Edited by: Yiping Zhong, Hunan Normal University, China
Reviewed by: John K. Kruschke, Indiana University Bloomington, United States; Katarzyna Sekścińska, University of Warsaw, Poland
This article was submitted to Cognition, a section of the journal Frontiers in Psychology
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Laboratory studies allow studying the predictors of bribe-taking in a controlled setting. However, presently used laboratory tasks often lack any connection to norm violation or invite participants to role-play. A new experimental task for studying the decision to take a bribe was designed in this study to overcome these problems by embedding the opportunity for bribe-taking in an unrelated task that participants perform. Using this new experimental task, we found that refraining from harming a third party by taking a bribe was associated with lower offered bribes and higher scores of the participants on the honesty-humility scale from the HEXACO personality inventory. A trial-level analysis showed that response times were longer for trials with bribes and even longer for trials in which bribes were accepted. These results suggest that taking a bribe may require overcoming automatic honest response and support the validity of the honesty-humility scale in predicting moral behavior.
Corruption a considerable negative impact on economic and social development (
The majority of existing studies model corruption using a modified version of the “Trust game” in which a player in the role of a company or a citizen can transfer funds to a second player, usually called “public official.” The public official can then reciprocate this “bribe” by increasing the final reward of the first player at the expense of other players – usually representing the public (e.g.,
Some existing studies use neutral language in the instructions, whereas others explicitly mention corruption and bribes. Explicitly mentioning bribes was shown to decrease the prevalence of corrupt behavior (
In the present study, we introduce a new experimental task that overcomes this problem, because in the task, taking a bribe is always clearly associated with a norm violation, even when neutral language is employed. Moreover, unlike the tasks used in previous studies, in our new task participants face many opportunities to take bribes of various sizes as well as trials without any bribes. This allows us to examine how bribe-taking develops during the task, as well as what factors affect the length of decision whether to take a bribe or not. At the beginning of the task, participants are instructed to perform the task according to simple rules. For their work, they receive a fixed payment each time they perform the task, regardless of whether they do it in accordance with the rules or not – as do many public officials and employees in general. However, when they perform the task in violation of the given rules, another party is harmed. In the described setting, participants are motivated to perform their task correctly, based on the assumption that they do not want to deliberately harm the other party. The relevant behavioral norms follow from the rules of the task that participants are asked to perform, and there is, therefore, no need for role-playing. A bribe is introduced seamlessly into this setting as an offer of a financial reward for violating the rules. Participants can thus earn an additional reward (i.e., the bribe) by performing the task in violation of the rules while causing harm to the other party. Alternatively, participants can pass on this opportunity and simply perform the task according to the instructions. For simplicity, we focus solely on the behavior of the bribee in the current study. No participant is assigned the role of a briber, and therefore there is no possibility to study the decision to offer a bribe or the development of a relationship between briber and bribee. However, these aspects can be added to the game relatively easily in future implementations. Moreover, their current absence does not preclude us from studying a bribee’s decision-making in general, as there are also real-life instances of corruption in which an official is offered bribes without knowing the bribers and without the need to interact with them any further – such as when a bribe is included inside an application or together with documents the official is supposed to check.
The structure of the described experimental task in a simplified manner captures the various characteristics of bureaucratic work such as repetitive tasks, fixed payment, little to no quality control, straightforward rules of conduct, and opportunities for accepting bribes. At the same time, the description of the game and its rules can be kept abstract and neutral, without any references to public officials and bribery.
In the remainder of the paper, we describe in detail the new experimental task and present the results of its implementation. To demonstrate support for the validity of the new task for studying bribe-taking, we first show that participants are more likely to break the rules of the game when offered a bribe and that they are even more likely to do so in the case of larger bribes. Furthermore, we show that the perception of accepting a bribe as immoral and rejecting it as moral is associated with a lower likelihood of taking the bribe. We also conceptually replicate some previously found associations between gender, personality traits, and corrupt behavior using the new task. In particular, many previous studies found that women tend to act less dishonestly than men in general (
Afterward, we take advantage of the fact that our task offers participants many opportunities to take or decline bribes of different sizes during the experiment and we test a hypothesis about moral self-licensing and corrupt behavior. According to the moral licensing theory (
Finally, we attempt to contribute to the ongoing debate about whether people are intuitively honest or whether they need deliberation to overcome impulses to act dishonestly when given a chance (
Materials, data, the R script used for analysis, and preregistration of study hypotheses can be found at
We collected data from 200 participants (73% female, median age = 23 years, 84% university students) recruited from the participant pool of our experimental laboratory. The experiment was conducted as the second part of a larger set of unrelated studies and took approximately 50 min. All participants were compensated for their participation in the whole set of studies with 100 CZK (∼4 USD). Moreover, they could gain additional money in the experiment, as described in more detail below. The experiment was conducted on computers using a custom-written Python program. The data collection took place in a laboratory with groups of up to 13 people seated at workstations separated by dividers. The whole experiment was conducted in Czech. Participants gave an informed consent at the beginning of the experiment and they were debriefed regarding the purpose of the research at its end.
Participants were asked to sort by color 100 objects moving one by one from the left to the right on computer screens. Participants had maximum of 3 s for classifying each object. The objects were combinations of three shapes and three colors (e.g., blue circle, yellow square, orange triangle, and so on). The sorting was done by pressing one of three buttons on a keyboard (“1,” “2,” and “3” on the numerical part of the keyboard). Each of the three buttons was randomly assigned to a combination of color and shape. The assignment changed after each classification. For each sorted object (that is, even incorrectly sorted), participants received a fixed amount of 3 points. At the beginning of the task, 2000 points were assigned to a well-known Czech charitable organization providing humanitarian aid. When the object was not classified to the corresponding
To test the moral licensing effect, participants were divided into two groups with differing distributions of bribes. In the
The information about the number of the current trial and the points gained for oneself and the charity was displayed during the whole trial (see
An illustration of a computer screen seen by a participant. The top row shows information about the number of the current trial, the total number of trials, and the number of points currently assigned to the charity organization. In the middle of the screen, an object (a yellow square in this case) is moving from the left side of the screen to the right. The current participant’s reward in points is shown to the right of the screen. In the bottom row, a participant sees which shapes and colors are assigned to keys “1,” “2,” and “3” in this trial (in this example, 1 is yellow circle, 2 is orange square, and 3 is blue triangle). If the participant presses “1,” the object would be sorted by its color, that is correctly, and the participant would gain 3 points. If the participant presses “2,” the object would be matched to a wrong color, and would cause a loss of 200 points for the charity, but it would be sorted according to its shape, allowing the participant to gain the 100 points marked on the object in addition to the 3 points awarded for each sorted object.
In total, participants classified 94.9% of the trials correctly according to color. Participants therefore classified most of the objects according to the instructions. When an object was classified according to shape (as was the case in 38.9% of the trials), in most cases it was in a situation where classifications according to color and shape were aligned; 33.8% of trials were classified correctly according to both criteria. When the classification criteria were misaligned, and participants were offered a bribe, they took it in 15.1% percent of cases.
Thirty-nine percent of participants classified all the objects according to color and thus did not lose any of the 2000 points for the charity. The distribution of the final outcome for the charity was highly negatively skewed (skewness = −3.29) with a mean of 936 points (
After the experiment, we asked the participants about their perceptions of sorting the objects according to color or shape in cases where the object was associated with a bribe. On a scale from one to four (1 – certainly not, 2 – rather not, 3 – rather yes, 4 – certainly yes), participants did not generally perceive taking the bribe as despicable (
Perception of the task by participants. The answers were coded on a 4-point scale and their means are shown by black points. Error bars represent 95% confidence intervals. The transparent points display individual answers.
We found a predicted association between the HEXACO honesty-humility scale and the final outcome from the task. Participants scoring higher on the honesty-humility scale had a lower final reward for themselves (
We used mixed-effect logistic regression for analyzing bribe-taking on a trial level and mixed-effect linear regression for analyzing reaction times (
To assess the effect of bribes on classification, we performed an analysis with all trials that were classified correctly according to one of the two criteria (shape and color) excluding the trials in which both criteria were aligned. Classification based on shape (i.e., corresponding to taking a bribe when it was present) was used as the dependent variable and the presence of a bribe and the trial order as well as their interaction as predictors. We found out that the data were autocorrelated and hence we also included the classification on the previous trial as a covariate. This also means that the first trial was not included in the analysis. The partial autocorrelation coefficients were not significantly different from zero for higher trial lags. We included participant random slopes for both predictors. Participants seemed to be less likely to classify objects according to shape in later trials (
Next, we repeated the analysis using trials with misaligned criteria but including only trials with bribes. That is, the analysis was performed only for the trials in which participants had to decide between taking a bribe and leaving it. We included trial order and bribe size and their interaction as predictors and the main effects also as random slopes for participants. Participants were less likely to take the bribe in later trials (
The effect of bribe size on the probability of taking the bribe. The figure shows the average probability of taking a bribe in trials with misaligned classification criteria depending on the bribe size. The displayed points are computed across all participants and trials separately for the low bribe condition (black squares and saltire) and high bribe condition (red dots and cross). It is possible to see that the probability of taking a bribe increases with larger bribes and that participants are more likely to classify an object according to its shape when a bribe was present (squares and dots) than when it was absent (saltire and cross). Note that the abscissa ends at the probability of 0.25, not 1.
Since the two conditions differed in their bribe sizes, we conducted the analysis testing the effect of the condition using only the larger half of bribes that was held constant between the two groups. We included only random slopes for the trial order owing to convergence issues.
Finally, we conducted an analysis of log-transformed response times. The distribution of response times (
Participants classified the objects faster in later trials {
Next, we performed an analysis of response times using only the trials in which bribes were offered and taking them was associated with a loss to the charity. We used only the significant predictors from the previous model and added a response on a given trial as another predictor. Even for the selected trials, participants classified the objects faster in later trials {
In the present study, we introduced and attempted to validate a new laboratory task for studying a participant’s decision to take a bribe. The new task was created to overcome the limitations of tasks commonly used in previous studies. One limitation of these tasks lies in the fact that they model corruption in a way in which the corrupt behavior itself does not violate any obvious norms of conduct. If the task is then described using purely neutral language, it models corruption and bribe-taking only as simple economic transactions in which another party can be harmed, as in the well-known Dictator or Ultimatum games (
The participants in our study clearly understood the instructions, as is evident from the fact that they correctly classified objects in almost 95% of the trials. That the incorrect classification occurred only in 5% of trials may suggest that corrupt behavior in the task was extremely rare. However, one must take into account that bribes were offered only in 20% of all trials and only in two-third of these trials was the classification according to shape different from the classification according to color. From the trials with misaligned classification criteria, participants took the offered bribe and harmed the charity in approximately 15% of cases on average. Nevertheless, almost 40% of participants did not take any bribe that would harm the charity.
In line with our expectations, we found that taking bribes was associated with the size of a bribe, with the perception of the rule violation as not moral, and with lower scores on the honesty-humility scale from the HEXACO personality inventory.
The correlation with the honesty-humility scale (and an absence of any correlation with the other five personality traits) suggests that our task is in fact associated with honesty and not with the related concepts such as agreeableness (
In the present study, women were somewhat less likely to take bribes than men. The results seem to support general proclivity for less corrupt behavior in women as described by
Trial-level analysis showed that response times were longer for trials with bribes and even longer for trials with bribes that were accepted, suggesting that acting corruptly requires overcoming an automatic honest response. It could be argued that the trials with bribes were less frequent than the trials without bribes, and hence the longer response times might have been simply associated with a higher cognitive load. Since participants mostly classified the objects according to their color, it is possible that classifying an object according to its shape to take a bribe was less automatic, and thus resulted in longer response times. This possibility does not, however, fully explain the effect of the bribe size on response times. Participants took longer to decide whether to take bribes of higher values, regardless of whether they took the bribe or not. This suggests that the decision concerning larger bribes was not just less automatic, but it was also associated with a stronger motivational conflict. This result is in line with previous findings that identify approximately 40% of participants as unconditionally honest, 25% as unconditional cheaters, and the rest as susceptible to the situational factors (
We found no support for the moral licensing hypothesis in the present study. However, as can be seen from
In its current implementation, the game focuses solely on the behavior of a bribee, not a briber. There is, therefore, no interaction between a potential briber and the bribee, and participants are not told from whom they receive the bribes. This is comparable to many real-life cases of corruption where a public official, who for example issues permits or conducts inspections, is offered bribes in exchange for making speedy favorable decisions or ignoring non-compliance. The public official in these instances has only a very limited interaction with the bribers and sometimes does not need to interact with them directly at all. This understandably differs from the situation where an official has repeated interactions with the briber, such as in a relationship of a customs officer or a politician with a company for which they repeatedly provide illegal benefits. In the latter case, the outcome of the briber as well as their mutual trust with the official may play a role in the bribee’s decision-making, which could be therefore better modeled with a task in which these factors are present (e.g.,
That bribes were not offered by another person, but were generated randomly,
We introduced a new experimental task for the laboratory study of decision to accept bribes. The task includes a norm violation that is absent in most of the previously used tasks. Using the new task, we found that bribes, especially larger ones, are tempting even for those who do not take them at the end – as shown by longer reaction times not only on trials with bribes present in general, but also on those in which bribes were not accepted. These results are in line with previous findings that identified a substantial proportion of conditional cheaters among participants of various experiments. People belonging to this group should be of primary interest to any potential anticorruption interventions (
This study was carried out in accordance with the recommendations of “Ethical Principles of Psychologists and Code of Conduct by American Psychological Association” with informed consent from all subjects. The study does not involve vulnerable populations, poses no risk for participants and uses no deception. All subjects gave informed consent in accordance with the Declaration of Helsinki. In accordance with Czech laws and institutional regulations of the University where the study took place, the protocol was exempt from the approval process by the IRB and it was not necessary to obtain written consent from the subjects.
Both authors designed the study and wrote the manuscript. ŠB coded the experimental task and analyzed the data. MV collected the data.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Results for a test of an additional preregistered hypothesis concerning the charity incurring a loss are not reported here for the sake of brevity and can be found at
Partial correlations controlling for the low bribe and high bribe conditions yielded virtually the same estimates in all participant-level analyses. Earnings for oneself were correlated somewhat more strongly with sincerity (
The honesty-humility scale contains an item specifically asking about taking a bribe (“I would never accept a bribe, even if it were very large”). When honesty-humility is analyzed without this question, it still predicts participants’ earnings (
The estimates can be interpreted as effects of the increasing from the minimum value of a given variable to the maximum.
OR – odds ratio.
Including random slopes for bribe sizes also did not significantly improve the base model without any random slopes.
The reaction times depend partly on a momentary state of a participant. For example, if participants do not pay much attention for a while, this will result in slower reaction times for several trials. The reaction times of subsequent trials are therefore usually correlated with each other more than with trials that are further away. Inclusion of reaction times on preceding trials is used to remove the resulting autocorrelation of residuals. The reported analyses yield virtually the same results as the analyses without inclusion of the reaction times for preceding trials as covariates. The only exception are the effects of trial order and charity earnings at the point of the decision, which are both themselves autocorrelated and are therefore reduced when the additional covariates are included.
However, the participants were not informed of the nature of the bribes’ origin.