Trust and Punishment: The effects of Punishment on Cooperation
in The Netherlands and India
Wasan, Raghav
Student number: 11376902
Date: August 12
th, 2017
Credits: 15
Supervisor: Dr. A. (Ailko) van der Veen
Faculty of Economics and Business
Specialization track: Behavioral Economics & Game Theory
University of Amsterdam
Statement of Originality
This document is written by Raghav Wasan, 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
Acknowledgement ... v
Abstract ... 1
Introduction: The Economic Man? ... 1
Literature Review: Interaction between punishment, trust and cooperation ... 3
The current study: Approach and predictions ... 8
Method ... 10
Participants ... 10
Design and Procedure ... 11
Results ... 15
Result 1 ... 16
Result 2 ... 19
Result 3 ... 21
Discussion and Conclusion ... 24
Results ... 24
Limitations and improvements ... 26
Appendix A: Repeating analysis with invalidated data: Comparison ... 29
Appendix B: Experiment (including instructions) ... 31
v Acknowledgement
Firstly, I would like to thank my supervisor and mentor, dr. A. (Ailko) van der Veen for his continued guidance for my Master Thesis. He was always supportive and encouraging me through his insights and ideas that helped me approach my thesis with confidence and rigor for which I am very grateful.
I would also like to thank Elisabeth Ennemoser, my colleague from Sociology Department, who’s support helped me shape my research idea.
Lastly, this thesis would not be possible without the critique and guidance of Ceren Abacioglu, my friend and colleague from Psychology Department, whose valuable insights helped me properly structure my thesis framework and analysis.
1 Abstract
Much of the existing literature has focused either on role of punishment on cooperation or role of trust on cooperation but not many studies have been conducted on the interaction between trust and punishment, and its effect on cooperation behavior. I conduct a public goods game in a high trust and a low trust society and explore this relationship to see where punishment is more effective. Unfortunately, I do not find significant evidence on whether punishment works better in enhancing cooperation in a high trust society or in a low trust society. However, some of the results found in previous literature are reaffirmed in this study – punishment increases overall contribution and also helps reduce free riding.
1. Introduction: The Economic man?
Homo economicus: The Economic man! Economics has been propagated throughout the years on the theory of self-interested rational human who is capable of weighing pros and cons of his actions and inactions, who takes fully informed decisions which result in best possible outcomes for himself. In traditional economic theory, the economic agent acts in his rational self-interest and puts his needs above everything else, a point that has been continuously illustrated by the Prisoner’s Dilemma game backed by Nash Prediction: both prisoners would always choose to defect. Behavioral economics strives to find a balance between homo economicus and homo sociologicus and assumes the agent to be irrational, and as such, a social animal – an agent capable of cooperating with other homo sapiens and working toward a common goal.
“Human societies represent a spectacular outlier with respect to all other animal species
because they are based on large-scale cooperation among genetically unrelated individuals” (Fehr & Fischbacher, 2004, as cited in Henrich et al., 2003). Fehr and Fischbacher (2004) also
point out that cooperation between humans is largely a function of social norms. These social norms are a function of a belief system that is shared amongst large groups of people which govern individual behavior as a part of that collective. Much of this cooperation is also enforced by the use of sanctions, whether legal or social – read, threat of credible punishment – which ensure that these constituted norms are rarely ever violated. Previous literature points out that punishment does indeed enhance cooperation, but its effect and viability fluctuates with differences in societies (Balliet & Lange, 2013). Difference in the levels of cooperation have
2 been also explained through differences in cultures, where different levels of willingness to cooperate have been accepted, and thus, different levels of punishment have been exhibited (Henrich et al., 2006). Related to this willingness to cooperate, one key cultural element that is present in all existing societies, as pointed out by Bond and colleagues (2004) is, therefore, the belief of other person’s goodwill – trust. Trust is commonly defined as, “the willingness to
accept vulnerability on the basis of the positive expectations of the intentions or behaviors of others”, (Rousseau, Sitkin, Burt, & Camerer, 1998, p. 395). Balliet and Lange (2013) point out
that this difference in the effect of punishment in enhancing cooperation is also a major function of the given trust in a society. Yet, no study so far has shown consistent evidence on whether punishment works better for cultures that have high (versus low) levels of trust in maintaining (and even possibly enhancing) cooperation. The current study, therefore, aims to answer the research question, “Does punishment work better in enhancing cooperation in a high trust
society or a low trust society?”
The rationale behind investigating this question is to study the relationship between trust and punishment and its effect on overall cooperation. But what we do not yet have evidence for is how do punishment and trust interact with each other in its effect on overall cooperation. Can people in a low trust society depend on some form of sanction to help them cooperate more, or would punishment work better in a high trust society in maintaining (or even possibly enhancing) cooperation? Yamagishi (1986) aimed to address this issue by inducing the idea that a sanctioning system could function as a public good. This system, when put in place, would ensure higher cooperation levels. He posited that it would work better in a low trust society as opposed to a high trust society in increasing overall cooperation in a society. I investigate his theory in relation to role and power of social norms and critique his stance on the relationship between trust and punishment. Based on this critique and investigation I develop my hypothesis and predictions.
In order to find evidence for my research question I play a one shot public goods game between a high trust and a low trust society based on the classification of World Value Survey Database1
. The rationale behind using this experimental game and not any other game is that
1 WVS is a body of social scientists and scholars from all over the world who study the impact of changing social values on a society’s culture and political outlook. It is headquartered in Vienna, Austria. The data from this body is used in a variety of fields like economics, sociology, anthropology etc. For more information, visit (http://www.worldvaluessurvey.org/WVSContents.jsp).
3 Yamagishi (1986) also uses a public goods game to study a similar proposition. Furthermore, another study interestingly describes public goods games as the mother of all cooperation models2
.
Therefore, a study investigating interaction between punishment and trust on cooperation should use a public goods game.The results of this experiment unfortunately are not conclusive for my research question. I find that punishment does increase overall contribution levels in both the societies, and also successfully deters free riding which is in line with previous literature but it does not provide significant evidence whether it increases contributions more in a high trust society or a low trust society.
The remainder of my study is organized as follows: Chapter two gives an overview of related literature, and critiques Yamagishi’s (1986) stance on interaction of punishment with trust, which is central to the core of my hypothesis. Chapter three outlines my hypothesis and predictions based on the literature review. Chapter four describes the methodology – participants used, their demographic characteristics, experimental design and payoff calculation method. Chapter five presents the results of my findings and the observed behavior of the participants. Finally, I discuss my results, their congruency with presented theory, limitations in my experimental design, possible improvements, and finally the conclusion.
2. Literature Review: Interaction between trust, punishment and cooperation
Much of the existing literature on cooperation has either focused on the effect of trust on cooperation (Ahmed & Salas, 2008; Buchan & Croson, 2004) or on the role of punishment in enhancing cooperation (Fehr & Gachter, 1999; Fehr, & Gachter, 2002; Fehr & Fischbacher, 2004). Fehr and Gachter in their 1999 paper studied the role of punishment in enhancing cooperation. Their results showed that average contribution levels in punishment treatment were higher than in no-punishment treatment illustrating the viability of punishment in enhancing cooperation. Ahmed & Salas (2008) studied levels of trust and cooperation in
2
I take this description from prof. dr. C.M. Matthijs Van Veelan’s paper, “Group selection, kin selection, altruism
and cooperation: When inclusive fitness is right and when it can be wrong”, He takes this description from an
email conversation with Michael Doebali, and gives him credit for it. I thought it is an apt description so I borrowed it here.
4 Sweden and India, and found a positive correlation between cooperation and trust within the counties. Following the steps from Fukuyama (1995), their study categorized the countries into high trust and low trust societies. According to Fukuyama (1995), the higher the level of trust the less corrupt, more economically developed and more socially prosperous the country will be. Sweden having high trust levels whereas India was classified as having low levels of trust. Using a one-shot public goods game their study, therefore, examined the impact of trust on cooperation and found that their results supported their expectation of positive correlation between trust and cooperation with average contribution levels being higher in Sweden than in India. In a similar manner, Buchan & Croson (2004) also tested out the role of trust on cooperation using the categorization as proposed by Fukuyama (1995) in USA and China. However, they failed to replicate the results from Ahmed & Salas (2008).
While these two studies examined the role of trust on cooperation, a cross-societal study carried out by Henrich et al. (2006) found evidence that the role (and willingness) of punishment in ensuring cooperation was indeed different in each society. Most of the studies thus far, however, studied cooperation using small groups of participants all belonging to ‘industrialized populations’, a limitation applicable to both Alcock & Mansell (1977) and Fox & Guyer (1977). Overcoming this limitation, Henrich et al. (2006) looked at 15 different societies (including tribal societies) and found differences in game behavior between these societies and the industrialized participant pools. They used a Third Party Punishment Game (3PPG) and found significant difference between the mounts these societies were willing to accept and willing to punish given their own society’s trust levels which created ‘cultural’ equilibria for cooperation, showcasing that the interaction between trust, punishment and cooperation was different for different societies. As pointed out earlier, while the cited literature has stressed on the effects of trust and punishment on cooperation, none of these studies really did look at how punishment might interact with the level of trust in its effect on cooperation.
Yamagishi (1986, 1988) posited that punishment would work better in a low trust society. According to him, people are willing to cooperate with each other once they are assured of others’ cooperation. People, in general, are aware that cooperation will lead to a mutually beneficial outcome and that not cooperating with each other would lead to undesired outcomes; even though people realize this they still fail to cooperate with each other. Citing reason for this failure to contribute toward a common goal, Pruitt and Kimmel (1977) theorized that
5 people generally do not trust3
each other to act towards that common objective. It is the positive expectation of having one’s vulnerability reciprocated, that dictates if one chooses to cooperate or not. If people do not believe that their goodwill/ cooperation will be positively reciprocated, then one will not be willing to cooperate. This is to say that mutual cooperation would exist only when people are convinced that their goodwill will not be exploited. In contrast, if in a society people have high trust levels, that is, when they know that they will seldom be exploited for their trusting behavior, cooperation becomes mutual. But how would people cooperate if they had low trust levels?
To address this problem, Yamagishi (1986) came up with a modified version of the existing goal/expectations theory proposed by Pruitt and Kimmel (1977). The goal/expectations theory states that a common goal for cooperation would have to be backed with positive expectations of others’ cooperation for people to actually cooperate, that is, there would have to be mutual trust. One drawback, however, of this theory and of existing literature on the effects of trust on cooperation in a social dilemma situation is that this, as pointed out earlier, has only been demonstrated using small groups of participants (Alcock & Mansell, 1977; Fox & Guyer, 1977). Fox & Guyer (1977) found an inverse relationship between group size and level of cooperation suggesting that smaller the group the larger the cooperation levels. Alcock & Mansell (1977) also found an interesting result in that people’s non-cooperation was a dominant strategy not because people only cared about own selfish interest but because they were also more loss averse – attaining a net loss by cooperating while others chose not to cooperate. As such, if the group was small enough for an individual’s actions to be visible to other group members then it could create interdependency between members where the impact of their decisions would also be visible which would make it easier for mutual cooperation to exist. However, it would be difficult to replicate similar results in larger groups where an individual’s decision would hardly be visible to others and also have minimal impact on their outcome, making it difficult for people to trust each other to cooperate and work toward a common goal.
Yamagishi’s (1986) solution to this problem was to have a sanctioning system as a public good. Typically, participants would contribute their experiment money to this system just as they would to a public good. The sum of this contribution from all members would then be allocated
3
6 to the smallest contributor to the public good whose payoff then would be reduced as a punishment. As such, the function of this sanctioning system as a public good would be to punish those who chose to contribute least as compared to other people’s contribution to public goods. Therefore, this system would ensure that nobody would want to be the lowest contributor which would push up the contribution levels. So, while having a common goal in place, this would also back people’s expectations that their trust would not be exploited by non-cooperators. As further argued by Yamagishi (1986), the premise of inducing a sanctioning system as a public good, where people voluntarily contribute to the success of the sanctioning system as a means to ensure cooperation, requires people to realize the necessity of having a sanctioning system in place. If people continue to cooperate then we would not require a sanctioning system in the first place. The necessity to have a system in place is to realize the lack thereof of reasonable cooperation levels in that given society. For example, if members of the society continue to pay dutifully their taxes then one need not put a punitive system in place. It is only when they begin to evade taxes that a punitive system must be put in place to check the evaders. This system is disincentive enough for members to cooperate and does not only benefit them directly through their own cooperation but also benefits them indirectly through other member’s cooperation, thereby enhancing cooperation levels as a whole. Therefore, by assuring people that their trust would not be violated through a sanctioning system, higher levels of cooperation could be achieved and this approach would be more effective in enhancing cooperation in a low trust society than in a high trust society.
In order to test this exact rationale, Yamagishi (1986) conducted an experiment where he used a modified version of a standard public goods game to test whether high trusting or low trusting people contributed more to the sanctioning system. This design had three fundamental differences from the public goods game developed by Fehr and Gachter4 (1999). First, participants contributed to a sanctioning system as a separate public good rather than directly punishing their fellow participants from their contribution. Second, and the major drawback of this design was that the only members that got punished through this sanctioning system were the one with the lowest contribution in that group, and as such this design did not allow for
7 anti-social punishment5. Third, the payoff function in the sanction treatment was designed such that one could still earn more than the given endowment even after being punished for their (non) contribution. This removed any incentive for a free rider to contribute to the public good or even to the sanctioning system. As such, even though low trusting members contributed significantly more to the sanctioning system than high trusting members and supported for Yamagishi’s hypothesis, this result was still inconclusive as to whether a sanctioning system would work better in a high trust or a low society due to this design bias.
Punishment as a form of public good or as a means to ensure higher cooperation can only work if all the people are willing to contribute in the punishment of the non-cooperators (Yamagishi, 1986). If the punishment efforts of one individual are reciprocated by other group members then punishment as a means to ensure higher cooperation would function well. This is to say that if the total benefit of ensuring higher cooperation outweigh the total cost of punishment itself then one would be willing to bear this cost (Yamagishi, 1988). In contrast, if a group member does not believe that their punishment efforts will be reciprocated by other group members, or if these other group members themselves continue to free ride on the benefits provided by those who do punish then the total cost of punishment for the punisher outweighs the total benefit gained by him. This would discourage any form of punishment of the non-cooperators at the hands of the non-cooperators and the sanctioning system as a provision/public good would fail to ensure higher cooperation in a low trust society.
Whether punishment works better in a high trust society or low society may also be looked at through the lens of social capital and the formation of norms within a society. Here, I define social capital as – “the idea that the benevolence6 of others in one’s social network is a valuable resource that provides benefits (and sometimes cost) to individuals in terms of both economic and social exchange”, (Balliet, & Lange, 2013, as cited in Adler & Kwon, 2002). Thus, people
in a society can derive positive economic and social value from each other’s benevolence. Coleman (1988) posits that social norms can be extremely powerful form of social capital. To
5
When a group member punishes their fellow member for contributing at least as much as them. In other words, when a group member punishes their fellow member for behaving more pro socially than themselves it is called anti-social punishment (Hermann, Thoni & Gachter, 2008).
6
The word ‘benevolent’ is derived from Latin with the prefix meaning ‘good’ (bene) and the root meaning ‘wish’ (velle). It is also translated as ‘goodwill’
8 quote an example from his study, “Effective norms that inhibit crime make it possible to walk
freely outside at night in a city and enable old persons to leave their houses without fear for their safety”. Social capital in a society is built up of plethora of such norms guiding the social
and economic relationships that the members of a society have with each other which are further strengthened or weakened by expectations, obligations and trustworthiness that they exhibit toward each other. Once these norms are internalized they form the core of social capital and then regulate its members’ behavior toward each other; these norms then may be put in effect through internal sanctions, like social sanctions in a collective society, or external sanctions, like legal sanctions ensuring that the norms are followed (Coleman, 1988).
As such, when in a high trust society cooperation becomes an internalized norm, through expectations and each other’s trustworthiness, people would be less willing to violate this norm. If this norm were to be violated, then the cooperators would be willing to bear the cost of punishment as they would trust other cooperators to contribute to the sanctioning of the non– cooperators ensuring previous levels of cooperation. Gintis (2008) conducted a study to understand the effect of punishment on cooperation on 15 different societies found that most of the low contributors responded to symbolic penalties as much as monetary penalties which induced a sense of guilt in them for having violated the cooperation norm. This study interestingly also found evidence for anti-social punishment in the later rounds, where the non-contributors punished the high-non-contributors believing that they punished them in the previous rounds for their low - or no - contributions. This behavior led to a decrease in the overall cooperation levels in the subsequent rounds making the role of punishment less effective in low trust societies while making it more effective in high trust societies through norm enforcement.
3. The current study: Theoretical approach and predictions
I conduct a one shot Public Goods game experiment with participants from India and The Netherlands. According to ‘World Value Survey data report (2010 - 2014)’ on trust, which asks the question, “Generally speaking, would you say that most people can be trusted or that you
need to be very careful in dealing with people?”, India is classified as a low trust society
whereas The Netherlands is classified as a high trust society. In their survey only 16.7% of the Indians said that they could trust other people making it a low trust society while 66.1% of the
9 Dutch said that they could trust other people making it a high trust society, thus, corroborating the choice of the countries. The experiment consisted of two treatments, namely ‘no punishment’ and ‘punishment’ treatment. Subjects were asked how much they would be willing to contribute to the public given how much their group members would contribute.
Looking at Yamagishi’s proposition of low trust societies reacting better to punishment as well as the role that social capital and norms play in a society the primary hypothesis of my study is:
H0: Punishment enhances cooperation more in The Netherlands than in India H1: Punishment contributions will be higher in The Netherlands than in India
I use the Public Goods game model as used by Fehr and Gachter (1999). The participants can choose to contribute or not contribute a part (or whole) of their initial endowments to the public goods, which is the sum of contributions made by all group members. This total sum is then multiplied by a number greater than 1 but smaller than N (the number of participants in a single group) and then equally divided amongst all group members. This number is called Marginal Per Capita Return (MPCR). In line with Nash theory, if all agents act rationally they would contribute zero in each stage and maximize their individual earnings. However, I expect members to contribute on average more than zero in both the treatments, as has also been found in other numerous studies (Ahmed & Salas, 2008; Buchan & Croson, 2004, etc.). Given the level of trust within the countries, there will be a greater number of free riders in Indian participants than in Dutch participants. If the participants apply the logic of backward induction, then neither set of participants should contribute more than zero, however, since India is a low trust country fewer Indian participants would trust their group member to contribute to the public good, and therefore, choose to also not contribute and in turn free ride. In contrast, since The Netherlands is a high trust society, its participants would expect their group members to free ride less on their contributions. Free rider here is defined as a participant who contributes zero for all 21 (0 to 20) entries in the contribution schedule7
(Fischbacher, Gachter & Fehr, 2001).
7 A contribution schedule is a table which lists down all possible contribution values of other group members
10 Also, as hypothesized, since in a high trust society (The Netherlands) cooperation is an internalized social norm the mean amount spent on punishing their group members for their perceived low contribution will be higher in The Netherlands than in India. While I expect to see this, I also expect to see higher contributions by both sets of participants when faced with the threat of credible punishment as compared to when they are not (just as it has been seen in previous cooperation and punishment experiments), thus, preserving the sanctity of the role of punishment in enhancing cooperation (Fehr & Gachter, 1999; Fehr & Gachter, 2002; Gächter, Renner & Sefton, 2008; etc.).
I also predict that the Dutch participants would contribute more in the presence of threat of credible punishment than their Indian counterparts rendering my hypothesis true that punishment does indeed enhance cooperation more in high trust societies than in low trust societies.
4. Method
Participants
As the current study investigated the effect of trust levels on cooperation within two selected countries, I gathered responses from Dutch and Indian participants that reside in their countries of origin, and not abroad (e.g., Indians living in The Netherlands), thus, necessitating conduction of the experiment online than in a laboratory. Table 1 below, presents the demographic data of the participants. The survey was distributed through Facebook, WhatsApp as well as email to maximize its outreach. There were a total of 60 participants in the online experiment; 27 (45%) Dutch and 33 (55%) Indians. However, six observations were dropped due to invalid data entries on part of the participants making it a total of 54. Some of these discarded participants entered punishment amounts of more than 10 Euros in the first stage of ‘punishment’ treatment which was not allowed and some of them chose to punish upper limit of the contribution schedule with maximum punishment amounts. This clearly showed that they either did not understand the game or filled their data randomly. Out of these six, two were Indians and four were Dutch making it a subtotal of 23 Dutch and 31 Indians. Out of the remaining total participants, 21 (39%) were female and 33 (61%) were men. The percentage of females amongst the Dutch as well as Indian participants was 39%. 91.3% of the Dutch
11 participants were aged between 18-27 and 93.5% of the Indian participants fell into the same age bracket, and all participants had at least Bachelors level education, except for eight Dutch participants. The subject pool for the experiment was, therefore, largely symmetrical in its demographic characteristics.
Table 1 - Summary of Participants
Nationality
Population Characteristics Indian Dutch
Sex Men Women 19 (61.3%) 12 (38.7%) 14 (61%) 9 (39%) Trust8 Yes No Not sure 5 (16.1%) 11 (35.5%) 15 (48.4%) 15 (65.2%) 2 (8%) 6 (26%) Age (in years)
18-22 23-27 28-32 4 (12.9%) 25 (80.6%) 1 (3%) 6 (26.1%) 15 (65.2%) 2 (8%) Education High School Bachelors Masters 0 13 (41.9%) 18 (58.1%) 8 (34.8%) 14 (60.9%) 1 (4.3%)
Notes: Numbers in parenthesis indicate the percentage of the demographic characteristics. The table shows the
demographic data for 54 participants broken down over sex, trust, age and education.
Design and Procedure
I use Strategy Method for my Public Goods online experiment. In most public goods game experiments, experimenters use Direct Response Method where subjects take their decision when it is their turn to do so. Fehr and Gachter (1999), Andreoni (1988, 1995) use this method in their studies. Typically, in the Direct Response Method the participants are asked how much they would contribute to the public good without knowing how much their group members
8 The participants were asked the same question that was used in the World Value Survey, “Generally speaking,
would you say that most people can be trusted or that you need to be very careful in dealing with people?”. ‘Yes’
corresponds to participants believing that people can be trusted; ‘No’ corresponds to participants believing that they cannot be too careful in dealing with people; ‘Not sure’ corresponds to participants not being sure whether people can be trusted or not.
12 have contributed. This method allows the experimenters to observe direct responses of a participant to the responses of their fellow participants. While this method is good to test behavioral tendencies of the participants it does not produce data points for every possible node in the experiment which the strategy method allows us to do (Brandts & Charness, 2011; Fischbacher, Gachter & Quercia, 2012). In Strategy Method, the subjects are asked how much they would like to contribute to the public good depending upon how much their group members contribute. Typically, this involves giving a participant a ‘contribution schedule’ which enlists all the possible contribution levels of their fellow participants in their group, thereby giving us more data points than would be obtained through Direct Response Method.
Since I use Strategy Method for my experiment, there is no learning behavior on the part of the participants so we will not see the same trajectory of contributions as seen in some of the existing literature which uses Direct Response Method (Eg. Fehr, & Gachter, 1999; Andreoni, 1988, 1995; etc.). Furthermore, playing a one shot Public Goods game also captures a twofold effect: the risk of one trusting his group members, and interdependence of participants in attaining their payoffs (Ahmed & Salas 2008). While this game effectively captures the risk of trust in the ‘no-punishment’ treatment, it also captures the relationship between punishment and trust, and its effect cooperation in the ‘punishment’ treatment (Yamagishi, 1986). The key comparison in my approach lies in, first, comparing data from stage one of ‘no punishment’ treatment, and stage two of ‘punishment’ treatment’– whether punishment actually increases contribution levels in India and in The Netherlands, and second, comparing data from stage two between the two countries to see if punishment increases contribution more in India or in The Netherlands.
The experiment was administered online for Dutch and Indian participants using an online survey tool Qualtrics for two weeks to which I had access through University of Amsterdam. Participants were randomly matched with other participants from their own country in order to induce trust for their own country members and to study its effect on cooperation. Once the data was collected, the participants were provided with their total earnings information via email and informed if they had won any money or not. The winners were separately asked to provide their bank details in order to be transferred their winning amount. Five participants were randomly chosen for payment from the top 20 earners which consisted of eleven Indians and nine Dutch. They were paid a total of 30 Euros from 4 to 8 Euros.
13 At the beginning of the experiment the participants were asked, along with a few control questions, to indicate whether they could trust other people or not. For this, I used the official question as is asked in World Value Survey (WVS; http://www.worldvaluessurvey.org/) in order to correlate this data with their contribution levels. Furthermore, in order to provide a frame of reference of the amount of money they were initially endowed with the participants were also given the stable exchange rate between EUR and INR (1 EUR = 75 INR).
The participant groups (one group consists of N=2 participants) were asked to indicate their level of contribution given the contribution level of their group member. The payoff formula used for this treatment is the same as originally used by Fehr and Gachter (1999).
It is given as follows: 𝜋"#= 𝜔"− 𝑔" + 𝑎 𝑔"+ (1) where, 𝜋"#= 𝑇𝑜𝑡𝑎𝑙 𝑒𝑎𝑟𝑛𝑖𝑛𝑔𝑠 𝑜𝑓 𝑡ℎ𝑒 𝑝𝑙𝑎𝑦𝑒𝑟 𝑖 𝜔" = 𝐸𝑛𝑑𝑜𝑤𝑛𝑚𝑒𝑛𝑡 𝑔" = 𝑃𝑙𝑎𝑦𝑒𝑟 𝑖’𝑠 𝑐𝑜𝑛𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑎 = 𝑀𝑎𝑟𝑔𝑖𝑛𝑎𝑙 𝑃𝑒𝑟 𝐶𝑎𝑝𝑖𝑡𝑎 𝑅𝑒𝑡𝑢!𝑛 𝑤𝑖𝑡ℎ 𝑎 = 0.6 𝑔"+ = 𝑇𝑜𝑡𝑎𝑙 𝑐𝑜𝑛𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑜𝑓 𝑝𝑙𝑎𝑦𝑒𝑟 𝑖 𝑎𝑛𝑑 𝑗
The participants were informed in the instructions that they would endowed with an initial earnings of 20 Euros to play the game with and that they would be randomly matched with a participant from their own country. Once this was done, they were then given a ‘contribution schedule’ (from 0 to 20 Euros) of their group member and asked to indicate their level of contribution for the given level of contribution of their group member. According to Nash theory, if each participant acts rationally then they would apply logic of backward induction and opt for the unique sub perfect Nash equilibrium and contribute zero (𝑔" = 0) and attain
individual maximized earnings of 20 Euros. However, the social optimum would dictate that each participant contribute their entire endowment to the public good (𝑔" = 20) and attain a socially maximized total earnings of 24 Euros.
14 They were further informed in the instructions that once they had filled out the ‘contribution schedule’, either of the participant’s ‘contribution schedule’ would be randomly chosen to calculate the ‘total earnings’, 𝜋"# for both the participants. Since 𝜋"# was calculated for each
given value (from 0 to 20) in the ‘contribution schedule’, it was averaged to compute total average earnings for both participants.
The second treatment in the experiment was the ‘punishment treatment’ which consisted of two stages. The payoff formula for this treatment is also the same as originally used by Fehr and Gachter (1999).
It is given as follows:
𝜋"I= 𝜋"#∗ 1 −L#NM − 𝑃" (2)
where, 𝑃+ = 𝑝𝑢𝑛𝑖𝑠ℎ𝑚𝑒𝑛𝑡 𝑎𝑚𝑜𝑢𝑛𝑡 "𝑝𝑒𝑛𝑡 𝑏𝑦 𝑔𝑟𝑜𝑢𝑝 𝑚𝑒𝑚𝑏𝑒𝑟 𝑗
and 𝑃" = 𝑝𝑢𝑛𝑖𝑠ℎ𝑚𝑒𝑛𝑡 𝑎𝑚𝑜𝑢𝑛𝑡 𝑠𝑝𝑒𝑛𝑡 𝑏𝑦 𝑔𝑟𝑜𝑢𝑝 𝑚𝑒𝑚𝑏𝑒𝑟 𝑖
In this treatment, the participants were given the opportunity to punish their group member for the given ‘contribution schedule’. The maximum punishment points player 𝑗 could assign to player 𝑖 was 0 ≤ 𝑃+ ≤ 10 Euros and vice versa. The participants were informed that this 10
Euros was not a part of the initial endowment of 20 Euros that they were given to play the game with. This was only relevant for the first stage and was designed to make sure that 𝜋"# ≥ 0
always, thereby not dis-incentivizing voluntary contribution on part of player 𝑖 even when 𝜋"I could go below zero. The participants were again endowed with an initial earnings of 20 Euros to play the game. Just as in ‘no punishment’ treatment they were then given a ‘contribution schedule’ (from 0 to 20 Euros) of their group member.
The crux of this treatment was the first stage. The first stage asked the participants how much they would be willing to spend from 0 to 10 Euros to punish their group member for all contribution levels from 0 to 20. The primary purpose of this first stage was to prime them with the possibility of being punished for their contribution and therefore, the opportunity to punish their group members for their contribution. The contributions from this stage were also used to calculate the payoffs for this treatment. Once the participants completed this round they moved
15 on to the second stage. Stage two of this treatment was exactly the same as stage one from ‘no punishment’ treatment; the key difference here was that the participants were informed that they could be punished for their contribution. They were then asked them how much they would be willing to contribute for the given level of contribution of their group member given that they themselves too could be punished for their level of contribution.
The unique sub perfect Nash equilibrium for this treatment is the same as ‘no punishment’ treatment where group members contribute zero to public goods in the second stage, and do not assign any punishment points to their group member in the first stage and thereby, attain a total earnings of 20 Euros.
𝜋"# of the participants in this treatment was calculated in the same manner as in ‘no-punishment’ treatment. The pairing of the participants was once again randomized, so that they could be partnered with different members as in from ‘no punishment’ treatment. Once this was done the ‘contribution schedule’ (let us call this CS2) of either of the participants from the second stage was randomly chosen and 𝜋"# for both was calculated using CS2. Since both
participants had also indicated their punishment levels in the first stage for the given ‘contribution schedule’ of their group member, the contribution levels indicated in CS2 were then matched with these punishment levels 𝑃" and 𝑃+ to calculate total earnings 𝜋"I. Just as in the ‘punishment’ treatment, since 𝜋"I was calculated for each given value in the ‘contribution schedule’, it was averaged to give total average of 𝜋"I for both of them. This total averaged
𝜋"# from the ‘no punishment treatment’ and 𝜋"I were further averaged to finally give the
average earnings from both the treatments for each participant.
5. Results
As expected, the variation was clearly found in the trust levels9 of both the societies. Amongst the Indian participants only 5 people (16.1%) said that they could trust other people, 15 (48.4%) were not sure and 11 (35.5%) of them said that they “can’t be too careful in dealing with
people”. In contrast, 15 (65.2%) of the Dutch participants answered that they could trust other
people, only 2 (8.7%) said that they “can’t be too careful in dealing with people” while the rest
9
16 of the participants couldn’t say. These numbers correspond and confirm the official trust levels as recorded in the latest report of World Value Survey (WVS; http://www.worldvaluessurvey.org/) as noted in one of the previous sections.
Result 1: The opportunity to punish group members caused a significant increase in the mean contributions in ‘punishment’ treatment for both sets of participants. Indian mean contributions of their initial endowment go up by 10% (from 29% in ‘no punishment’ treatment to 39% in ‘punishment’ treatment). Dutch mean contributions of their initial endowment go up by 8% (from 30% to 38%).
Figure 1A. Mean Contributions of Indian participants
Notes: ‘No punishment’ Contribution line shows the participants’ mean contribution in the first treatment. ‘Punishment Contribution’ line shows participants’ mean contribution in second treatment. ‘Perfect contribution’ line shows the own mean contribution had members contributed the same amounts as in the contribution schedule. Paired t-test indicates a significant increase in their mean contributions with t= -76.135 and p<0.001.
0 2 4 6 8 10 12 14 16 18 20 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 O w n M ean C on tr ib u ti on Contribution Schedule No punishment Contribution Punishment Contribution Perfect Contribution
17 Figure 1B. Mean Contributions of Dutch participants
Notes: ‘No punishment’ Contribution line shows the participants’ mean contribution in the first treatment. ‘Punishment Contribution’ line shows participants’ mean contribution in second treatment. ‘Perfect contribution’ line shows the own mean contribution had members contributed the same amounts as in the contribution schedule. Paired t-test indicates a significant increase in mean contributions of the Dutch with t=-8.370 and p<0.001.
Mean contribution data for ‘no punishment’ and ‘punishment’ treatment for Indian and Dutch participants is presented in Figure 1A and Figure 1B above, and Table 2 below. Both, the table as well as the Figures show an increase in mean contributions for both sets of participants from ‘no punishment’ to ‘punishment’ treatment. I conducted a Shapiro-Wilk Normality Test to see if the mean contributions by Indian and Dutch participants for both treatments were normally distributed. For Indian participants I obtained W Statistic of 0.939 with p=0.214 for ‘no punishment’ treatment, and W Statistic of 0.942 with p=0.244 for ‘punishment’ treatment, respectively. Similarly, for Dutch participants I obtained W statistic of 0.928 with p=0.125, and W statistic of 0.949 with p=0.336, respectively, suggesting that the null hypothesis could not be rejected and that the contribution data was normally distributed. This result validates the use of a paired t-test, which can only be used if the data if normally distributed.
Conducting a paired t-test showed that presenting the participants with an opportunity to punish their group members for their given contribution schedule led to a significant increase in their mean contributions with t= -76.135 and p<0.001 for Indian participants and t=-8.370 and
0 2 4 6 8 10 12 14 16 18 20 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 O w n M ean c on tr ib u ti on Contribution Schedule No Punishment Contribution Punishment Contribution Perfect Contribution
18
p<0.001 for Dutch participants, respectively. Thus, replicating the effect and role of
punishment in enhancing cooperation as also found in the previous literature.
Table 2 - Mean Contribution for both treatments Mean contribution by Indian
Participants Mean contribution by Dutch Participants
Contribution Schedule No punishment opportunity Punishment opportunity No punishment opportunity Punishment opportunity 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 2.68 (4.79) 2.94 (4.49) 3.19 (4.32) 3.58 (4.21) 3.94 (4.04) 4.10 (4.06) 4.32 (4.32) 4.97 (4.72) 5.29 (4.93) 5.65 (5.01) 5.74 (5.32) 6.16 (5.59) 6.68 (5.75) 6.94 (6.08) 7.32 (6.63) 7.87 (7.08) 8.10 (7.41) 8.45 (7.92) 8.84 (8.37) 9.00 (8.67) 9.06 (8.78) 4.61 (6.29) 4.81 (6.05) 5.10 (5.86) 5.39 (5.72) 5.74 (5.60) 6.00 (5.37) 6.35 (5.50) 6.65 (5.51) 6.97 (5.54) 7.39 (5.57) 7.71 (5.73) 8.03 (5.89) 8.58 (6.27) 8.97 (6.47) 9.29 (6.66) 9.68 (7.05) 9.90 (7.25) 10.26 (7.55) 10.81 (8.12) 10.97 (8.36) 11.16 (8.60) 1.74 (3.15) 2.39 (3.20) 2.70 (3.20) 3.17 (3.58) 3.61 (3.61) 4.26 (3.78) 5.09 (3.59) 5.39 (3.64) 5.22 (5.87) 5.57 (4.07) 6.39 (5.19) 6.52 (5.32) 7.17 (5.70) 7.30 (5.76) 7.78 (6.06) 7.96 (6.20) 8.13 (6.36) 8.52 (6.79) 8.78 (6.94) 8.78 (7.01) 8.87 (7.18) 2.52 (3.46) 3.22 (3.23) 3.65 (3.01) 4.04 (2.90) 4.43 (2.79) 5.09 (3.03) 5.91 (3.27) 6.30 (3.32) 6.39 (3.74) 6.78 (3.94) 7.74 (4.83) 8.22 (4.80) 8.83 (4.96) 9.13 (5.10) 9.83 (5.36) 10.35 (5.60) 10.74 (5.86) 11.35 (6.21) 11.87 (6.43) 12.13 (6.61) 12.43 (6.80)
19 Mean 5.94 (5.83) 7.83 (6.43) 5.97 (4.96) 7.66 (4.54)
Notes: Numbers in parentheses are standard deviations. This table shows the difference in the mean contributions made by both set of participants in ‘no punishment’ and ‘punishment’ treatments. Paired t-test shows that the mean contributions between Indian and Dutch participants in ‘no punishment’ treatment are statistically similar with t=-0.269 and p=0.790. Paired t-test also indicates no significant difference between their mean contributions in ‘punishment’ treatment with t=0.730 and p=0.473.
Result 2: There was no significant difference in the contributions between Dutch and Indian participants in the ‘no punishment’ treatment. However, I also did not find any significant increase in the contributions by Dutch participants over Indian participants in the ‘punishment’ treatment.
Figure 2A. Mean Contributions in ‘no punishment’ treatment
Notes: This figure shows the difference between mean contributions made by Indian and Dutch participants in the second stage of ‘punishment’ treatment. Paired t-test shows no significant difference between their contribution behavior with t=-0.269 and p=0.790. 0 2 4 6 8 10 12 14 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 M ean C on tr ib u ti on s Contribution Schedule Indian Dutch
20 Figure 2B. Mean Contributions in 'punishment' treatment
Notes: This figure shows the difference between mean contributions made by Indian and Dutch participants in the second stage of ‘punishment’ treatment. Paired t-test shows no significant difference between their contribution behavior with t=0.730 and p=0.473.
Column 3 and 5 in Table 2, and Figure 2A and 2B given above summarize the effect of punishment on enhancing contribution levels for Indian and Dutch participants. Paired t-test between mean contributions of Dutch and Indian participants in ‘no punishment’ treatment showed no significant difference between their contribution behavior with t=-0.269 and
p=0.790 indicating similar contribution behavior between the set of participants. Next, I
compared their contribution behavior in ‘punishment’ treatment to see whether or not punishment significantly increased contributions more for Dutch than for Indians. The result thus obtained, using a paired t-test did not show any significant difference between their contribution behavior with t=0.730 and p=0.473. Hence, I did not find any statistical support for my first hypothesis that punishment would increase contributions significantly more amongst Dutch participants than amongst Indian participants.
In order to ascertain the reason behind this result I conduct another set of analysis. Conducting a paired t-test between Indian and Dutch participants in ‘no punishment’ treatment I interestingly find that trusting Indian participants contributed significantly more than their Dutch counterparts with t=-20.634 and p<0.001. On the contrary, paired t-test reveals that non-trusting Dutch participants contributed significantly lot more than their non-non-trusting Indian counterparts with t=3.282 and p=0.037. Added to this, a paired t-test between Indian and Dutch participants who are not sure if they can trust others or not in ‘no punishment’ treatment shows
0 2 4 6 8 10 12 14 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 M ean C on tr ib u ti on s Contribution Schedule Indians Dutch
21 that Dutch participants contribute significantly more than their Indian counterparts with
t=3.277 and p=0.003. I also find insightful results by conducting a similar analysis for both
sets of participants for ‘punishment’ treatment. Trusting Dutch participants contributed significantly more than their trusting Indian counterparts with t=4.074 and p<0.001, unlike in the ‘no punishment’ treatment. But we also find a difference between non-trusting Indian and Dutch participants with non-trusting Indians contributing significantly more with t=-9.257 and
p<0.001. Similar analysis on ‘not sure’ participants for ‘punishment’ treatment shows that
Indian participants contribute significantly more than Dutch participants with t=-4.377 and
p<0.001.
Hence, it is this counteracting and inconsistent contribution behavior on part of trusting and non-trusting sets of participants in both treatments that give me insignificant overall results due to which I have no support for my first hypothesis.
Result 3: There were more free riders amongst Indian participants than in Dutch participants in both treatments. Indians also spent significantly more on punishment than their Dutch counterparts. On average, Indians spent 22% of the endowed 10 Euros to punish their group members while Dutch participants spent 18% on punishment. However, this did not significantly reduce free riding amongst Indian participants like it did amongst their Dutch counterparts.
As predicted, there were more free riders amongst Indian participants than amongst Dutch in both treatments. A total of eleven participants chose to free ride in the absence of punishment. Out of these eleven, five (45%) were Dutch and six (55%) were Indians, respectively. Conducting a paired t-test between free riders for both sets of participants shows that when they were presented with the threat of credible punishment, overall free riding went down significantly by 55% from eleven free riders to five free riders with t=-3.026 and p=0.012, thereby, ascertaining the effectiveness of punishment in curbing the free rider problem.
22 Figure 3. Mean Amount spent on Punishment
Notes: This graph shows the difference between the mean punishment contributions made by both sets of participants in the first stage of ‘punishment’ treatment. Wilcoxon signed rank sum test, which is a non-parametric version of the paired t-test, shows a significant difference between the mean punishment amounts contributed between both groups with p=0.0024 but with Indians punishing significantly more than their Dutch counterparts.
I then look at mean amounts spent on punishment by both sets of participants to study their punishment behavior. Figure 3 above and Table 3 below, present participants’ mean punishment amounts spent for the given contribution schedule. Conducting a Shapiro-Wilk Normality test shows that the null hypothesis could not be rejected for Indian participants with mean punishment contributions being normally distributed with W statistic of 0.922 and
p=0.098. It was, however, rejected for their Dutch counterparts with W statistic of 0.845 and p=0.003 suggesting that their mean punishment contribution data was not normally distributed.
This result invalidates the use of a paired t-test which can only be used if the data is normally distributed.
A Wilcoxon signed rank sum test, which is a non-parametric version of the paired t-test, shows that there was a significant difference between the mean punishment amounts contributed by Indian and Dutch participants with p=0.002 with but with Indians punishing significantly more than their Dutch counterparts. Thus, as can be seen our second hypothesis that high trust societies would have higher punishment contributions is rejected.
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6 6.5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 O w n M ean P u n is h m en t Contribution Schedule Indians Dutch
23 Given this result, I find that Dutch free riders reacted more strongly to this threat of credible punishment than did Indian free riders. Out of the five Dutch free riders in “no punishment’ treatment only one continued to free ride in the ‘punishment’ treatment, significantly decreasing free riding amongst Dutch participants by 80% with t=2.152 and p=0.042. But free riding did not significantly go down amongst Indian participants with t=1.438 and p=0.160. Four out of the six original free riders still continued to free ride.
What I also interestingly find is that amongst Dutch participants only 13% (3 out of 21) chose not to punish at all (𝑃" = 0) at the first stage of ‘punishment’ treatment. This was
proportionately higher amongst Indian participants with 38% (12 out of 31) choosing not to punish at all (𝑃" = 0). Furthermore, Dutch participants spent a mean amount of 1.84 Euros
with their median mean punishment contribution being 1.70 Euros. Indian mean punishment contribution was skewed upwards at 2.21 Euros with their median mean punishment contribution being 1.90 Euros. In order to see if there was a difference in the number of members taking responsibility of punishment I conducted a t-test between both sets of participants at the first stage of ‘punishment’ treatment. We find that there is a significant difference between the number of participants contributing zero with t=2.426 and p=0.018. Hence, even though Indian participants punished significantly more than their Dutch counterparts the responsibility of punishment was shared more equally amongst Dutch participants.
Table 3 - Mean Punishment contribution Contribution
Schedule
Mean Punishment Contribution by Indians
Mean Punishment Contribution by Dutch 20 19 18 17 16 15 14 13 12 0.03 (0.18) 0.10 (0.40) 0.19 (0.65) 0.29 (0.90) 0.45 (1.21) 0.55 (1.43) 0.94 (1.93) 1.10 (2.18) 1.42 (2.81) 0.09 (0.42) 0.09 (0.42) 0.09 (0.42) 0.13 (0.46) 0.13 (0.46) 0.13 (0.46) 0.48 (1.16) 0.52 (1.24) 0.57 (1.34)
24 11 10 9 8 7 6 5 4 3 2 1 0 Mean 1.52 (3.00) 1.90 (3.26) 2.39 (3.48) 2.81 (3.53) 3.00 (3.53) 3.29 (3.67) 3.48 (3.88) 3.90 (3.89) 4.06 (4.00) 4.58 (4.51) 5.13 (4.66) 5.23 (4.81) 2.21 (2.76) 0.70 (1.72) 0.96 (1.92) 1.43 (2.09) 1.96 (2.69) 2.22 (2.68) 2.57 (2.79) 3.30 (3.20) 3.70 3.35) 4.17 (3.58) 4.78 (3.67) 5.00 3.81) 5.78 (3.83) 1.85 (1.98)
Note: Numbers in parentheses are standard deviations. This table shows the mean punishment amounts contributed by both sets of participants for the given contribution schedule. Paired t-test shows no significant difference between the contributions made by Indian and Dutch participants with t=0.730 and p=0.473.
6. Discussion and Conclusion
Results
In Yamagishi’s (1986) view punishment, as a form of public good, would assure people of other people’s cooperation. As such, punishment would work better in enhancing cooperation in a society where a mechanism would be required to uphold people’s trust. I induced the argument that since in a low trust society it is difficult for people to trust each other, none of its members would be willing to contribute to punishment as a public good, and punishment as a system would fail to ensure higher cooperation in a low trust society. The results of this study, however, are mixed.
I unfortunately do not find statistical evidence to support my first hypothesis. This result is presented in Result 2. As expected, I find that in the absence of punishment both sets of participants showed statistically similar contribution behavior, irrespective of the level of trust.
25 But as per my prediction that punishment would be more effective in increasing contribution significantly more in The Netherlands than in India, I do not find any significant difference in their mean contribution levels in ‘punishment’ treatment. This seems to be a direct result of variation between the contribution behavior of trusting, non-trusting and unsure sets of participants. We see this difference due to a lack of a large enough representative sample size. Table 1 shows the difference in the number of participants who trust, do not trust and are not sure. The small sample of trusting Indian participants compared to a larger sample of trusting Dutch participants and a small sample of non-trusting Dutch participants compared to a larger sample of non-trusting Indian participants causes their means to be skewed upwards which causes the main result to be insignificant.
Result 1 and 3, however, are in line with previous literature. Threat of credible punishment does indeed significantly increase overall cooperation, irrespective of given level of trust in a society, and it also manages to significantly decrease overall free riding. But frequency of free riders and their reaction to threat of credible punishment differs between both sets of participants. Even though Indian participants contributed statistically as much as their Dutch counterparts to public goods when faced with threat of credible punishment most of the Indian free riders from the first treatment continued to free ride in the second treatment. As such, there was a significant decrease in free riding amongst Dutch participants but not amongst Indian participants.
The Netherlands being a high trust society, cooperation is a social norm. The Dutch participants who chose to deviate and violate the social norm in the first treatment reacted positively to punishment, and restored order by contributing to the public good in the second treatment. Gintis (2008) conjectured that this may be due to symbolic punishment or inducement of guilt amongst participants which makes them increase their contribution but my experimental model and design makes it difficult to align the results with his conjecture because inducing guilt would require some form of prerequisite social interaction and thereby, building up of reputation effect. This, though, is not the aim of my study. It does, however, align with Coleman’s (1988) theory that violation of the social norm would cause members of a high trust society to undertake the cost of punishment in order to correct cooperation behavior. We see that the cost of punishment was more equally borne by Dutch participants than by Indian participants. Even though our second hypothesis is rejected one insight can be drawn from the
26 finding above. Not all Indian participants believed that punishment would be effective in enhancing cooperation. As such, as many as 38% of the Indian participants chose not to bear the cost of punishment which would ensure higher cooperation. On the contrary, 87% of the Dutch participants chose to bear this cost of punishment, diffusing the responsibility of ensuring higher cooperation more or less equally and thereby, also indicating that they trusted the sanctioning system to enhance cooperation.
This may explain that why even though the Dutch punished less, they contributed as much as their Indian counterparts, that is, they shared the responsibility of contribution to public goods also more equally. This result can be attained by the significant decrease in free riding amongst Dutch participants. Furthermore, even though Dutch participants spent significantly less on punishment, they achieved the same statistical level of cooperation as their Indian counterparts, thus, also ensuring higher individual payoffs. Amongst Indian participants most of the original free riders still continued to free ride and chose to maximize their own payoffs, leaving the responsibility of contributing to the public good along with the bearing of the cost of punishment to other society members.
Limitations and improvements
In order to get significant results from my study, the sample size would have to consist of at least 66 participants10. My study had 60 participants, out of which 6 observations had to be discarded due to invalid data entries. This reduced my sample size to 54 participants which led to inconclusive results. It consisted of 31 Indian participants and 23 Dutch participants most of whom had at least a Bachelor’s level education, which is not nearly a representative sample11 of the societies in question. Therefore, given the sample size, first, we do not find significant results, and second, even if we did we could not draw any conclusions about how the effect of punishment co-varies with the level of trust in its effect on cooperation for these given societies. One definite conclusion we may allow ourselves to draw is that the effect of punishment does in fact vary with the given level of trust in a society, as has been seen in previous literature
10
Conducting a cox sample size analysis, at significance level of 5%, for a standard deviation of 0.5 we get a minimum sample size of at least 66 that is required in order to obtain significant results.
11
The official trust levels reported through the World Value Survey report are reported for similar demographic characteristics – for people between age 18-30, and education level consisting of at least a university degree.
27 (Henrich et al. 2006). Therefore, it is hard to substantiate any claim from this study except for the ones already substantiated in previous studies.
Since, in order to gather responses from people residing in their own country the experiment was conducted online. This limited the number of sessions I could conduct the game for to only one. Having more than one session would lead to a decreasing participation rate since it would be difficult to get participants to actively fill out the experimental survey without having a controlled environment like a laboratory. In a tradeoff between having more sessions or having more observations, I chose to have more observations. This also necessitated the use of Strategy Method which also impacted the design of the game that was played – a one shot public goods game. As such, this study does not allow for behavioral learning or strategy correction on part of the participants as is normally the case in a more ‘real world’ situation. Furthermore, since the experiment was conducted online, I had less control on the mental state of participants while filling out the experimental survey which, as pointed out earlier, led to invalidation of some observations.
In order to so isolate the role of punishment in high and low trust societies in enhancing cooperation, the experiment design was kept simple. Of course, there are various other factors that may dictate why or how people choose to trust or not trust their fellow society members. In that, we may not draw conclusions simply based on this study. The Netherlands has a well-defined social welfare system which is absent in India. As such, The Netherlands has far more social equality, trust and stronger social norms than India (Knack and Keefer, 2008). Similarly, there may be more cultural variables that dictate how both the societies function and it may also have a consequence for how both countries perceive trust, punishment and cooperation. These would have to be incorporated in the study to truly observe the effect of punishment in any society. Hence, any future study aimed at asking this question will have to also address these social and cultural variables before being able to draw any reasonable conclusions.
My study findings do reinforce the role of punishment in enhancing cooperation, given its interaction with different levels of trust. I found evidence of punishment enhancing cooperation and reducing free riding. I also found evidence of the difference in amount of punishment required in enhancing cooperation in both India and The Netherlands. India required more punishment contribution than The Netherlands to achieve the same level of cooperation levels.
28 Unfortunately enough, I did not find a significant difference in the contribution levels as predicted.
This study also holds implications for public policy. It is imperative to keep economic framework not too far removed from the society it is implemented – it is human behavior that has built up pyramids, sent man to the moon and created the best The Beatles album that there is. It is this human behavior that may warrant punishment for inducement of cooperation but it is also this human behavior that projects itself as trustworthy through reciprocation of other people’s trusting behavior. In that, it is intriguing and increasingly important that we question the evolution of social norms and culture of a society to understand whether the society in motion is a consequence of continued deterrence or a consequence of continued faith in humanity.
29 Appendix A
Repeating analysis with invalidated data: Comparison
In order to analyze if there are any statistical differences in our results by inclusion of the invalidated data of six participants I conduct the same analysis with 60 participants. First, I conducted a Shapiro-Wilk Normality Test in order to determine that mean contribution data for both treatments was normally distributed for both sets of participants. The null hypothesis could not be rejected for either of the four mean contributions, ascertaining that the data was normally distributed.
Result 1: There was no significant difference caused by the inclusion of invalid participant data. We still find a significant increase in mean contributions from ‘no punishment’ to ‘punishment’ treatment for both sets of participants.
Conducting a paired t-test showed that presenting the participants with an opportunity to punish their group members for their given contribution schedule led to a significant increase in their mean contributions with p<0.001 for both sets of participants, thus, reinforcing the results obtained with 54 participants. Thus, there was no significant difference caused by the subtraction of invalid participant data.
Result 2: There was a significant difference in the contributions between Dutch and Indian participants in the ‘no punishment’ treatment due to inclusion of invalidated data. However, the results for ‘punishment’ treatment were statistically same.
Unlike in results with 54 participants, conducting a t-test with 60 participants, I found a statistically significant difference in mean contributions between Dutch and Indian participants in the ‘no punishment’ treatment with t=2.595 and p=0.0173. Investigating the reason for this variation in result, I find that the mean contribution in ‘no punishment’ treatment with the inclusion of invalid Dutch participant data, was 6.31 euros as compared to 5.97 Euros in the original analysis. Even though I found this difference, I did not find any significant increase in the mean contributions by Dutch participants over Indian participants in the “punishment’ treatment with t=0.667 and p=0.512 as compared to with t=0.730 and p=0.473 (with 54 participants) which as we can see has no statistically significant difference. Thus, invalidating
