MSc ECONOMICS
BEHAVIOURAL ECONOMICS AND GAME THEORY TRACK
MASTER THESIS
The Influence of Social Norms on Honesty, Fairness and
Redistribution Preferences.
By
MEGAN ROUX
11623659
14 August, 2018
Supervisor: dr. Ivan Soraperra
Abstract
Inequality and Corruption are currently central and interrelated issues that have become a popular area for discussion and debate. This paper experimentally studies the effect of society- level social norms on individual levels of honesty, fairness views and redistribution preferences when it is possible that income is earned through dishonesty means. The experiment is conducted with participants from three countries with varying corruption levels. The United Kingdom has the lowest Corruption Perception Index (CPI) in our sample. This is followed by the United States of America and then India, which has a high CPI. The experiment has two parts. The first part involves an effort-task. There are two treatments: in one the scores are automatically recorded and in the other are self-reported by participants who have the chance to lie. The results confirm that people from countries with higher levels of corruption (India), are more dishonest. The second part of the experiment involves a third party redistributing the points earned by participants in the first part. Redistribution increases greatly when there is a chance that the points were earned dishonestly. This increase is the greatest among the Indian population. They are also the population group
Statement of Originality
This document is written by Megan Roux who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no
sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of Contents
1. Introduction 3
2. Existing Literature 6
3. Methodology Experiment design 9
A Players 10 B Players 12 Participants 13 Payoff 15 Research question 15 Hypotheses 15
4. Results A Players’ dishonest behaviour 17
B Players’ redistribution behaviour 20
B Players’ beliefs 26
5. Discussion 30
6. Reference List 34
7. Appendix Tables and Figures 36
1. Introduction
Corruption and inequality are widespread issues which currently demand a lot of attention in economics and politics, both in academia and in policy debate. There is a widening gap between the rich and poor in many societies around the world, resulting in dissatisfaction and tension between the have and the have-nots. Corruption is a persistent issue which undermines economic growth, environmental protection efforts and political stability (Banuri & Eckel, 2012).
Corruption and inequality are often linked. Corruption can negatively impact equality of income, (Banuri & Eckel, 2012). There are many examples, such as theft or tax evasion, where immoral or illegal behaviour has led to undeserved wealth. There is in fact a correlation between country level inequality, measured by the Gini coefficient, and perceived level of corruption, measured by the Corruption Perception Index (CPI) (Bortolotti, Soraperra, Sutter, & Zoller, 2017). Internationally there is a negative correlation between income inequality and social trust, (Algan & Cahuc, 2014). Oishi, Kesebir and Diener (2011) found a similar relationship within America, Americans trusted people less and believed people to be less fair in years that had higher national income inequality. Furthermore, how income is earned is shown to impact fairness concerns, (Alesina & Angeletos, 2005). Views about what is fair may change if there is mistrust about how ones fortune arose, (Bortolotti et al., 2017).
Although governments and organizations can implement checks to ward against corruption or dishonest behaviour, observation and control can be costly and there are many situations which require intrinsic honesty. Transparency International (2018) claim that “Corruption corrodes the social fabric of society”. That it weakens trust in the system and creates a “distrustful and apathetic public” which then makes corruption even harder to challenge. (Uslaner & Badescu, 2004) show there is a correlation between trust and corruption. Rothstein (2001) argues that by reducing corruption we can increase trust by creating trust “from above”. There is indeed evidence that societal and social norms have been shown to greatly affect individual honesty and that countries with higher levels of corruption and incidence of rule violation present higher levels of individual dishonesty, (Gächter & Schulz, 2016). Among these countries there is also less willingness to punish dishonesty, (Cameron, Chaudhuri, Erkal, & Gangadharan, 2009). Tolerance of inequality has also been shown to be related to the societal norm, (Roth & Wohlfart, 2016).
This paper studies how social norms impact the relationship between honesty and redistribution if income is possibly earned through dishonest methods. The central research question this paper seeks to answer is:
What are the effects of cultural and social norms on honesty and redistribution decisions if cheating was an option?
In order to study this effect, four testable hypotheses were set up. In short, the first two hypotheses describe that because individual honesty is influenced by societal social norms, there will be a higher level of cheating, and more suspicion of cheating, among participants from countries with higher corruption levels. The other two hypotheses concern redistribution preferences. The first is that baseline redistribution preferences (redistribution decisions made based on honestly earned income) are affected by social norms. Roth & Wohlfart (2016) analyse general social surveys in Europe and the United States of America and suggest that people who have experienced more inequality become more accepting of inequality and are less likely to favour redistribution policies. Based on this we expect that participants from countries with higher inequality may redistribute less often, at least in the baseline condition. Lastly, it is expected that when given a chance to redistribute between two people who had the opportunity to cheat, participants from countries with higher levels of corruption will redistribute more often because they are more likely to suspect cheating occurred. However, there is a possibility that the opposite occurs. That although participants from countries with higher corruption levels may suspect more cheating, they may also be more accepting of cheating and so could redistribute less often not wanting to punish the suspected cheaters. This study focuses on cultural and social norms and is set up as a cross-country study. Given time and cost constraints, only 3 countries are included in the study. The United States of America (USA), the United Kingdom (UK) and India were selected because of their varied corruption levels. Corruption is measured by Transparency International (2017a) using the corruption perception index, or CPI. The United Kingdom has the lowest corruption of the three counties, followed by the USA and India, which has reasonably high corruption. Further discussion about the corruption and inequality levels of these countries is included in the methodology section.
In the experiment there are two broad groups of participants: A players and B players. A players participate in a real effort task. They are divided into two treatments: Honest and Self-Report. In the Honest treatment the computer calculates the player’s points based on their actual performance and in the Self-Report treatment the players report their performance and are awarded points accordingly. This is an effort to mimic real-world scenarios where workers might overstate their performance to earn a bonus or promotion. A mean comparison of the scores found that overall cheating indeed takes place as scores in the Self-Report treatment are significantly higher. Cheating is also found to be the most pervasive among the Indian group, followed by the American group. No cheating is found to occur among the participants from the UK. This offers support for the first hypothesis that country wide corruption levels could impact individual honesty.
The second part of the experiment involved participation by the B players. They performed a redistribution task, redistributing the points earned by two A players who were from the same country as they were. Again, they were divided into the two same treatments. B players in the Honest treatment redistributed points between two A players who were in the Honest treatment and B players in the Self-Report treatment did the same task but between A players who were in the Self-Report treatment. B players were completely informed about the task that A players performed and the rules they followed. It was found that in general, redistribution rates increased substantially in the Self-Report treatment. They followed the same pattern as in the scores in the first part of the experiment. Redistribution increased the most among Indian participants, followed by American participants. Redistribution rates among the English participants did not change significantly. This offers support for the hypothesis that participants from countries with higher levels of corruption, redistribute more often if cheating was a possibility.
The rest of the paper is structured as follows. Section 2 provides a summary of existing, related literature. Section 3 describes the methodology; the experimental design and data collection. Section 4 presents the results of the experiment for A and B participants. Lastly section 5 offers a discussion of the results and limitations and concludes the paper.
2. Existing Literature
Why are people dishonest? Standard economic theory predicts that rational individuals will be dishonest if the benefits of the act outweigh the costs. However, during experiments the majority of participants have been shown not to cheat or at least, not to cheat to the full extent, (Rosenbaum, Billinger, & Stieglitz, 2014). This occurs even if there is no chance of being caught. Why? From a psychological perspective, an important input to the decision to be dishonest is the internal reward. People internalize the norms and values of their society and compare their own behaviour to this internal benchmark, (Henrich et al., 2001; Mazar, Amir, & Ariely, 2008).
Therefore, there should be a link between the norms of the society in which an individual lives and their intrinsic honesty. A cross-cultural study by Gächter & Schultz (2016) showed that societal institutions and cultural values did impact individual intrinsic honesty and rule following. In societies with a higher prevalence- and less punishment- of rule violations, dishonesty is considered more justifiable in certain situations. They found, using an anonymous die-rolling task, that individuals from countries with a higher prevalence of rule violation reported higher scores. They conclude that people “benchmark their justifiable dishonesty with the extent of honesty they see in their societal environment.”
Individual attitudes towards dishonesty also impact how one responds to dishonesty in others. The false consensus effect describes how individuals project their behaviour onto others, (Ross, Greene, & House, 1977). Irlenbusch & Ter Meer (2013) showed that people who engage in a certain behaviour are more likely to believe that someone else has engaged in that behaviour too. They studied this in the context of a repeated public goods game and found that participants who contributed more were also more likely to believe the self-reported high contribution levels of others. These inflated beliefs about the contributions of others reduced the likelihood of punishment. Although not about cheating specifically, these findings suggest that people from countries with high corruption, who act more dishonestly, may also suspect more cheating from their peers. Furthermore Uslaner & Badescu (2004) show there is a correlation between trust and corruption levels in a society. Therefore people who live in a country with higher corruption may trust their peers less.
Societies which have lower levels of corruption and rule violations have shown a higher propensity to punish this sort of behaviour. Cameron, Chaudhuri, Erkal, & Gangadharan (2009) conducted an experimental cross-cultural study, among university students, about bribery and punishment thereof. They found much higher rates of bribery among the participants from India (ranked by the CPI as having high levels of corruption) than the ones from Australia. Participants from India also showed a correspondingly lower willingness to engage in costly punishment against those players who participated in bribery.
Another important issue surrounding corruption and dishonesty is the effect this has on inequality. There are many cases where wealth is earned through unethical or illegal methods. Examples include tax evasion or exploitation of workers and cutting-corners in business to boost profits. There is a correlation between inequality, measured by the Gini coefficient, and the CPI, (Bortolotti et al., 2017). Globally, there is also a negative correlation between income inequality and social trust, (Algan & Cahuc, 2014). A study done within America also found that in years that had higher national income inequality, people trusted their peers less and believed people to be less fair, (Oishi et al., 2011). Additionally, how income is earned affects fairness views with respect to redistribution, (Alesina & Angeletos, 2005). It is considered fairer if it is believed that individual effort has determined income rather than if income was determined by luck or corruption. Furthermore, the fairer a participant believes an outcome to be, the less they redistribute, (Alesina & Angeletos, 2005). How much a person can be held responsible for their fortune, or indeed misfortune, plays a big role in whether it is considered fair. Bortolotti et al., (2017) showed in a redistribution experiment that redistribution increased if there was a chance that someone had earned their income dishonestly.
Redistribution from a participant that earned a high payoff, but who may have done so dishonestly, to a participant who earned a low payoff, but who most likely did so honestly, can be considered a sort of punishment to the dishonest player. Participants who are from a more corrupt country will, according to the false consensus effect and research in this area, be more likely to suspect a high payoff was the result of dishonesty. As already explored by Bortolotti et al., (2017) if someone suspects fortune has been earned illegally, they should be more likely to redistribute in favour of the honest party. Thus, participants from corrupt countries, because they suspect more corruption, should opt for more redistribution. However the study by Cameron et al., (2009) revealed that people from more corrupt countries are more tolerant of corrupt activity and so punish these activities less often. If this is the case, they should
redistribute less because they find the corrupt activity more acceptable. This experiment is designed to disentangle which of these effects are stronger.
Another thing that can influence an individual’s level of inequality aversion, and the extent to which they favour redistribution, is the level of inequality to which they are accustomed. A global survey revealed that individuals who in their lives have experienced higher levels of inequality are more accepting of it, and demand less redistribution, (Roth & Wohlfart, 2016). This study recorded and used the level of inequality that prevailed during what they call respondents’ “impressionable years", when they were between 18 and 25 years old, as an independent variable and controlled for current economic conditions. This suggests that there is some causation between the inequality levels to which people are accustomed and their redistribution preferences. Thus, participants from countries with high inequality levels could in this experiment favour less redistribution.
3. Methodology
Experiment design:
The experiment made use of a between-subject design. Participants were randomly assigned to one of four different groups. Participants were assigned to the role of Player A or Player B. They were also placed in either the Honest or the Self-Report treatment
A players participated in an effort task for which they earned points based on their performance. Those allocated to the Honest treatment had their answers automatically checked by the computer. Those in the Self-Report treatment simply reported which questions they answered. They therefore had the opportunity to lie and over-report their performance.
B players had the opportunity to redistribute the earnings of two A players. B players in the Honest treatment redistributed between two A players who were also in the Honest treatment and B players in the Self-Report treatment redistributed between two A players who were in the Self-Report treatment.
To summarize, the figure below shows the four different types of participants in the study and briefly describes their roles. Further details of the roles will be offered in the next section. Information on the number of participants assigned to each of these roles, within each country, can be found in table A-1 of the appendix. Also available at the end of the appendix, are all instructions, tasks and questions that were given to the participants during the experiment.
Figure 1: Participant types
Roles
Player A Player B
Treatments:
Honest
Effort task: the computer automatically checks answers
and gives a score.
Redistribution of points: earned by two A players in
the Honest treatment.
Self-Report Effort task: participants
themselves report their score.
Redistribution of points: earned by two A players in
the Self-Report treatment.
A Players
A players participated in a real-effort task, used in an experiment on honesty by Mazar et al., (2008). Participants were provided with 12 matrices, each with 12 numbers (e.g. 6.48). Two of the numbers in each matrix summed exactly to 10. Participants had to find these two numbers. They were given three minutes to solve as many matrices as possible, after which they were directed to the next screen. In the Honest treatment subjects indicated which of the two numbers they chose in each matrix. The computer checked their answers and allocated their points accordingly. In the Self-Report treatment subjects did not have to input a solution but simply indicated which matrices they had solved. Players were told to report the outcome truthfully, although the experimenter could not verify their claim and they were allocated points based on the number of matrices they reported as having solved.
For every 3 matrices solved (or reported as solved), players received 1 point. At most they could earn 4 points if they solved all 12 matrices. For simplicity, henceforth the number of matrices that an A player solved will be referred to as their score. The table below shows how a player’s score was converted into points.
Table 1: Points earned from matrices solved Matrices Solved (score) Points earned
1, 2 0
3, 4, 5 1
6, 7, 8 2
9, 10, 11 3
12 4
At the end of the experiment, points were converted into money. Purchasing Power Parity was used in order to keep incentives comparable across countries. More detail is available in the payoff section of this methodology.
An effort task was chosen in order to mimic many real-world situations where people may misreport their effort level or ability. For example, a worker who misreports the number of hours they have worked in order to be paid more or increase their chances of a promotion. The fact that the player’s report cannot be verified by the experimenter mimics the real-world fact that often monitoring of workers is extremely costly. This particular effort task is useful because it is relatively simple and can be easily performed by people in all parts of the world because it does not depend on any knowledge or language skills which might be specific to different regions. Another task which is commonly used to test for dishonesty is the die rolling task. This task is useful because it is easy to understand, can be conducted quickly and requires participants to make a simple non-strategic decision. It also allows for gradual dishonesty. With both these tasks you cannot verify individual dishonesty although analysis of the means and distributions of scores, between those in the Honest and Self-Report treatments, gives you information about dishonesty levels.
A disadvantage of the chosen task is that people may cheat out of efficiency concerns. One way to combat this would have been to have their cheating directly affect another participant, rather than just the experimenter. However, this introduces increasingly strategic decisions where individuals may base their decision to cheat on how honest they suspect the other player will be. While this in an interesting area for future research in a cross-cultural setting, it is not the topic of the study here.
Participants were then asked a series of questions about their age, socioeconomic status, gender, living situation, area of study and the extent to which they consider themselves religious. In addition, some questions on belief in honesty of others and individual norms of honesty from the world value survey were used, (Inglehart, R., C. Haerpfer, A. Moreno, C. Welzel, K. Kizilova, J. Diez-Medrano, M. Lagos, P. Norris, 2014).
B Players
B players had to decide if, and how, to redistribute the points earned between two A players. Half redistributed between two A players in the Honest treatment. The other half redistributed between two A players who were in the Self-Report treatment. B players were fully informed about the task A players performed and the rules they followed. They were also informed that the A players would be from the same country as they were.
The redistribution decisions worked in the following way: If one A player solved 3 matrices, they received 1 point. If the other solved 10, they received 3 points (see table 1). Player B then had the opportunity to redistribute a total of 4 points. They could do so in intervals of 0.5 points.
The experiment made use of the strategy method, a procedure whereby participants make a decision for every possibility. B participants therefore each made 14 redistribution decisions, one for every possible combination of points between A players. This allowed for data on all possible scenarios. Participants were not informed which decision would be relevant for the A players. Using the strategy method is advantageous because it enables collection of a range of data which can be transformed into panel data. It also protected against the possibility of only having a few A player responses and therefore only a small range of real scores in the effort task. B players were not paid for individual decisions but were paid a fixed rate, which was equal to the maximum amount an A player could receive. This prevents B players from feeling disadvantageous inequality aversion when comparing their payoffs to A players. Also, because their payoff is completely independent of their responses they had no reason to allow selfish motives to interfere with their decisions. The number of questions asked to participants was limited in order to avoid lazy or bored responses.
they think may have cheated. They also had to answer the same series of biographical and value-based questions that A players answered.
Participants:
This study focuses on cultural and social norms and so it is a cross-country study. Given time and cost constraints, only 3 countries were included in the study. The United States of America (USA), the United Kingdom (UK) and India were selected as the countries from which participants were recruited. As already discussed, these countries were chosen because they represent a spectrum of corruption levels. Transparency International is a global organization which researches and reports on corruption as well as works with governments to try and curb its influence. Their most common index to measure corruption is the Corruption Perception Index (CPI), (Transparency International, 2017a). For each country the CPI measures the perceived corruption of the public sector by analysts, business people and experts. The most recent data (2017) ranks the UK as 8th (with a score of 82 out of 100), the USA as 16th (75/100)
and India 81st (40/100) out of a total of 180 countries.
Transparency International (2017b) also conduct large surveys of regular citizens, ask about their experiences of corruption, and compile a global corruption barometer. Asked about suspected corruption in a variety of public sectors and institutions, the UK consistently report very low levels. Reported levels in the USA are consistently higher than those of the UK. In India, reported levels are higher in the vast majority of cases. For example, more than 40% of Indians report that most or all of their government officials and local councillors are involved in corruption. This is considerably higher than the approximately 22% of UK citizens, and 30% of Americans, who report this level of corruption. Moreover, 54% of respondents in India report that most or all of their police force are involved in corruption, this is as opposed to 11% in the UK and 27% in the USA.
For information on inequality levels, refer to the World Income Inequality database by UNU-WIDER (2017) and the World Inequality Database (2014) . The most recent available data point by the World Income Inequality database (WIID) for India is 2011, so 2011 is used for comparison. In this year India recorded the highest Gini coefficient of 50.3, followed by the USA (Gini coefficient=38.9) and then the UK (Gini coefficient=33). However, it’s noted that inequality levels have risen steeply in the USA since 2011. The World Inequality Database
(2014) offers more recent data, measuring and comparing, among other things, pre-tax national income. Inequality levels in the three countries follow a similar pattern to corruption levels. The UK has the lowest inequality of these three countries, with a 40% share of income belonging to the top 10%. This is in comparison to 47% in the USA and 56% in India. The world average is 52%.
It is difficult in my study to disentangle the effects of inequality and corruption levels, because within my 3 country sample they are perfectly correlated. Also, other than their corruption and inequality levels, there are many differences between these three countries and it is hard to measure the full range of unobservable factors that might influence decision-making. In order to control for this problem as much as possible, a number of demographic and other questions are asked about trust and community. However, this by no means solves the problem and I am careful not to overstate my results.
The experiment was set up and conducted using Qualtrics and participants from each of these countries were recruited via Amazon Mechanical Turk (MTurk). A total of 345 participants were recruited, 114 from India, 123 from the USA and 108 from the UK. For more information about the participants, their allocation to the various treatment groups, and their socio-demographics refer to table A-1 in the appendix.
MTurk is increasingly popular among social scientists as a data collection technique, (Litman, Robinson, & Abberbock, 2017). It proved a useful tool for this study for a number of reasons. Firstly, it allows MTurk workers more privacy than in the laboratory which is especially important given the topic of study, in particular for A players. Participants in the Self-Report treatment knew that there was no possibility the experimenter could determine if they were dishonest in their reporting. Secondly, it allowed for data collection from participants from various countries. The tasks that participants performed were relatively simple and short so concerns about understanding were minimal. Monetary incentives were fairly low, as is shown in the following section, so there were concerns about effort and concentration. To address this problem, a number of questions were inserted throughout the study to check concentration. If participants answered the questions incorrectly, they were directed to the end of the study and their responses were excluded.
Payoffs:
All participants who completed the study were paid the same base payment of $0.20. Additionally, 3 participants from each country were randomly selected to earn a bonus payment that was dependent on their performance and choices. Within each country, two A players were randomly matched to a B player and their final payoffs were determined by the B player’s redistribution decisions. The bonus payments were standardized by purchasing power so that incentives were the same for all participants in all the countries. The table below shows exact dollar payoffs for different roles arranged according to country.
Table 2: Participant Dollar Payoffs
India United States of
America United Kingdom
A Players (per point earned after
the final allocation by Player B)
$1/ point $3/ point $3/ point
B Players $4 $12 $12
Research Question:
The experiment was designed in order to answer the following research question:
What are the effects of cultural and social norms on honesty and redistribution decisions if cheating was an option?
In order to study this effect, a number of testable hypotheses were set up. Hypotheses:
1) Individual honesty is influenced by the societal social norm. Thus, there will be a higher level of cheating among participants from countries with higher corruption levels.
The effort task the A players completed was designed to test this hypothesis. Mean comparisons of scores between the Honest and Self-Report treatments are used to measure cheating behaviour.
2) Participants from countries with higher levels of corruption will suspect higher levels of cheating among their peers.
B players were asked to predict how many A players, out of 100, may have cheated. These responses are used to test this hypothesis.
3) Baseline redistribution preferences (redistribution decisions made based on honestly earned income) are affected by social norms. Participants from countries with higher inequality will redistribute less.
This hypothesis is inspired by Roth & Wohlfart (2016) who analysed general social surveys in Europe and the United States of America. Their findings suggest that people who have experienced more inequality become more accepting of inequality and are less likely to favour redistribution policies. Analysis of the choices made by B Players in the Honest treatment tests this hypothesis.
4) When given a chance to redistribute between people who had the opportunity to cheat, participants from countries with higher levels of corruption will redistribute more often because they are more likely to suspect cheating occurred.
There is also a possibility that although participants from countries with higher corruption are more likely to suspect cheating occurred, they may also be more accepting of dishonest behaviour. This could result in them redistributing less often. A comparison of the B players redistribution choices between the Honest and the Self-Report treatments will test this hypothesis.
4. Results
In this section, the results of the experiment are presented, starting with the behaviour of A players followed by the redistribution choices of the B players.
A Players’ dishonest behaviour
Result 1: Many participants cheat in the Self-Report treatment. There is a significant increase in the mean score between the Honest and the Self-Report treatment. However, participants do not cheat to the full extent, many in the Self-Report treatment report low scores.
In Figure 2, Overall shows the mean scores of all participants in the Honest and Self-Report treatments. There were 149 A players in total, 74 were in the Honest treatment and 75 were randomly assigned to the Self-Report treatment. The mean score for participants in the Honest treatment was 6 and in the Self-Report treatment, 7.28. The difference is significant at a 5% level, (t-test p=0.013). This is evidence that cheating took place in the Self-Report treatment. In addition, Figure 3 shows the distribution of scores in the Honest and Self-Report treatment. A Mann-Whitney test comparing the Honest and Self-Report treatments (see Overall in Table 3) reveals that the medians of the two distributions are significantly different (p=0.018). However, it is also clear that participants do not cheat to the full extent possible. The mean score for the Self-Report treatment is only 7.28, this is far from the maximum score of 12. Within Figure 3, Overall also reveals that 25% of scores are 4 or lower. For robustness, regression analysis is offered in Table A-2 of the appendix. Model 1 shows that there is a significant increase in the scores in the Self-Report treatment (OLS regression p=0.013).
Figure 2: Mean Scores by Treatment Group
Table 3: Mann-Whitney Test on Scores
Nationality Group N Rank Sum Expected P-value
American Honest 27 625.5 742.5 0.042**
Self-Report 27 859.5 742.5
English (UK) Honest 20 460 440 0.624
Self-Report 23 486 506
Indian Honest 27 613 715.5 0.058*
Self-Report 25 765 662.5
Overall Honest 74 4930 5550 0.018**
Self-Report 75 6245 5625
Notes: Mann-Whitney test comparing the distribution of scores of the Honest and Self-Report groups of each nationality. The results reveal that for the American and Indian groups there is a significant difference between the medians of the Honest and Self-Report treatments. There is no significant difference between the two English (UK) treatments. *P<0.10, **P<0.05, ***P<0.01.
Notes: Bar graph showing the mean scores divided into nationality and treatment groups.
Overall represents the mean score not yet divided into groups. T-tests were conducted to
compare the means. *P<0.10, **P<0.05, ***P<0.01.
Figure 3: Distribution of Score by Treatment Group
Result 2: There are higher rates of cheating among Indian participants than among American participants, and these rates are both higher than among English (UK) participants. The largest, and most significant increase in scores is between the Honest and Self-Report treatments for Indian participants, followed by American participants. Again, in none of the three countries do participants cheat to the full extent.
Figure 2 also shows the mean scores of participants in the two treatments, separated into their nationalities. From this figure it can be seen that the largest, and most significant, increase in scores between the two treatments, is among the Indian participants. The mean score increases from 4.85 to 7.32, a significant increase of 2.47 (t-test p=0.014) among the Indian population. By contrast, in the American population scores only increase by 1.52 (t-test p=0.053). Among the UK population, the mean score actually decreases, by 0.476, in the Self-Report treatment. Although this decrease is not significant (t-test p=0.567) this result does indicate that the incidence of cheating in the UK was very low or non-existent. Figure 3 also shows the median score in the UK decreased and a Mann-Whitney test (table 3) reveals no significant difference between the two median scores (p=0.624). No participants in the UK Self-Report group reported scores higher than 10. On the other hand, the Mann-Whitney test (table 3) revealed significant differences in the medians of the distributions of scores between the Honest and
Notes: Box Plots showing the distribution of scores between the different nationalities and treatments. Overall represents the scores before they are divided into the country groups.
Self-Report treatments for both the US (p=0.042) and Indian participants (p=0.058). Figure 3 indicates that the median score increased in the Self-Report treatment for both those groups.
Again, for robustness, refer to the regression analysis offered in Table A-2 of the appendix. Dummy interaction terms of Indian and American participants in the Self-Report treatment, reveal that there is a significant increase in scores in both these cases (OLS regression, p=0.023 and p=0.078, respectively). The increase is again larger and more significant for Indian participants in the Self-Report group than for Americans. This is true in model 2 which offers no biographical controls, and in model 3 where the controls are added (OLS regression, p=0.059 and p=0.079, respectively). It is also worth noting the significantly lower scores (OLS regression p=0.023) of the Indian participants in the Honest treatment. This is shown in model 2 by the coefficient of the Indian dummy variable. This is possibly caused by low effort levels.
B Players’ redistribution behaviour
There were 196 B players who participated in the experiment. Each player made 14 redistribution decisions, which means in total there were 2, 744 redistribution decisions. In 784 of these decisions, both A players had the exact same number of points to begin with. Only in 8.3% of these instances did any redistribution take place. In 1,960 cases there was inequality to begin with. These are the cases analysed and from which a panel data set was developed. In 5% of instances, B players increased inequality. In 28.2%, B players reduced inequality and in the majority of cases, 66.8%, B players chose not to redistribute at all. Graph A-1 in the appendix shows this ratio as a pie chart.
Figure 4 and 5 display the average extent of inequality reduction among the various treatment groups. The vertical axis shows the level of redistribution by using a measure called the Inequality Reduction (IR) index. This was a measurement technique used by Bortolotti et al., (2017). IR is defined by the following equation:
𝐼𝑅 = 1 −|𝜋𝑝𝑜𝑠𝑡
1 − 𝜋
𝑝𝑜𝑠𝑡2 |
|𝜋𝑝𝑟𝑒1 − 𝜋 𝑝𝑟𝑒2 |
Where 𝜋𝑡𝑖 are the points for A player 𝑖 = {1,2} before or after (𝑡 = 𝑝𝑟𝑒 or 𝑡 = 𝑝𝑜𝑠𝑡) the redistribution by the B player. If 𝐼𝑅 = 0, no redistribution took place while if 𝐼𝑅 = 1, player B effected an equal split of the earned points. If 0 < 𝐼𝑅 < 1, the B player reduced the original inequality. The IR index can be interpreted as the percentage of the initial inequality that has been eliminated. Negative values occur if initial inequality was increased by the B player.
Result 3: There is more redistribution, greater inequality reduction, in the Self-Report treatment where the participant may have earned their income from cheating. This occurs more often in situations where the high-scorer is suspected of cheating.
In figure 4, notice firstly that overall the degree of inequality reduction increases between the Honest and Self-Report treatments. This is verified by model 1 in table 4 which shows there is a significant increase (OLS regression p=0.011) in the aggregate level of the IR index between the Honest and Self-Report treatments.
Inequality is, in general, reduced the most for pairs where one participant earned 0 and the other earned 2 or higher. In these cases, inequality is reduced by between 20% and 30%. This is in line with reports from many participants who wrote that they aimed to leave scores as they were but wanted to ensure that even weak participants, those who scored 0, received some points. With all these pairs, there were only very small increases in inequality reduction compared to the Self-Report treatment. This is an interesting result seeing as scores of 0 are almost guaranteed to be honest responses and B participants in the Self-Report treatment could have used this opportunity to reward these participants for their honesty. However, there is little evidence that this occurred.
For three pairs of scores in the Honest treatment, negative IR values are observed. These indicate increased inequality. This can be attributed to a few things. Although only 5% of the 1,960 redistribution decisions increased inequality, these absolute values are much larger (sometime as big as -6). This can greatly skew the means. A few participants who chose to increase inequality wrote that they aimed to give an extra bonus to participants who did particularly well in the task. Others claimed to have redistributed entirely randomly. The possibility that participants made random decisions, due perhaps in part to the fact that their
decisions did not impact their payoff, is a limitation in this study and adds some noise to the data.
Regardless, there is still a strong response to possible dishonesty. Some of the largest increases in IR levels are for pairs where one participant scored 4 points. When asked who they suspected of cheating, 50% of participants in the Self-Report treatment responded that they were suspicious of either scores’ higher than 3, or perfect scores (in other words, 4 points). This shows that when a participant suspects that a high score was a result of cheating they are more willing to take points away from the high-scoring A player.
Figure 4: Redistribution Levels across Pair Groups and Treatments
Notes: The Inequality Reduction index (IR) compares the level of inequality pre and post redistribution. IR=0 represents no redistribution, IR=1 indicates that B players redistributed points equally. IR<0 indicates that participants increased inequality. On the horizontal axis we see the different combinations of points by A players. Overall represents the overall IR levels, before they are separated into pair types or nationalities. T-tests were conducted to compare the IR indexes between the treatments within each pair. *P<0.10, **P<0.05, ***P<0.01. A t-test could not be conducted on the overall group because it includes repeated observations from the same subjects,
Table 4: Panel Data Analysis of Redistribution Decisions
Dep. Var: IR Model 1 Model 2 Model 3
Self-Report treatment (d) 0.120** (0.047) -0.026 (0.078) 0.116 (0.096) Indian (d) -0.025 (0.092) -0.035 (0.094) American (d) -0.068 (0.080) -0.093 (0.078) Self-Report × Indian 0.259** (0.116) 0.235** (0.116) Self-Report × American 0.190* (0.110) 0.144 (0.110) Male (d) -0.090* (0.052) Age 0.001 (0.003) Middleclass (d) -0.017 (0.050) Urban (d) -0.088 (0.076) Economics (d) -0.038 (0.050) Religious (d) 0.088 (0.057) Center (d) -0.060 (0.058) Right (d) -0.032 (0.071) Community (d) -0.047 (0.046) Income Equality (d) 0.096** (0.047) Trust (d) 0.040 (0.072) Self-Report × Trust -0.215** (0.092) Fairness (d) -0.024 (0.063)
Individual Honesty norms 0.008
(0.145) Constant 0.074** (0.033) 0.106 (0.067) 0.175 (0.121)
Group variable Response ID Response ID Response ID
No. of observations 1,960 1,960 1,960
No. of groups 196 196 196
Prob > chi2 0.0114** 0.0045*** 0.000***
Notes: OLS regression with individual level random effects. These results are estimated using panel data generated using the many redistribution decisions each participant made. *P<0.10, **P<0.05, ***P<0.01.
Result 4: In the Honest treatment there are no significant differences in the IR across nationalities. Therefore, there is no evidence that a countries’ inequality levels affect individual preferences for redistribution, at least when income (or in this case, points) are earned fairly.
Figure 5 shows the mean IR level for the three nationalities in the Honest and Self-Report treatment. Firstly, refer to the Honest treatment. These values represent baseline preferences for inequality reduction when income, (or ‘points’, in the context of this experiment) is earned fairly, based on effort or ability but not on luck or dishonesty. Small differences between the nationalities are observed. The largest change is between the American and English (UK) populations. Americans’ inequality reduction is the lowest between the 3 nationalities at only 4%, while it is the highest, at 11%, for the English (UK) group. All this is despite the fact that, as discussed in the introduction, India records the highest country-wide Gini coefficient and inequality rates. Previous research (Roth & Wohlfart, 2016) concludes that individuals who live in societies with high levels of inequality, prefer lower levels of redistribution. This study does not provide evidence that prevailing society-level inequality rates impact individual preferences for inequality reduction. Model 2 and 3 in table 4 provide support for these conclusions. In neither case are the dummy variables for the Indian (OLS regression p=0.783 in model 2 and p=0.712 in model 3) and American (OLS regression p=0.394 in model 2 and p=0.234 in model 3) participants in the Honest treatment significant.
Figure 5: Redistribution Levels across Nationalities and Treatments.
Result 5: The largest, and most significant increase in inequality reduction, between the Self-Report and Honest treatment, is for Indian participants, followed by American participants. Among the UK population, inequality reduction is very similar between the Honest and Self-Report treatments.
When cheating is possible, there are far higher aggregate levels of inequality reduction in the American and Indian populations. As shown in figure 5, in the Honest treatment, the average inequality reduction among Americans is 4%, this increases to 20% in the Self-Report treatment. In the Indian group, average inequality reduction increases from 8% to 31%. Within the English (UK) group, there is a very small decrease, from 11% to 8%, between the two treatment groups. Table 4 offers some statistical support. The interaction dummy for Indian participants in the Self-Report treatment is large and highly significant in Model 2 (OLS regression p=0.026). The results are robust after introducing a wide range of socio-demographics control variables in Model 3 (OLS regression, p=0.042). The interaction dummy
Notes: The Inequality Reduction (IR) index compares the level of inequality pre and post redistribution. IR=0 represents no redistribution, IR=1 indicates that B players redistributed points equally. The bar graph shows the mean IR index within each nationality and treatment group. T-tests could not be conducted on this sample because each participant makes multiple decisions. To test for significance we look at the OLS regressions done on the panel data set in table 4.
for American participants in the Self-Report treatment is also large and significant in Model 2 (OLS regression, p=0.085). However, once we introduce the Model 3 control variables it is no longer significant (OLS regression, p=0.191). This implies that perhaps there is another variable that controls the increase in inequality reduction in the American population. Trust is a dummy variable which takes value 1, if participants answered in the survey that “Most people can be trusted”. It takes a value of 0, if participants answered that they “Need to be careful” when dealing with people. The interaction term of Trust and those in the Self-Report treatment is large, negative and significant (OLS regression, p=0.020). This indicates that participants, in the Self-Report treatment, who are more trusting, redistribute less when cheating is possible. The majority of Americans (see table A-1 in the appendix) reported that when dealing with people you “Need to be careful”. Only 38.2% of the American sample say that most people can be trusted. This is the lowest level among the three nationalities. It is conceivable then that this relationship is part of what drives the large, significant coefficient on Americans in the Self-Report treatment and is why the coefficient is no longer significant in Model 3.
Participants from India not only appear to cheat the most frequently but also redistribute the most in the Self-Report treatment. The pattern is the same for the American and the English (UK) participants, for example, the English (UK) participants cheat the least, and redistribute the least. This appears to indicate that participants have correct beliefs about the levels of cheating in their countries. The next section will discuss the beliefs of the B players.
B Players’ beliefs
B players were asked some questions to assess what they believed about the behaviour of the A players. They were asked how many A players, out of 100, they suspect may have cheated as well as how acceptable, on a scale of 1 to 10, they thought cheating was in this study. It is worth nothing however, that beliefs were not incentivized.
Result 6: Indian participants suspect the highest level of cheating, this is followed by the American and then English (UK) participants. Indians also consider cheating more acceptable in this study than the other nationalities.
38% cheated, Americans suspect 48% and Indians suspect 56%. The difference in the mean scores, between the English (UK) and the Indian group, is highly significant (t-test, p=0.006). The other differences between the other nationality group were large although not significant (t-test mean comparison between American and English (UK) groups gives a p=0.112, t-test mean comparison between American and Indian groups gives a p=0.217).
Table A-5 (See appendix) offers a statistical analysis of these beliefs. It can be seen that the effect of the dummy for being an Indian participant, is large, positive, significant and robust even when a variety of controls are added, across Models 1 (OLS regression, p=0.006), 2 (OLS regression, p=0.046) and 3 (OLS regression, p=0.077). The dummy for American participants is not robust and although it is significant in Model 2 (OLS regression, p=0.042), it is not in Model 3 (OLS regression, p=0.186), when the Trust dummy is added. The dummy for trust is once again, large, negative and very significant (OLS regression, p=0.022).
Members of the Indian population also reported that cheating in this study was more acceptable than did the other two nationalities. Figure 7 illustrates this. There is a significant difference is the means between the American and Indian groups (t-test, p=0.042). However OLS analysis (table A-5 in the appendix) of these results show no significant effect of being Indian on reported acceptability of cheating (OLS regression, model 2 p=0.562 and model 3 with controls, p=0.108).
Figure 6: The Percentage of A Players that the B Players Suspected of Cheating.
Figure 7: Reported Acceptability of Cheating by Nationality
Notes: B players in the Self-Report treatment were asked how many A players, out of 100 they suspected of cheating. The bar graph records the means and t-tests were
conducted to compare the means. *P<0.10, **P<0.05, ***P<0.01.
Notes: B Players in the Self-Report treatment were asked how acceptable they considered cheating in this experiment. Their answers were on a scale of 1 to 10, where
Result 7: Suspicions of cheating have no observable effect on redistribution decisions in the Self-Report treatment. Reported acceptability of cheating also has no effect, with one exception. Among the Indian Self-Report group, there is a small negative relationship between how acceptable cheating is considered and the level of inequality reduction.
Table A-3, in the appendix, shows that the percentage of participants that B Players (those in the Self-Report treatment) suspected of cheating has no effect on redistribution decisions. The variable Suspected Cheating, which measures the percentage of A players whom B players thought might cheat, has an extremely small and insignificant effect on the IR index in the Self-Report treatment (OLS regression, model 1 p=0.435). Between the nationalities the level of cheating (see figure 2) and level of suspected cheating (see figure 6) follow the same patterns and so it would appear that people have correct beliefs about cheating in their country. Inequality Reduction in the Self-Report condition again follows the same pattern (see figure 5). However there is no clear indication that suspected cheating has a direct effect on inequality reduction choices.
In Table A-4, in the appendix, the effect of reported Acceptability of Cheating on the IR Index in the Self-Report treatment is shown. Although the variable itself is not significant (OLS regression, model 2 p=0.830), the interaction term with participants from India is. There is a small, and significant, negative relationship within this treatment group (OLS regression, model 2 p=0.039). The more acceptable cheating is considered to be, within the Indian, Self-Report group, the less inequality reduction takes place. However this is not robust when more controls are added in model 3 (OLS regression, model 3 p=0.104) and although the Indian group find cheating more acceptable (figure 7), they overall redistribute more often (figure 5).
5. Discussion
Inequality worldwide is on the rise and it has led to a lot of debate, not only about what is efficient but also about what is ethical and what is fair. Fairness views impact decisions about redistribution and alleviation of inequality, (Alesina & Angeletos, 2005). Fairness means different things to different people and it depends a great deal on the situation and how the inequality arose to begin with.
There is evidence that earning one’s fortune through effort and hard work is considered fairer than if it arises through good fortune, (Alesina & Angeletos, 2005). Previous research has also shown that if income is potentially earned through dishonest methods it is considered less fair and more redistribution takes place, (Bortolotti et al., 2017).
This paper attempted to add to the existing research by looking at how societal social norms, specifically corruption and inequality levels, impact the existing findings. To this end, populations were sampled from three countries, the USA, the UK and India.
There are a couple of obstacles to interpreting the results. Firstly, there are a slew of unobservable characteristics and cultural differences between the countries of study. Many control variables were introduced into the regressions which look at, among other things; politics, trust and reported individual norms of honesty. It would be a mistake however to assume that this is an exhaustive list of the range of contributing factors. Secondly, inequality and CPI are perfectly correlated among the three countries chosen. It is therefore impossible to disentangle which had the major effect on participants’ redistribution decisions. To solve this problem, ideally there would have been four countries in the study that offer a combination of high and low corruption and inequality levels. However this was unfeasible. Caution is therefor employed when discussing results.
The first major finding is that cheating rates are higher among the Indian sample. India is also the country with the highest corruption levels. The increase in scores from the Honest to the Self-Report treatment is the greatest among the Indian group (t-test p=0.014). This is shown using a mean comparison in Figure 2. The American group also experiences a considerable and
English (UK) population, (t-test p=0.567). These results are supported using OLS regression and are robust even when a large variety of controls are added. This offers support for the first hypothesis, that there is more intrinsic honesty in countries with lower levels of corruption. Reported Individual Norms of Honesty by all the participants in the study reveal a similar trend. Participants were asked to rank how acceptable they considered certain immoral or illegal activities like cheating on a test or evading taxes. Their answers were converted into a score out of 10, where 10 would indicate extremely high norms of honesty. On average, UK participants scored 7.9, the American participants 7.8 and the Indian participants 6.2.
Of course, this effect could also be caused by inequality levels which follow the same pattern. There could also be other unobservable factors which drive this behaviour. Three countries provides a small sample to work with and so we cannot draw very strong conclusions. However these results do support the large cross-cultural study conducted by Gächter & Schulz (2016). They did honesty experiments in 23 countries and found a strong link between the prevalence of rule violations and individual honesty.
There was also a difference in how distrustful members of the different countries were. Participants in the Indian group suspected that over half of A players would cheat and over report their score. OLS regression analysis (Table A-5) shows that this result is robust (model 3, p=0.077). The American group suspected 48% and the English (UK) group suspected 38%. These results are displayed in Figure 6. These results support the second hypothesis that individuals who experience high corruption levels are more distrusting and would suspect higher levels of cheating. This is in line with findings by Uslaner & Badescu (2004) who showed that there is a correlation between corruption levels in a society and trust.
The third hypothesis in this study was about baseline redistribution preference. In this case ‘baseline’ refers to the Honest treatment where cheating could not take place. Previous research has indicated that inequality levels in a society could impact individual preferences for redistribution and that people who live in an unequal society often support less redistribution, (Roth & Wohlfart, 2016). Although India has the highest level of inequality among these 3 country groups, they demand, in the Honest treatment, more redistribution than the American group does. However, in general in the Honest treatment there are no significant differences in the redistribution preferences across nationalities, (OLS regression table 4 model 3, for the American dummy p=0.234 and for the Indian dummy p=0.712). Therefore, this paper presents
no evidence that a countries’ inequality levels affect individual preferences for redistribution. At least not when income (or in this case, points) is earned fairly, through effort and ability, and not through luck or dishonesty.
The second major finding is the change in redistribution patterns that occurs when income (or points) may have been earned in a dishonest manner. The level of inequality reduction, in general, increases in the Self-Report treatment. Many of the most significant changes happen when one participant scored 4 points. This is in line with reports from more than 50% of participants, that they were suspicious that a score of 4 was earned through cheating. This supports the findings of Bortolotti et al., (2017) who report that participants redistribute more if income was potentially earned dishonestly.
The level of inequality reduction increases the most among the Indian population. They redistribute significantly more in the Self-Report treatment (OLS regression table 4 model 2, p=0.026). There is also a significant increase among the American population (OLS regression table 4 model 2, p=0.085). The level of inequality reduction does not change significantly among the English (UK) population (OLS regression table 4 model 2, p=0.735). Figure 5 graphically shows this relationship. These findings support the claim that people from corrupt societies will redistribute more. Hypothesis four suggests that the reason for this is that people from more corrupt societies are more suspicious and are more likely to believe cheating occurred. In the sample, this is certainly true. Indian participants suspected a far higher percentage of participants cheated. However, OLS regression analysis (Table A-3) does not reveal that suspicions about cheating have any significant effect of redistribution decisions (OLS regression model 1, p=0.435).
It is therefore difficult to draw concrete conclusions about why there is more redistribution among the Indian group. This is further confused by the fact that participants in the Indian group report, to a greater extent than the other groups, that cheating in this experiment is acceptable. This can be seen in Figure 7. One might suppose that if cheating is considered more acceptable that you might redistribute less, given the fact that redistribution can be seen as a kind of punishment for cheating. Table A-4 uses OLS regression to study this effect and finds no robust significant relationship between individual rating of the acceptability of cheating and
To summarize, the possibility of cheating does increase redistribution and it does so more in countries that have higher levels of dishonesty and who are more likely to believe a high income (or score) is a result of cheating. However it is impossible to make firm assertions that it is the expectation of cheating that drives the redistribution decisions. What is clear from this study and previous research is that corruption has a serious negative impact on its citizens. Citizens can become more dishonest and less trusting of their peers. It can be even more detrimental, if it is found that this impacts on their willingness to punish dishonesty and corruption. However based on this research it does not seem necessarily to be the case. If individuals from corrupt countries are indeed willing to punish people they suspect of dishonesty, there is hope that citizens in countries with high corruption could push for change. Politicians who seek to fight growing inequality based on corruption might want to put more power into the hand of citizens who may demand redistribution. Further research, ideally with a larger set of countries, would help clarify the relationship.
6. References:
Alesina, A., & Angeletos, G.-M. (2005). Fairness and Redistribution. The American Economic Review, 95(4), 960–980.
Algan, Y., & Cahuc, P. (2014). Trust, growth, and well-being: New evidence and policy implications. Handbook of Economic Growth, Vol.2, 49-120.
https://doi.org/10.1016/B978-0-444-53538-2.00002-2
Banuri, S., & Eckel, C. (2012). Chapter 3 Experiments in Culture and Corruption: A Review, (May), 51–76. https://doi.org/10.1108/S0193-2306(2012)0000015005
Bortolotti, S., Soraperra, I., Sutter, M., & Zoller, C. (2017). Too Lucky to be True Fairness views under the shadow of cheating. Working Paper, (October), 1–55.
Cameron, L., Chaudhuri, A., Erkal, N., & Gangadharan, L. (2009). Propensities to engage in and punish corrupt behavior: Experimental evidence from Australia, India, Indonesia and Singapore. Journal of Public Economics, 93(7–8), 843–851.
https://doi.org/10.1016/j.jpubeco.2009.03.004
Gächter, S., & Schulz, J. F. (2016). Intrinsic honesty and the prevalence of rule violations across societies. Nature, 531(7595), 496–499. https://doi.org/10.1038/nature17160 Henrich, J., Boyd, R., Bowles, S., Camerer, C., Fehr, E., Gintis, H., & McElreath, R. (2001).
In search of Homo economicus: Behavioral experiments in 15 small scale societies. American Economic Review, 91(2), 73–78. https://doi.org/10.1257/aer.91.2.73
Inglehart, R., C. Haerpfer, A. Moreno, C. Welzel, K. Kizilova, J. Diez-Medrano, M. Lagos, P. Norris, E. P. & B. P. et al. (2014). World Values Survey: Round Six - Country-Pooled, (June), 1–21.
Irlenbusch, B., & Ter Meer, J. (2013). Fooling the Nice Guys: Explaining receiver credulity in a public good game with lying and punishment. Journal of Economic Behavior and Organization, 93, 321–327. https://doi.org/10.1016/j.jebo.2013.03.023
Litman, L., Robinson, J., & Abberbock, T. (2017). TurkPrime.com: A versatile
crowdsourcing data acquisition platform for the behavioral sciences. Behavior Research Methods, 49(2), 433–442. https://doi.org/10.3758/s13428-016-0727-z
Mazar, N., Amir, O., & Ariely, D. (2008). The Dishonesty of Honest People: A Theory of Self-Concept Maintenance. Journal of Marketing Research, 45(6), 633–644.
Science, 22(9), 1095–1100. https://doi.org/10.1177/0956797611417262
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. https://doi.org/10.1016/j.joep.2014.10.002
Ross, L., Greene, D., & House, P. (1977). The ‘false in social consensus perception effect’: An egocentric bias and attribution processes. Journal of Experimental Social
Psychology, 13(3), 279–301. https://doi.org/10.1016/0022-1031(77)90049-X
Roth, C., & Wohlfart, J. (2016). Experienced Inequality and Preferences for Redistribution. SSRN Electronic Journal, 1–87. https://doi.org/10.2139/ssrn.2809655
Rothstein, B. (2001). Social Capital in the Social Democratic Welfare State. Politics & Society. https://doi.org/10.1177/0032329201029002003
Transparency International (2017a). Corruption Perception Index 2017 [data file]. Retrieved from https://www.transparency.org/news/feature/corruption_perceptions_index_2017 Transparency International (2017b). Global Corruption Barometer 2017 [data file]. Retrieved
from https://www.transparency.org/research/gcb/gcb_2015_16/0 Transparency International (2018). What is Corruption? Retrieved
from https://www.transparency.org/what-is-corruption
United Nations University World Institute for Development Economics Research, UNU-WIDER (2017). World Income Inequality database WIID3.4 [data file]. Retrieved from https://www.wider.unu.edu/database/world-income-inequality-database-wiid34
Uslaner, E. M., & Badescu, G. (2004). Honesty, trust, and legal norms in the transition to democracy: Why Bo Rothstein is better able to explain Sweden than Romania. In Creating social trust in post-socialist transition (pp. 31–51). New York: Palgrave Macmillan. https://doi.org/10.15713/ins.mmj.3
World Inequality Database (2014). World Inequality Database WID [data file]. Retrieved from https://wid.world/data/
