Stretching the Truth to Appear Honest to Yourself or Others?
A Meta-Analysis on the Role of Implausibility Aversion in Dishonest Behavior Inge A.M. ter Laak
University of Amsterdam 23 June 2017
Student number: 6058787 Number of ECTS: 24
Research Master’s supervisor: Eftychia Stamkou, University of Amsterdam
Abstract
Dishonesty is a common and costly problem of society. According to the theory of self-concept maintenance, people tell small “truth stretching” rather than severe lies to maintain an honest self-concept. To the contrary, neoclassical economic theory proposes that dishonesty is determined by weighing the benefits of lying against the probability and consequences of detection. Experimental paradigms like the die game and matrix task are commonly used to study dishonesty and generally support the theory of self-concept maintenance. However, these studies often do not account for plausibility concerns that might prevent people from telling implausible lies. In this meta-analysis, based on Gerlach, Theodorescu and Hertwig (2017), including 84 die games and 43 matrix tasks, it was studied whether “implausibility aversion” predicted dishonesty. The results showed that “implausibility aversion” predicted maximizing when all paradigms were combined, but not within all paradigms separately. Possible
explanations for these findings were investigated. Finally, it was discussed how these insights can be used to diminish societal dishonesty.
Acknowledgements
I would like to thank my external supervisor Philipp Gerlach for offering me an internship position at the Max Planck Institute for Human Development in Berlin, for including me in a very interesting project and supporting me during and after my internship. The Max Planck Institute was an inspiring working environment that offered many opportunities to gain scientific knowledge and research experience. I would also like to thank my supervisor Eftychia Stamkou for enthusiastically accepting to supervise me and supporting me during my internship abroad. Finally, I would like to thank the Jo Kolk foundation for supporting me and other women working in science.
Table of Contents
Abstract……… ii
Acknowledgements……….. iii
Table of Contents………...……….. iv
List of Tables……… vi
List of Figures……….. vii
List of Appendix Tables………... viii
I. Stretching the Truth to Appear Honest to Yourself or Others? A Meta-Analysis on the Role of Implausibility Aversion in Dishonest Behavior……….... 1
II. Method………. 4 Data collection………... 4 Criteria……….. 4 Measures………... 4 Implausibility aversion………. 5 Maximizing………... 5 Data Analysis……….………... 6 Robustness……… 6 Heterogeneity………... 6 Main analyses……… 7 III. Results……….. 7 Confirmatory Analyses………. 7 Heterogeneity……….... 7 Robustness……… 9 Meta-regression……… 12 Exploratory Analyses……… 14 IV. Discussion……… 16
Limitations and Directions for Future Research……… 18
Conclusion……… 21
References………. 23
Footnotes………... 26
List of Tables
Table 1. Mean differences (MD) of average maximum overreporting for all paradigms combined and the separate paradigms with number of studies k, 95% CIs,
z-values, and p-values………... 9
Table 2. Adjusted mean differences (MD) of average maximum overreporting by Trim and Fill for all paradigms combined and the separate paradigms with number of studies k (adjusted number of studies by Trim and Fill), 95% CIs,
z-values, and p-values……… 12
Table 3. Regression coefficients of the meta-regression model of average maximum overreporting regressed on probability of obtaining the highest outcome (prob. max) for all paradigms combined with standard errors (SE), 95% CIs,
List of Figures
Figure 1. Forest plots of average maximum overreporting for (1) the single die game, (2) the multiple dies game, and (3) the matrix task. Studies with larger squares had a lower standard error and had therefore larger weights in the random effects model. The horizontal bars indicate the corresponding 95%
CI………. 10
Figure 2. Funnel plot of mean differences against their standard errors. The plot includes both the observed studies (grey dots) and the “unpublished” studies imputed by Trim and Fill (white dots). The dotted triangle
represents the pseudo-CI, the area in which 95% of the studies would fall in absence of publication bias. The dotted vertical line represent the mean adjusted by Trim and Fill. The white contour represents the region in which the studies would need to be in order to achieve the 5% confidence
level………. 11
List of Appendix Tables
Table A1. Overview of the included single die games with the probability of honestly obtaining the highest outcome (prob. max) and the reference of the corresponding article. The code corresponds to the forest plot in Figure 1. For concision, a reference is only mentioned once for an article
and the same reference is subsequently indicated by (-)………... 9 Table A2. Overview of the included multiple dies games with the probability of
honestly obtaining the highest outcome (prob. max) and the reference of the corresponding article. The code corresponds to the forest plot in Figure 1. For concision, a reference is only mentioned once for an article
and the same reference is subsequently indicated by (-)………... 12 Table A3. Overview of the included matrix tasks with the probability of honestly
obtaining the highest outcome (prob. max) and the reference of the corresponding article. The code corresponds to the forest plot in Figure 1. For concision, a reference is only mentioned once for an article and the
Stretching the Truth to Appear Honest to Yourself or Others?
A Meta-Analysis on the Role of Implausibility Aversion in Dishonest Behavior
The widespread idea that everybody lies on a daily basis has been contradicted by recent self-report findings that most individuals lie occasionally and only a small part of the population tells the majority of lies (Serota, Levine, & Bolster, 2010). This illustrates that most people seem lying averse (Erat & Gneezy, 2012) and more inclined to tell “truth stretching” lies as opposed to severe lies, even when this results in a lower payoff (Shalvi, Dana, Handgraaf, & De Drue, 2011). This “truth stretching” or partial lying is also reflected in several everyday issues like tax misreporting or dodging. As pointed out by Erat and Gneezy (2012), this seemingly ‘innocent’ lying behavior can have a major impact on society when a considerable amount of people
engages in it. In order to take countermeasures, it is therefore important to determine why people lie and which circumstances promote dishonesty.
“Truth stretching” has been explained by the theory of self-concept maintenance (Mazar, Amir, & Ariely, 2008), which holds that individuals are motivated to maintain an honest self-concept. Lying can negatively affect this, which makes it psychologically (or intrinsically) costly (Abeler, Becker, & Falk, 2014). To maintain a positive self-concept and still benefit from lying, individuals stretch the truth to a point that they can still categorize their behavior as honest in general (Mazar et al., 2008). In this view, the limit up to which a behavior is still seen as honest is malleable and depends on the context. For example, tax misreporting to avoid paying an extra 100 dollars may be more malleable than directly stealing 100 dollars. An increasing body of evidence supports this theory by showing that individuals often only partially lie (for an overview see Rosenbaum, Billinger and Stieglitz, 2014) or only lie to the extent to which self-justifications are available (Shalvi et al., 2011).
To the contrary, the neoclassic economic model assumes that unethical behavior is the result of a rational cost-benefit analysis (Becker, 1986). The model proposes that people are more likely to behave dishonestly when the probability of detection and the severity of
punishment are low, and the rewards are high. However, evidence on “truth stretching” seems to contradict this preposition. People commonly do not maximize their payoff even when this causes them to miss out on a higher reward (Shalvi et al., 2011). Does this mean that the economic model of dishonesty can be disregarded? Kajackait and Gneezy (2015) argue against this by showing that many participants did engage in a simple cost-benefit analysis, but only when the probability of detection was low. This shows that “truth stretching” behavior might actually be motivated by reputational concerns rather than maintaining an honest self-concept; a small lie is more plausible than a ‘maximal’ lie and will therefore go more likely undetected. In line with this, Gamliel and Peer (2013) argue that many psychological studies on dishonesty do not account for the risk of detection, although people are aware of this risk in daily life. This awareness might also be present in studies on dishonesty, especially because participants are often tempted to lie with other people present (e.g. participants, experimenter) who could potentially detect the lie. Participants might therefore avoid lying to such an extent that their claim becomes implausible. However, most studies on dishonesty did not account for this “implausibility aversion”, which might have influenced the results on the levels of dishonesty.
The aim of the current study is to assess whether “implausibility aversion” is a neglected factor in psychological research on dishonesty. To this end, a meta-analysis is conducted to study if “implausibility aversion” predicts maximal dishonesty (maximizing) in two common
experimental paradigms: the matrix task and the die game. The meta-analysis is based on the dataset of Gerlach et al. (2017), and included 68 die games with a single die (single die game),
16 die games with multiple dies (multiple dies game), and 43 matrix tasks,. In the matrix task participants are required to find two numbers in matrices of 12 two-digit numbers (e.g., 5.31) that sum up to exactly 10.00 (Mazar et al., 2008). The time given is insufficient to solve all matrices and participants are paid according to the number of matrices they claim to have solved. Lying is not directly observed, but one of the numbers in the matrix is a hidden id, which makes it possible to compare each participant’s actual and reported performance. On average,
participants reported solving 40-60% of the matrices whereas they actually solved 35%. This non-maximal overreporting is commonly explained in terms of self-concept maintenance. However, participants might actually refrain from reporting all matrices because they consider this to be implausible.
In the die game participants privately roll one die and report the number of pips to the experimenter (Fishbacher & Föllmi-Heusi, 2013). The experimenter then pays participants according to their reports (e.g., reporting 1 paid $1, reporting 2 paid $2, etc.). Participants tend to partially lie by reporting higher pips (e.g., 4 instead of the rolled 2) but not the maximum (e.g., 6 regardless of what they rolled). The authors reasoned that participants shied away from reporting the highest pip to avoid appearing greedy or dishonest. However, similar to the matrix task, participants might have reported a more plausible outcome. It is therefore hypothesized that maximizing will decrease when “implausibility aversion” increases.
Still, reporting 6 pips in the single die game is a “smaller” lie than reporting 20 matrices in the matrix task.¹ It is therefore hypothesized that participants show more “implausibility
aversion” in the matrix task compared to the single die game. However, when the number of dies increases, the probability of rolling the maximum outcome decreases and people might become more hesitant to lie maximally. In line with this, Suri, Goldstein and Mason (2011) showed that
dishonesty levels and maximizing decreased substantially in the die game with 30 dies compared a single die. Therefore it is hypothesized that participants show more “implausibility aversion” in the multiple dies game compared to the single die game.
Method Data Collection
The meta-analysis was based on a subset of the data of Gerlach et al. (2017). Through an extensive Google Scholar search (scholar.google.com) the authors sought to integrate all experiments from articles that cited the seminal articles of Mazar et al. (2008), and Fischbacher and Föllmi-Heusi (2013) who introduced the matrix task and the die game, respectively. Both published and unpublished studies (e.g., working papers, discussion papers, and theses) were integrated to diminish the effect of possible publication bias.
Criteria. The authors only integrated one-shot, fully anonymous, incentivized
experiments, in which misreporting could not be sanctioned. In addition, only experiments for which the outcome of the game was not predetermined were included. Finally, only experiments in which the number of possible outcomes was higher than the number of participants were included, to avoid inclusion of underpowered studies. For all articles that met these criteria, Gerlach et al. (2017) contacted the corresponding authors to work with the primary data. Only the studies for which the primary data were shared were included in the current meta-analysis, because these were required to estimate the measures of “implausibility aversion” and
maximizing. Based on these criteria 68 single die games, 6 multiple dies games, and 43 matrix tasks were selected for the current meta-analysis (see Appendix A for the included articles). Measures
Implausibility aversion. In none of the experiments participants were directly asked how plausible they thought it would be to honestly obtain the highest outcome. Therefore,
“implausibility aversion” was indirectly estimated by a measure that was available for all three experimental paradigms: the probability of obtaining the highest outcome. The lower the probability of obtaining the highest outcome was, the more unlikely it became to obtain this outcome through honest means. In this case it was expected that maximizing appeared less plausible to participants.
For the die game, the probability of actually rolling the maximum outcome could be calculated. Let 𝑥 be the outcome of a single die roll with 𝑞 sides. The probability of rolling the maximum number of pips with 𝑘 dies is given by
𝑃(𝑥 = 𝑥𝑚𝑎𝑥) = ( 1 𝑞)
𝑘
.
For the matrix task, the probability of obtaining the highest outcome was estimated by the percentage of participants that actually solved all matrices divided by the total number of participants. Let 𝑥 be the number of solved matrices, 𝑦 the number of participants that actually solved all matrices, and 𝑁 the total number of participants. The probability of solving all matrices is then given by
𝑃(𝑥 = 𝑥𝑚𝑎𝑥) = 𝑦 𝑁.
Maximizing. Lying to the maximum extent was measured by the average maximum overreporting. This was calculated by the mean difference between the number of participants who claimed the maximum outcome and the number of participants who actually were eligible for this maximum. For the matrix task, participants eligible for the maximum were those who actually solved all matrices. For both die games, the number of participants eligible for the
maximum was estimated based on the expected number of participants actually rolling the highest pip.
Data Analysis
Because various designs and manipulations were used in the included studies, a random-effects model was estimated. Robustness and heterogeneity methods were used to account for possible publication bias and variation in effect sizes, respectively.
Robustness. Although the article search covered both published and unpublished studies, the sample might have been biased because low-effect size studies were less likely published or reported. To assess whether the sample was affected by selective reporting of significant studies only (publication bias), the effect sizes per study were be plotted against their standard errors. If studies were not symmetrically distributed around the average effect size, this suggested
publication bias (Borenstein et al., 2009). In this case, the Trim and Fill method was used to correct this bias. This iterative procedure removed the studies with relatively large effect sizes and standard errors, and added hypothetically “unpublished” studies until the mean effect sizes were symmetric around the mean (Duval & Tweedie, 2000a; 2000b). Previous sensitivity analyses suggested that this method correctly adjusted the data in 55-90% of the cases (Idris, 2012).
Heterogeneity. Because of the variation in study designs, heterogeneity in effect sizes was expected for each paradigm. To test this, the 𝜏2statistic was estimated for each paradigm
separately using the DerSimon and Laird (1986) method. This statistic quantified the amount of between-study variation in effect sizes. Cochran’s Q was estimated to test the null-hypothesis that the studies within each experimental paradigm shared the same effect size (i.e., 𝜏2 = 0;
common effect size and a random-effects model was required. Research on the reliability of this method in random-effects models suggested a reliability of 85.3-94.3% (Guolo & Varin, 2007; IntHout, Ioannidis, & Borm, 2014).
Main analyses. A meta-regression analysis was performed with the three paradigms as subgroups. Average maximum overreporting was regressed on the probability of obtaining the highest outcome. A main effect was expected of probability of obtaining the highest outcome on average maximum overreporting, which would provide evidence for the hypothesis that
“implausibility aversion” predicts dishonesty. In addition, an interaction effect was expected between paradigm and probability of obtaining the highest outcome on average maximum overreporting. Q-tests were performed to compare the effect sizes between paradigms. The null-hypothesis of this test was that the effect sizes of average maximum overreporting were equal across paradigms. If this was rejected and the effect size of average maximum overreporting was significantly lower for the for the matrix task than for the single die game, it supported the hypothesis that participants showed more “implausibility aversion” in the matrix task than in the single die game. If, in addition, the effect size of average maximum overreporting was
significantly lower for the multiple dies game than the single die game, it supported the hypothesis that participants showed more “implausibility aversion” in the multiple dies game than in the single die game. Multiple comparisons were corrected for by the Bonferroni correction.
Results Confirmatory Analyses
The overall level of average maximum overreporting was relatively low, MD = 8.57%, 95% CI [6.76%, 10.38%], z = 9.27, p < .001, which indicated that approximately 8.57% of
participants overreported to the maximum extent. The mean differences for the separate paradigms were summarized in Table 1. The significance level was divided by the number of paradigms to correct for multiple comparisons, α = .05/3 = .017. In line with expectations, average maximum overreporting was higher lower in the matrix task than in the single die game, Q(1) = 34.43, p < .001. Furthermore, average maximum overreporting was lower in the multiple dies game compared to the single die game, Q(1) = 7.41, p =.007. For the multiple dies game, average maximum overreporting was not significant at the α = .05 level, which indicated that in this game the proportion of participants that maximized did not differ from the proportion that would be expected from an honest sample. However, the sample size of the multiple dies game was relatively low (N = 16), which could have distorted the results. Also, the mean difference of the multiple dies game showed a positive trend in average maximum overreporting, suggesting that a proportion of participants tended to maximize in this game.
Heterogeneity. Using the DerSimon and Laird Method, Cochrane’s Q suggested overall heterogeneity in effect sizes, τ² = 0.008, Q(167) = 1088.64, p < .001. Heterogeneity in effect sizes was also found within each paradigm, for the matrix task, τ² = 0.002, Q(42) =186.53, p < .001, for the single die game, τ² = 0.006, Q(67) = 193.37, p < .001, and for the multiple dies game, τ² = .008, Q(15) = 57.08, p < .001. These results suggested that a random-effects model was required. Heterogeneity in the studies is also shown in the forest plot in Figure 1. See Appendix A for an overview of the corresponding studies. The mean difference varied from MD = 44.44% for Wib12a in the predicted direction to MD = -16.67% for Uti13b in the opposite direction. The next section explored whether this heterogeneity within the paradigms could be accounted for by possible publication bias. Furthermore, the section on meta-regression explored
whether the differences in effect sizes between paradigms could be explained by the probability of obtaining the highest outcome.
Table 1
Mean differences (MD) of average maximum overreporting for all paradigms combined and the separate paradigms with number of studies k, 95% CIs, z-values, and p-values.
Robustness. To assess possible publication bias, the mean differences were plotted against their standard errors in the funnel plot in Figure 2. The funnel plot includes both the observed studies and the added “unpublished” studies by Trim and Fill. The plot shows that studies were missing on the negative part of the mean difference. Negative mean differences represent the studies in which the reported outcomes were lower on average than would be expected from an honest sample. In accordance with this, Trim and Fill found evidence for publication bias and added 41 “unpublished” studies. As a result, average maximum overreporting reduced to MD = 3.02%, 95% CI [1.41%, 4.63%], z = 3.67,
p < .001. This suggests that, on average, the proportion of participants reporting the highest outcome was lower than proposed by the original meta-analysis.
k MD (%) 95% CI z p
Lower Upper
All paradigms 127 8.57% 6.76% 10.38% 9.27 <.001
Single die game 68 12.89% 10.49% 15.30% 10.49 <.001
Multiple dies game 16 4.99% -0.16% 10.15% 1.90 .057
1.
3.
Figure 1. Forest plots of average maximum overreporting for (1) the single die game, (2) the multiple dies game, and (3) the matrix task. Studies with larger squares had a lower standard error and had therefore larger weights in the random effects model. The horizontal bars indicate the corresponding 95% CI.
Figure 2. Funnel plot of mean differences against their standard errors. The plot includes both the observed studies (grey dots) and the “unpublished” studies imputed by Trim and Fill (white dots). The dotted triangle represents the pseudo-CI, the area in which 95% of the studies would fall in absence of publication bias. The dotted vertical line represent the mean adjusted by Trim and Fill. The white contour represents the region in which the studies would need to be in order to achieve the 5% confidence level.
The three paradigms were also separately inspected using the Trim and Fill method. The adjusted mean differences are shown in Table 2. Trim and Fill suggested publication bias for all three paradigms. No studies were added to the multiple dies game, but the mean difference was adjusted to a significant positive value. However, because the number of included multiple dies games was low (N = 16), single studies had a relatively strong effect on the adjustments of Trim and Fill and might have distorted the results. Therefore the
adjustments for this paradigm needed to be interpreted with care.
For the die roll game with a single die Trim and Fill added five “unpublished” studies and suggested that the mean difference was higher than proposed by the original analyses. For
the matrix task one “unpublished” study was added, correcting the mean difference to a slightly lower value than originally proposed.
Table 2
Adjusted mean differences (MD) of average maximum overreporting by Trim and Fill for all paradigms combined and the separate paradigms with number of studies k (adjusted number of studies by Trim and Fill), 95% CIs, z-values, and p-values.
Meta-regression. A meta-regression model was fitted in which average maximum overreporting was regressed on the probability of obtaining the highest outcome for all paradigms combined. The results are summarized in Table 3. In line with expectations, the probability of obtaining the highest outcome significantly predicted average maximum overreporting. That is, when the probability of obtaining the highest outcome decreased by 10%, the proportion of participants overreporting to the maximum extent decreased by 5.5%. This model with a single predictor accounted for a relatively high amount of heterogeneity, R = 40.76%. This supported the first hypothesis that maximizing decreased when implausibility aversion increased.
k (adj. k) adj. MD (%) 95% CI z p
Lower Upper
All paradigms 127 (168) 3.02% 1.41% 4.63% 3.67 <.001 Single die game 68 (73) 14.05% 11.06% 16.49% 11.24 <.001 Multiple dies game 16 (16) 0.05% 0.29% 9.71% 2.08 .038 Matrix task 43 (44) 3.71% 2.28% 5.14% 5.08 <.001
Table 3
Regression coefficients of the meta-regression model of average maximum overreporting regressed on probability of obtaining the highest outcome (prob. max) for all paradigms combined with standard errors (SE), 95% CIs, z-values, and p-values.
B (SE) 95% CI z p Lower Upper Constant 0.04 (0.01) 0.02 0.05 4.24 <.001 Prob. max 0.55 (0.12) 0.40 0.69 7.25 <.001 Observations k = 126, N =11026 Explained heterogeneity R² = 40.76%
To explore whether the differences in average maximum overreporting between paradigms could be explained by the probability of obtaining the highest outcome, a meta-regression model fitted an interaction effect between probability of obtaining the highest outcome and paradigm on average maximum overreporting. Counter to expectations, only a significant interaction effect was found between the multiple dies game and the probability of obtaining the highest outcome on average maximum overreporting, b = 2.29 (SE = 1.14), 95% CI [0.06% , 4.52%], z = 2.01, p = .044. This indicated that when the probability of obtaining the highest outcome decreased with 10% in the multiple dies game, average maximum overreporting decreased with 22.9%. In this multiple dies game, average maximum overreporting accounted for R² = 18.04% of heterogeneity. However, for the other two paradigms no interaction effect was found, indicating that the probability of obtaining the highest outcome did not significantly predict average maximum overreporting in the single
die game and the matrix task. In the next section possible explanations for this lack of interaction effect are investigated.
Exploratory Analyses
The finding that there was no interaction between average maximum overreporting and probability of obtaining the highest outcome for the matrix task and the single die game was unexpected. A possible explanation for this is that there was not enough variation in the probability of obtaining the highest outcome within the single die game and the matrix task. For the single die game, the maximum possible number of pips varied little, as shown in Figure 2. The majority of these games used a six-sided die, which resulted in the fact that most of the studies had an equal probability of obtaining the highest outcome. The matrix task also lacked variation. In only five of 43 matrix tasks at least one participant solved all
matrices. Because of this, the lower limit of the predictor variable (e.g., probability of obtaining the highest outcome) was not reached in most of the studies, which resulted in a floor effect. This floor effect was reflected in the fact that for only five studies the probability of obtaining the highest outcome was nonzero. This lack of variation in the single die game and the matrix task might explain why the probability of obtaining the highest outcome significantly predicted average maximum overreporting for all paradigms combined, but not for the model that was fitted to each paradigm separately. In the next section a
meta-regression model with less predictors was fitted to investigate whether this diminished the effects of lack of variation.
Figure 3. Bar plot of the number of sides of the dies used in the single die games.
For the fitted meta-regression model with an interaction between the probability of obtaining the highest outcome and paradigm, six predictors needed to be estimated. The data however may have lacked sufficient variation to estimate all six predictors. To reduce the number of predictors, another meta-regression model with less parameters was estimated. This model included paradigm and the probability of obtaining the highest outcome as predictors of average maximum overreporting. If the differences in average maximum overreporting between paradigms would become non-significant after including the probability of obtaining the highest outcome as a predictor, this would suggest that differences in maximizing between the paradigms were explained by “implausibility aversion”.
The regression coefficients and model fit of this model are summarized in Table 3. The probability of obtaining the highest outcome predicted average maximum overreporting, when including paradigm in the model. Furthermore, the separate paradigms did not predict average maximum overreporting when the probability of obtaining the highest outcome was included in the model. This suggested that the probability of obtaining explained the
differences in average maximum overreporting between the paradigms. This meta-regression model with only four predictors accounted for a substantial proportion of heterogeneity, R = 39.66%. However, the confidence interval of the estimate of the probability of obtaining the highest outcome varied between a value close to zero to a value greater than one, which indicated that the estimate of the effect was imprecise.
Table 3
Regression coefficients of the meta-regression model with paradigm and the probability of obtaining the highest outcome (prob. max.) as predictors of average maximum overreporting, with standard errors (SE), 95% CIs, z-values, and p-values.
B (SE) 95% CI z p
Lower Upper
Prob. max 0.60 (0.29) 0.04 1.16 2.09 .037
Single die game 0.03 (0.05) -0.07 0.13 0.57 .568
Multiple dies game 0.01 (0.05) -0.09 0.08 -0.14 .886
Matrix task -0.01 (0.04) -0.09 0.11 0.21 .835
Observations k = 126, N = 11026
Explained heterogeneity R² = 39.66%
Discussion
In the current meta-analysis it was investigated whether “implausibility aversion” predicted dishonesty in three commonly studied experimental paradigms, the single die game, multiple dies game, and the matrix task. In line with expectations, less people maximized
when it was more implausible to obtain the maximum outcome through honest means (e.g., matrix task and multiple dies game) compared to when this was more plausible (e.g., single die game). Robustness analyses provided evidence for publication bias: correcting for this bias decreased maximizing substantially for all paradigms combined. However, this
correction did not eliminate the differences in maximizing between the separate paradigms. A meta-regression model showed that “implausibility aversion” negatively predicted maximizing for all paradigms combined, which provided evidence for the hypothesis that maximizing decreased when “implausibility aversion” increased. However, in a meta-regression model with an interaction between paradigm and “implausibility aversion” the effect of “implausibility aversion” on maximizing disappeared. In this model, only an interaction was found between the multiple dies game and maximizing. This implied that “implausibility aversion” did not predict maximizing in the matrix task and the single die game and that the found relation between “implausibility aversion” and maximizing was driven by the multiple dies game only. This was counter to the expectation that
“implausibility aversion” explained the differences in maximizing between the paradigms. However, lack of variation in the possible maximum outcomes of the matrix task and the single die game might have caused the absence of an interaction between these paradigms and “implausibility aversion”. Exploratory analysis supported this notion: a meta-regression model with paradigm and “implausibility aversion” as predictors of average maximum
overreporting showed that “implausibility aversion” did explain the differences in maximizing between the three paradigms.
Taken together, these results suggested that “implausibility aversion” predicted maximizing. Although the results need to be interpreted with care because the estimated strength of the effect was relatively imprecise, these findings suggest that “implausibility aversion” is a factor that needs to be accounted for in research on dishonesty.
Limitations and Directions for Future Research
There were a number of limitations to the current study. One important limitation was the lack of variation in possible maximum outcomes, which might explain why
“implausibility aversion” did not interact with maximizing in the matrix task and the single die game. The lack of variation in the matrix task was caused by the fact that in most of these tasks none of the participants managed to honestly solve all matrices. The participants were probably not given sufficient time so that at least a proportion of the participants were able to solve all matrices. This resulted in a floor effect, because in the majority of matrix tasks the probability of obtaining the highest outcome was zero. Similarly, in most single die games a six-sided die was used, which might have caused the lack of variation in the extent of possible maximizing. As a result, it was difficult to detect a relation between “implausibility aversion” and maximizing for these paradigms separately. In future research, this lack of variation in the single die game could possibly be avoided by comparing maximizing in different multiple dies games with substantial variation in number of sides. For matrix tasks, participants could be given more time, so that at least some - but not all - participants manage to solve all matrices. In doing this, the observed floor effects could possibly be avoided.
Furthermore, although the results showed that people seemed to consider the
plausibility of their lie, it remains unclear what the mechanism behind this was. One possible explanation is that people wanted to avoid the actual consequences of being caught cheating (e.g., miss out on a reward). Alternatively, reputational concerns about being viewed as a liar might have played a role. In line with this second explanation, Gneezy, Kajackaite and Sobel (2016) showed that reputational concerns predicted dishonesty: more people lied when their outcome could not be observed by the experimenter compared to when the actual outcome could later be verified. Furthermore, in the matrix task a participant’s outcome depended on performance rather than chance (like in the die game). Therefore, participants might have
been motivated to avoid face loss (e.g., being viewed as bad performers), especially when they had to report their outcome to the experimenter. In all societies, and particularly in collectivist societies, people are motivated to avoid face loss and to maintain their reputation (Zhang, Cao, & Grigoriou, 2011). Therefore, dishonesty might have been motivated by different processes in the matrix task and the die game. An experimental setup which distinguishes these motivational concerns would be necessary to explore if dishonesty is indeed driven by different processes in these paradigms.
Another limitation was that the current meta-analysis was based only on correlational data. Hence, it could not be determined if “implausibility aversion” or self-concept
maintenance caused the observed behavior. That is, when an outcome became more
implausible, reporting this outcome might have also become less easily categorizable as an honest act. An experiment in which it is studied whether (dis)honest behavior is driven by the social motivation of reporting a plausible outcome to maintain an honest reputation or by a psychological process of maintaining an honest self-concept would be necessary to assess which of these theories best explains dishonest behavior. For instance, the measure of Mazar et al. (2008) whether someone’s self-concept is updated after a lie could be compared in games with different probabilities of honestly obtaining the highest outcome (i.e., single die game with 6 vs. 12 sides). If it is indeed harder to justify an implausibly lie, a larger
proportion of maximizers would update their self-concept in the 12-sided compared to the 6-sided game. This would provide evidence for the theory of self-concept maintenance. To the contrary, “implausibility aversion” would predict no differences in self-concept updating between these games. In addition, with a questionnaire it could be directly assessed how plausible people consider it to be to honestly obtain the maximum outcome. With these experiments the motivations behind dishonest behavior could be explored.
A final limitation is that no conclusions could be drawn on individual differences in dishonesty and susceptibility to implausibility aversion, because participants did not
participate in all three paradigms. Fischbacher and Föllmi-Heusi (2013) suggested that in the die game about 20% of the people maximized, while 39% of the people did not lie at all. This suggests that the population consists of different lying types, in which a proportion never lies, another proportion always maximizes, and a final proportion varies between “truth stretching” and maximizing according to how plausible it is to maximize. Data on individual dishonesty in all paradigms is necessary to assess if these types indeed exist and are stable in different contexts. This would shed more light on the question whether more people lie when it becomes plausible enough or whether some people are always honest regardless of context. Theoretical and Practical Implications
Although more research is required to estimate the impact of “implausibility aversion” on behavior, this meta-analysis showed that this factor has been neglected in research on dishonesty. This might be problematic for the ecological validity of studies that have drawn conclusions on dishonesty levels and underlying motivations, without considering the possibility that people were motivated to tell a plausible lie. That is, people might in fact act in line with a cost-benefit analysis; weighing the potential costs of telling an implausible lie (i.e., reputational concerns) to the corresponding benefits and comparing it to those of a “truth stretching” lie. Hence, the finding that most people “stretch the truth” rather than maximize might reflect plausibility concerns, rather than motivations for self-concept maintenance. Therefore it is important to not disregard the classical economic model in favor of the theory of self-concept maintenance until the role of “implausibility aversion” in dishonesty is further investigated.
If future research shows that the plausibility of a lie plays a substantial role in dishonesty, and that people indeed act in line with a cost-benefit analysis, this might be a
useful tool to counteract societal dishonesty. In that case, increasing the probability of detection could decrease dishonesty because the potential costs of getting caught might no longer outweigh the benefits. In line with this, criminology studies on deterrence showed that the certainty of punishment is a better predictor of criminal behavior than the severity of punishment (Wright, 2010). To reduce tax misreporting, for example, policy makers could stress the fact that all tax forms are checked on their plausibility. In addition, information on the motivations for dishonesty could also help reduce it. For instance, if it turns out that reputational concerns predict dishonesty, measures that damage people’s reputation when they lie might also help reduce dishonesty. In case of tax misreporting, policy makers could
introduce a system in which citizens’ reputations is damaged if they are caught, by putting them under stricter control in subsequent years. Similarly to reduce fare dodging in public transport, policy makers could introduce higher ticket checking frequencies to increase the detection probability rather than increasing fines. In addition, reputational measures in which fare dodgers have to use a differently colored ticket after getting caught might result in more people buying a ticket to avoid being perceived as a free rider by other travelers.
Conclusion
All in all, this meta-analysis showed that the plausibility of a lie influenced maximizing in economic experimental paradigms. Although it remains to be studied how large the effect is of “implausibility aversion” on dishonesty, the results suggest that this factor has not been studied sufficiently. People might not “stretch the truth” to maintain an honest self-concept, but rather to appear honest to others. This is an important distinction, because if people are motivated by coming across as honest rather than preserving an honest self-image, different measures are necessary to counteract societal dishonesty. External rewards and punishments might be more effective in this case than internal rewards and
punishments. Using these insights could potentially help lower societal dishonesty and the costs that society pays for it.
References
Becker, G. S. (1968). Crime and punishment: An economic approach. In The economic dimensions of crime (pp. 13-68). London, UK: Palgrave Macmillan. DOI:
10.1007/9781-349-62853-7_2
Borenstein, M., Hedges, L. V., Higgins, J. P., & Rothstein., H. R.(2009). Introduction to meta-analysis. Chichester, UK: John Wiley & Sons. DOI: 0.1002/9780470743386.refs Cochran, W. G. (1954). The combination of estimates from different experiments. Biometrics,
10, 101-129. DOI: 10.2307/3001666.
DerSimonian, R., & Laird, N. (1986). Meta-analysis in clinical trials. Controlled Clinical Trials, 7, 77-188. DOI: 10.1016/0197-2456(86)90046-2
Dufwenberg, M., & Gneezy, U. (2000). Measuring beliefs in an experimental lost wallet game. Games and Economic Behavior, 30, 163-182. DOI: 10.1006/game.1999.0715 Duval, S., & Tweedie, R. (2000). A nonparametric “trim and fill” method of accounting for
publication bias in meta-analysis. Journal of the American Statistical Association, 95, 89-98. DOI: 10.1080/01621459.2000.10473905
Duval, S., & Tweedie, R. (2000). Trim and fill: a simple funnel‐plot–based method of testing and adjusting for publication bias in meta‐analysis. Biometrics, 56, 455-463. DOI: 10.1111/j.0006-341X.2000.00455.x
Erat, S., & Gneezy, U. (2012). White lies. Management Science, 58, 723-733. DOI: 10.1287/mnsc.1110.1449
Fischbacher, U., & Föllmi-Heusi, F. (2013). Lies in disguise - An experimental study on cheating. Journal of the European Economic Association, 11, 525-547. DOI: 10.1111/jeea.12014
behavior. Journal of Applied Social Psychology, 43, 1281-1288. DOI: 10.1111/jasp.12091
Gerlach, P., Teodorescu, K., & Hertwig, R. (2017). The truth about lies. A meta-analysis on dishonest behaviour. Manuscript in preparation.
Gino F., Ayal S., Ariely D. (2009). Contagion and differentiation in unethical behavior: The effect of one bad apple on the barrel. Psychological Science, 20, 393–398. DOI: 10.1111/j.1467-9280.2009.02306.x
Gneezy, U. (2005). Deception: The role of consequences. The American Economic Review, 95, 384-394. DOI: 10.1257/0002828053828662
Gneezy, U., A. Kajackaite, and J. Sobel (2016). Lying aversion and the size of
the lie (Technical Report). San Diego, SDC: UCSD. DOI: 10.2139/ssrn.2852055 Guolo, A., & Varin, C. (2015). Random-effects meta-analysis: The number of studies matters.
Statistical Methods in Medical Research, 0, 1-19. DOI: 10.1177/0962280215583568 Idris, N. R. N. (2012). Performance of the trim and fill method in adjusting for the publication
bias in meta-analysis of continuous data. American Journal of Applied Sciences, 9, 1512. DOI: 10.3844/ajassp.2012.1512.1517
IntHout, J., Ioannidis, J. P., & Borm, G. F. (2014). The Hartung-Knapp-Sidik-Jonkman method for random effects meta-analysis is straightforward and considerably outperforms the standard DerSimonian-Laird method. BMC Medical Research Methodology, 14, 25. DOI: 10.1186/1471-2288-14-25
Kajackaite, A., & Gneezy, U. (2015). Lying costs and incentives. Games and Economic Behavior, 102, 433–444. DOI: 10.1.1.702.7982
Mazar, N., Amir, O., & Ariely, D. (2008). The dishonesty of honest people: A theory of self- concept maintenance. Journal of Marketing Research, 45, 633-644. DOI:
Peters, J. L., Sutton, A. J., Jones, D. R., Abrams, K. R., & Rushton, L. (2007). Performance of the trim and fill method in the presence of publication bias and between‐study
heterogeneity. Statistics in Medicine, 26, 4544-4562. DOI: 10.1002/sim.2889 Rosenbaum, S. M., Billinger, S., & Stieglitz, N. (2014). Let’s be honest: A review of
experimental evidence of honesty and truth-telling. Journal of Economic Psychology, 45, 181-196. DOI: 10.1016/j.joep.2014.10.002
Serota, K. B., Levine, T. R., & Boster, F. J. (2010). The prevalence of lying in America: Three studies of self-reported lies. Human Communication Research, 36, 2-25. DOI: 10.1111/j.1468-2958.2009.01366.x
Shalvi, S., Dana, J., Handgraaf, M. J., & De Dreu, C. K. (2011). Justified ethicality: Observing desired counterfactuals modifies ethical perceptions and behavior. Organizational Behavior and Human Decision Processes, 115, 181-190. DOI: 10.1016/j.obhdp.2011.02.001
Suri, S., Goldstein, D. G., & Mason, W. A. (2011). Honesty in an online labor market. Human Computation, 11, 61-66. DOI: 10.1.1.361.2135
Wright, V. (2010). Deterrence in criminal justice: Evaluating certainty vs. severity of punishment. Washington, DC: The Sentencing Project.
Zhang, X. A., Cao, Q., & Grigoriou, N. (2011). Consciousness of social face: The
development and validation of a scale measuring desire to gain face versus fear of losing face. The Journal of Social Psychology, 151, 129-149. DOI:
Footnotes
¹ Reporting 6 pips in a die game with a single six-sided die is an overstatement of 2.5 units compared to the expected outcome (6 – 3.5), while reporting all matrices as solved in a game with 20 matrices is an overstatement of 13 units on average (20 – 7).
Appendix A
This appendix contains a reference list of all studies that were included in the meta-analysis with the corresponding code (used in Figure 1) and probability of honestly obtaining the highest outcome. All conditions were coded as a single study.
Table A1. Overview of the included single die games with the probability of honestly obtaining the highest outcome (prob. max) and the
reference of the corresponding article. The code corresponds to the forest plot in Figure 1. For concision, a reference is only mentioned once for an article and the same reference is subsequently indicated by (-).
Code Study Prob. max Reference
Arb14a Study 1 .167 Arbel, Y., Bar-El, R., Siniver, E., & Tobol, Y. (2014). Roll a die and tell a lie – What affects honesty? Journal of Economic Behavior & Organization, 107, 153-172. DOI: 10.1016/j.jebo.2014.08.009
Arb14c Study 2 .167 -
Cad16 Study 3, incentive to lie .167 Cadsby, C. B., Du, N., & Song, F. (2016). In-group favoritism and moral decision-making. Journal of Economic Behavior &
Con13a Individual .167 Conrads, J., Irlenbusch, B., Rilke, R. M., & Walkowitz, G. (2013). Lying and team incentives. Journal of Economic Psychology, 34, 1-7. DOI: 10.1016/j.joep.2012.10.011
Con13b Team .167 -
Con13c Team-mixed, individual .167 -
Con13d Team-mixed, team .167 -
Con14a T1 .167 Conrads, J., Irlenbusch, B., Rilke, R. M., Schielke, A., & Walkowitz,
G. (2014). Honesty in tournaments. Economics Letters, 123, 90-93. DOI: 10.1016/j.econlet.2014.01.026
Con14b T3 .167 -
Con14c T5 .167 -
Clo14a Control .167 Clot, S., Grolleau, G., & Ibanez, L. (2014). Smug Alert! Exploring self-licensing behavior in a cheating game. Economics Letters, 123, 191-194. DOI: 10.1016/j.econlet.2014.01.039
Dai16a Station . 333 Dai, Z., Galeotti, F., & Villeval, M. C. (2016). Cheating in the lab predicts fraud in the field: An experiment in public transportation. Management Science, Articles in Advance, 1-20. DOI:
10.1287/mnsc.2016.2616
Dai16b Fine collection office .333 -
Fis13a Baseline .167 Fischbacher, U., & Föllmi‐Heusi, F. (2013). Lies in disguise - An experimental study on cheating. Journal of the European Economic Association, 11, 525-547. DOI: 10.1111/jeea.12014
Fis13b High stake .167 -
Fis13c 4,9 .167 -
Fis13d Externality .167 -
Fis13e Double anonymous .167 -
Gac16a Austria .167 Gächter, S., & Schulz, J. F. (2016). Intrinsic honesty and the
prevalence of rule violations across societies. Nature, 531, 496-499. DOI: 10.1038/nature17160
Gac16b China, high stakes .167 -
Gac16c China, low stakes .167 -
Gac16d Colombia .167 -
Gac16e Czech Republic .167 -
Gac16f Georgia .167 - Gac16g Germany .167 - Gac16h Guatemala .167 - Gac16i Indonesia .167 - Gac16j Italy .167 - Gac16k Kenya .167 - Gac16l Lithuania .167 - Gac16m Malaysia .167 - Gac16n Morocco .167 -
Gac16o The Netherlands .167 -
Gac16q Slovakia .167 -
Gac16r South Afrika .167 -
Gac16s Spain .167 -
Gac16t Sweden .167 -
Gac16u Tanzania .167 -
Gac16v Turkey .167 -
Gac16w United Kingdom .167 -
Gac16x Vietnam .167 -
Gin13h Study 2, mandatory regulation .167 Gino, F., Ayal, S., & Ariely, D. (2013). Self-serving altruism? The lure of unethical actions that benefit others. Journal of Economic Behavior & Organization, 93, 285-292. DOI:
10.1016/j.jebo.2013.04.005 Gin13i Study 2, no mandatory regulation .167 -
Gra13a Random income .125 Gravert, C. (2013). How luck and performance affect stealing. Journal of Economic Behavior & Organization, 93, 301-304. DOI: 10.1016/j.jebo.2013.03.026
Gun14b Study 2, Mturk .167 Gunia, B. C., Barnes, C. M., & Sah, S. (2014). The morality of larks and owls: Unethical behavior depends on chronotype as well as time of day. Psychological Science, 25, 2272-2274. DOI:
10.1177/0956797614541989
Gun14c Study 2, students .167 -
Jac15a BT .167 Jacobsen, C., & Piovesan, M. (2016). Tax me if you can: An
artefactual field experiment on dishonesty. Journal of Economic Behavior & Organization, 124, 7-14. DOI:
10.1016/j.jebo.2015.09.009
Jac15b TT .167 -
Mue15a Individual .167 Muehlheusser, G., Roider, A., & Wallmeier, N. (2015). Gender differences in honesty: Groups versus individuals. Economics Letters, 128, 25-29. DOI: 10.1016/j.econlet.2014.12.019
Mue15b Team .167 -
Ruf14 - .167 Ruffle, B. J., & Tobol, Y. (2014). Honest on Mondays: Honesty and
the temporal separation between decisions and payoffs. European Economic Review, 65, 126-135. DOI:
10.1016/j.euroecorev.2013.11.004
Sha11a Study 1, single-roll .167 Shalvi, S., Dana, J., Handgraaf, M. J., & De Dreu, C. K. (2011). Justified ethicality: Observing desired counterfactuals modifies ethical perceptions and behavior. Organizational Behavior and Human Decision Processes, 115, 181-190. DOI:
10.1016/j.obhdp.2011.02.001
Sha11c 3.50 exit .167 Shalvi, S., Handgraaf, M. J., & De Dreu, C. K. (2011). Ethical manoeuvring: Why people avoid both major and minor lies. British Journal of Management, 22, 16-27. DOI:
10.1111/j.1467-8551.2010.00709.x
Sha11d 2.50 exit .167 -
Sha12a Study 1, low time pressure .167 Shalvi, S., Eldar, O., & Bereby-Meyer, Y. (2012). Honesty requires time (and lack of justifications). Psychological Science, 23, 1264-1270. DOI: 10.1177/095679761244383
Sha12b Study 1, high time pressure .167 -
Sha12c Study 2, low time pressure .167 -
Sha12d Study 2, high time pressure .167 -
Sha13 - .167 Shalvi, S., & Leiser, D. (2013). Moral firmness. Journal of Economic
Behavior & Organization, 93, 400-407. DOI: 10.1016/j.jebo.2013.03.014
Uti13a Students .167 Utikal, V., & Fischbacher, U. (2013). Disadvantageous lies in
individual decisions. Journal of Economic Behavior & Organization, 85, 108-111. DOI: 10.1016/j.jebo.2012.11.011
Uti13b Nuns .167 -
Wib12a Placebo group .167 Wibral, M., Dohmen, T., Klingmüller, D., Weber, B., & Falk, A. (2012). Testosterone administration reduces lying in men. PloS one, 7, e46774. DOI:10.1371/journal.pone.0046774
Table A2. Overview of the included multiple dies games with the probability of honestly obtaining the highest outcome (prob. max) and the reference of the corresponding article. The code corresponds to the forest plot in Figure 1. For concision, a reference is only mentioned once for an article and the same reference is subsequently indicated by (-).
Code Study Prob. max Reference
Abe14b 4-coin, telephone .063 Abeler, J., Becker, A., & Falk, A. (2014). Representative evidence on lying costs. Journal of Public Economics, 113, 96-104. DOI:
10.1016/j.jpubeco.2014.01.005
Abe14c 4-coin, lab, telephone .063 -
Abe14d 4-coin, lab, click .063 -
Con15a F-t-F .063 Conrads, J., & Lotz, S. (2015). The effect of communication channels on dishonest behavior. Journal of Behavioral and Experimental Economics, 58, 88-93. DOI: 10.1016/j.socec.2015.06.006
Con15c C-lab .063 -
Con15d C-remote .063 -
Sch16a Control .028 Schurr, A., & Ritov, I. (2016). Winning a competition predicts dishonest behavior. Proceedings of the National Academy of Sciences, 113, 1754-1759. DOI: 10.1073/pnas.1515102113
Sch16c Study 1, losers .028 -
Sch16e Study 2, losers .028 -
Sch16f Study 3a, winners .028 -
Sch16g Study 3a, losers .028 -
Sch16h Study 3b, winners .028 -
Table A3. Overview of the included matrix tasks with the probability of honestly obtaining the highest outcome (prob. max) and the reference of the corresponding article. The code corresponds to the forest plot in Figure 1. For concision, a reference is only mentioned once for an article and the same reference is subsequently indicated by (-).
Code Study Prob. max Reference
Cai15a Study 1, eyes 0.00 Cai, W., Huang, X., Wu, S., & Kou, Y. (2015). Dishonest behavior is not affected by an image of watching eyes. Evolution and Human Behavior, 36, 110-116. DOI: 10.1016/j.evolhumbehav.2014.09.007
Cai15b Study 1, control 0.00 -
Far15a Study 2, piece rate 0.00 Faravelli, M., Friesen, L., & Gangadharan, L. (2015). Selection, tournaments, and dishonesty. Journal of Economic Behavior & Organization, 110, 160-175. DOI: 10.1016/j.jebo.2014.10.019
Far15b Study 3, tournament rate 0.00 -
Far15c Study 4, piece rate 0.00 -
Fri12 Main study 0.00 Friesen, L., & Gangadharan, L. (2012). Individual level evidence of dishonesty and the gender effect. Economics Letters, 117, 624-626. DOI: 10.1016/j.econlet.2012.08.005
Gra13b Performance income 0.12 Gravert, C. (2013). How luck and performance affect stealing. Journal of Economic Behavior & Organization, 93, 301-304. DOI: 10.1016/j.jebo.2013.03.026
Gro16a Gain frame 0.00 Grolleau, G., Kocher, M. G., & Sutan, A. (2016). Cheating and loss aversion: Do people cheat more to avoid a loss? Management Science, 62, 3428-3438. DOI: 10.1287/mnsc.2015.2313
Gro16b Loss frame 0.00 -
Gun14a Study 1 0.00 Gunia, B. C., Barnes, C. M., & Sah, S. (2014). The morality of larks and owls: Unethical behavior depends on chronotype as well as time of day. Psychological Science, 25, 2272-2274. DOI:
Hil16a Study 1a, loyalty 0.07 Hildreth, J. A. D., Gino, F., & Bazerman, M. (2016). Blind loyalty? When group loyalty makes us see evil or engage in it. Organizational Behavior and Human Decision Processes, 132, 16-36. DOI:
10.1016/j.obhdp.2015.10.001
Hil16b Study 1a, control 0.00 -
Hil16c Study 1b, loyalty 0.00 -
Hil16d Study 1b, control 0.00 -
Hil16g Study 3a, loyalty 0.00 -
Hil16h Study 3a, control 0.03 -
Hil16i Study 3b, loyalty pledge 0.06 -
Hil16j Study 3b, no loyalty 0.00 -
Hil16k Study 3b, control 0.00 -
Hil16n Study 5a, loyalty low competition 0.00 - Hil16o Study 5a, loyalty high competition 0.00 - Hil16p Study 5a, control low competition 0.00 -
Hil16p Study 5a, control low competition 0.00 - Hil16q Study 5a, control high competition 0.00 - Hil16r Study 5b, loyalty low competition 0.00 - Hil16s Study 5b, loyalty high competition 0.00 - Hil16t Study 5b, control low competition 0.00 - Hil16u Study 5b, control high competition 0.00 -
Kou14c Study 4, morning 0.00 Kouchaki, M., & Smith, I. H. (2014). The morning morality effect: The influence of time of day on unethical behavior. Psychological Science, 25, 95-102. DOI: 10.1177/0956797613498099
Kou14d Study 4, afternoon 0.00 -
Kou14e Study 3, light backpack 0.00 Kouchaki, M., Gino, F., & Jami, A. (2014). The burden of guilt: Heavy backpacks, light snacks, and enhanced morality. Journal of Experimental Psychology: General, 143, 414. DOI:
Kou14f Study 3, heavy backpack 0.00 -
Kou15a Study 1, exclusion 0.00 Kouchaki, M., & Wareham, J. (2015). Excluded and behaving unethically: Social exclusion, physiological responses, and unethical behavior. Journal of Applied Psychology, 100, 547. DOI:
10.1037/a0038034
Kou15b Study 1, inclusion 0.00 -
Lee15b Main study 0.00 Lee, J. J., Gino, F., Jin, E. S., Rice, L. K., & Josephs, R. A. (2015). Hormones and ethics: Understanding the biological basis of unethical conduct. Journal of Experimental Psychology: General, 144, 891. DOI: 10.1037/xge0000099
Maz08c Study 2, $0.5 control vs. recycle 0.00 Mazar, N., Amir, O., & Ariely, D. (2008). The dishonesty of honest people: A theory of self-concept maintenance. Journal of Marketing Research, 45, 633-644. DOI: 10.1509/jmkr.45.6.633
Maz08d Study 2, $2 control vs. recycle 0.00 - Maz08e Study 2, $0.5 control vs. honor code 0.00 - Maz08f Study 2, $2 control vs. honor code 0.00 -
Maz08g Study 3, control vs. recycle 0.00 -
Maz08h Study 4, first task 0.00 -
Rig14a Noncompetitive, self-grading 0.00 Rigdon, M. L., & D'Esterre, A. P. (2015). The effects of competition on the nature of cheating behavior. Southern Economic Journal, 81, 1012-1024. DOI: 10.4284/0038-4038-2012.301
