type of lies reported in the Serota questionnaire (as opposed to a diary study, as used by Kashy & DePaulo, 1996), might be the reason for the lack of correlation with social potency. It was shown before that people that are trying to be likable are more prone to telling others-oriented lies (Feldman et al., 2002). We suggested that these others-oriented lies are perhaps not measured by the Serota questionnaire. In the current study, we asked our participants to elaborate regarding the last lie they told. Based on these reports, it seems as if the Serota questionnaire is measuring both self oriented and others oriented lies.
A negative correlation was found between both lying frequency measures and the FT, an explicit measure for attitude toward lying. It seems that people who lie frequently show a less negative attitude towards lying in an explicit measure. The direction of causality is not clear. It is possible that frequent liars lie more because they explicitly find lying to be less negative. Another possibility is that these individuals report a less negative attitude towards deception as a mean of justifying their behavior. The fact that no correlation was found between lying frequency and an implicit measure of attitude towards lying can be seen as a support for the second explanation.
The Serota score and the YPI LIE scale had no effects in the Sheffield Lie Test. It seems like lying frequently did not reduce the cognitive cost of deception. One possible explanation is that, as claimed by Johnson et al. (2005), the cognitive cost of deception is malleable. Another explanation could be that there are many differences between Sheffield Lie Test and lying outside the lab, steaming from differences in motivation, initiation, risk, topic, etc. Perhaps people that lie a lot in everyday life do become better at lying, but the difference cannot be detected with this sort of laboratory task. In addition, the split-half reliability of the Sheffield Lie Test lie effects was again rather low. This gives us another support to the idea that lie effects in this task are nor very stable within participants.
With respect to the behavioral aspect of frequent lying, as oppose to our prediction, no effect was found for the Serota questionnaire, nor for the YPI LIE subscale, in the Dice task. In addition, as oppose to previous findings (Shalvi et al., 2011; Gino & Ariely, 2012; Shalvi, Eldar & Bereby-Meyer, in press), no difference was found between the multiple roll condition and the single roll condition. One possible explanation could be the effect of order; previous studies used the multiple and single roll conditions as between subjects variables. Here, due to the interest in
individual differences and the relatively small sample of frequent liars, we used a within subjects design. In addition, and again due to the limited sample, order was not counter-balanced. All subjects completed first the single roll and then the multiple roll condition. Moreover, in contrast to previous studies, we used a design of repeated dice rolls, to enable classification of honest and dishonest subjects. Given this different design, it is possible that subjects began to feel bad about cheating, and started limiting their over-reports after a while. This effect could override the effect of multiple versus single roll condition.
Given this limitation it is hard to draw conclusions regarding the effect of frequent lying on the decision to cheat on a task. In addition, using this sort of task the only classification of honest and dishonest subjects is based on statistics, and is hence falsely accusing some honest yet lucky participants. We decided to run another study, using a simpler design for the Dice task and adding another task that enables a clear-cut discrimination, in order to further investigate the behavioral correlates of frequent lying. In this study, we were also interested in the negative affect related to deceptive behavior.
Study 3
In study 2 we found a correlation between lying frequency and a less negative attitude towards deception. In addition, we found no indication for any relation between the Serota questionnaire and different validity measures, and were able to replicate the correlation between lying frequency and psychopathy found in study 1. With respect to the behavioral aspect, as oppose to our predictions, no difference was found between the single and the multiple roll condition. This fact made the interpretation of the behavioral data somewhat problematic. We decided to run another study in the lab, using a simpler design of the Dice Under Cup task, in which only one long multiple rolls block is administered. Using this design, we expected to find a correlation between lying frequency reports and cheating in the lab.
In addition, since the classification to honest and dishonest subjects in the Dice task is based on success rates, lucky subjects are falsely accused in being dishonest. To deal with this limitation, we added two other ways to classify subjects as honest or dishonest. First, we added a dishonesty questionnaire at the end of the experiment, after the payment, in which subjects were asked whether or not they cheated in the Dice task. Furthermore, we added another task, the Words Task (Wiltermuth, 2011), in which a clear cut classification can be made between honest and dishonest subjects. In this task subjects are requested to solve word jumbles in a certain order, and to report how many they solved in a row. The third word in unsolvable, so any report of three and above can be considered as dishonesty. Using these two new classifications, we expected to find that frequent liars are more prone to cheating in the lab.
Moreover, we suggested in study 2 that subjects didn’t cheat as much as they could because a negative affect was elicited by continuous deception. Here, we added a measurement
of negative and positive affect, the Positive And Negative Affect Schedule (PANAS, Watson, Clark & Tellegen, 1988), immediately after the Dice task. We expected high scores in the Dice task to correlate with negative affect, as previously found by Shalvi et al. (in press) for some conditions.
Method
Sampling: From the initial sample, we invited 50 subjects scoring 0 (~25% of the sample), 50 subjects scoring 1 (~40% of the sample), 63 subjects scoring 2 (100% of the sample), 72 subjects scoring 3-5 (100% of the sample) and 35 subjects scoring 6 and above (100% of the sample). Subjects who participated in study 2 were not invited again. In total, 270 invitations were sent. 33 subjects responded to the invitation and 31 were eventually scheduled. An additional sample of 20 subjects was recruited without screening, from the psychology department and other faculties in the University of Amsterdam.
Subjects: 51 subjects participated in the study, 37 females. Average age was 21.1 years (SD = 6.16). The experimenter was unaware of subjects’ Serota score throughout the scheduling and testing phases.
Tasks: The expended Serota, YPI, FT, DIT and SCL, all used in study 2, were used in the same manner.
Dice Under Cup. A modified version of the Dice-Under-Cup task was used. In this version, a
long multiple roll block was used. In each trial, subjects were requested to roll the dice three times inside the cup, but only report the first result. Each subjects repeated this 60 times.
PANAS. (Watson et al., 1988). Subjects were presented with 20 words related to positive and
negative affect, and were requested to indicate to what extent they feel this way on a scale of one to five.
Scrambled Words. (Wiltermuth, 2011). Subjects were presented with 9 scrambled words in
Dutch, and had 5 minutes to solve them (the words used were potlood, bloem, taguan, sokken,
steen, kleur, rusten, koekhe and ijsberg). Subjects were paid an extra 0.5 euro for each word they
solved, but the instructions indicated that the words should be solved in the order in which they appear, noting “if you successfully unscramble the first three words but not the forth you will only be paid for the first three words, even if you successfully unscramble the fifth, sixth and seventh words”. When the time was up, subjects were requested to indicate how many words they were able to solve in a row. Subjects were informed that the number they indicate will be used to calculate the payment.
Crucially, and unknown to the participants, the third word was unsolvable. This word could only be unscrambled to spell “taguan”, a large nocturnal flying squirrel. The first, second and 4-8 words were highly solvable. These words were selected from a larger sample of words, after piloting the words with 10 native Dutch speakers. The 8 words used were all solved by all pilot subjects, and the unsolved word was not solved by any of the pilot subjects. The ninth word was solvable but difficult (was only solved by four out of the ten pilot subjects).
Given this design, any reported score of three and above could be considered as cheating, enabling a clear cut between honest and dishonest people in this task.
Dishonesty Questionnaire. After receiving their payment, subjects were asked to answer the
“We are interested in the way people perform in the dice task. The money you earned is yours and will not be taken away. We want to ask you a short question regarding your performance in this task. Out of the 60 times you had to report your score, in how many times did you report a
higher score than you actually got? “
Results
The different measurements’ descriptive statistics and correlations with lying frequency are presented in table 4.
Serota: The mean was not significantly different from the mean in study 1, t = 0.62 , p = 0.53. Using the classification suggested by Feldman et al. (2002), 70.5% of the last lies told were classified as self oriented and 19.5% were classified as other-oriented (the rest were inconclusive). Comparing the two groups (self oriented vs. other-oriented last lie), the only significant difference found was that subjects in the self-oriented group were more impulsive, t = 2.59, p < 0.05.
Affective aspect. FT. Subjects generally rated words related to truth as more pleasant than words related to deception, t(49) = 20.53, p < 0.001.
Behavioral. Dice task. The average score in the Dice task was 3.63 (SD = 0.3). The distribution of responses deviated from chance distribution, χ²= 23.73, p < 0.001 (see figure 4). A positive correlation was found between the Dice score and the PANAS negative affect score, r = 0.34, p
Table 4
Descriptive statistics and correlations with lying frequency – Study 3.
M (SD) Cronbach’s
α Pearson Correlation with Serota
Pearson Correlation with YPI LIE
Serota - amount of lies 2.27 (2.68) .61 .27*
Serota – Difficulty of the last
lie 25.61 (31.3) - -.17 -.11
Serota – Emotionality of the
last lie 20.67 (30.12) - -.12 -.18
YPI LIE 7.86 (2.1) .55 .27*
YPI total score 89.76 (15.75) .88 .30* .43**
Dice under cup Mean 3.63 (0.3) - .48** .15
Honesty questionnaire 2.55 (7.1) - .33* .31*
Words task 2.8 (1.9) - .21 .06
FT 6.4 (2.2) - -.25 -.22
PANAS positive affect 22.75 (5.41) .82 -.05 .00
PANAS negative affect 16.08 (4.27) .69 .09 .23
SCL total score 122.96 (35.49) .94 .26 .54*
DIT – p score 34.25 (16.68) - -.21 -.05
DIT – stage 4 score 33.27 (15.8) - -.03 -.01
Figure 4 . Results distribution in the Dice Under Cup (Study 3). The line represents chance
distribution.
Dishonesty questionnaire. 12 subjects reported some degree of over-reporting (e.g. dishonesty)
in the dishonesty questionnaire. The mean over-reporting rate admitted was 2.55 (SD = 7.09). A highly significant positive correlation was found between the Dice score and the dishonesty questionnaire, r = 0.49, p < 0.001, indicating that people who admit over-reporting in the Dice task generally got a higher score.
Next, similar to study 2, subjects were divided to honest and dishonest according to their dice score. Taking an alpha of 10%, 15 subjects showed a score deviating from chance. 8 of these subjects also admitted over-reporting in the honesty questionnaire. Significantly more subjects in the dishonest group admitted over reporting in the dishonesty questionnaire, χ² = 4.58,
p < 0.05. As another mean of classification for honest and dishonest we used the dishonesty
questionnaire: all subjects admitting over reporting were classified as dishonest.
One way ANOVA’s were used to compare the scores of the two groups (honest and
dishonest subjects) in the two different variables (deviating total score and admitting
over-reporting). For variables in which there was a significant difference between males and females, gender was entered as a covariate. For the total score discrimination, a significant difference was
