Discrimination and team performance:
The differences between random and non-random
income heterogeneity on free riding behavior
Eveline V.A. Loozen – 10867910 // 28 June 2015 – final version // MSc. Business Economics - Organization Economics // Dhr. Prof. Dr. Randolph Sloof
2
Statement of originality
This document is written by Student Eveline Loozen, who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than 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 to the work, not for the contents.
3
Discrimination and team performance:
The differences between random and non-random
income heterogeneity on free riding behavior
By Eveline Loozen
Abstract
Team production is very important for an organization to prevail above the market. But it also creates free riding behavior. To alleviate this problem as much as possible it is important to know how to best structure incentives within teams. This thesis studies the influence of discrimination on team performance due to measure the willingness of individuals to contribute to a team project under different degrees of income heterogeneity. To find out to what extent the feeling of being treated unfairly has influence on the behavior of people I used a one shot public goods experiment with a team size of two and three different rounds: a round without income differences, a round with random income differences and a round with income differences based on gender (non-random). It seems that non-random differences in income lead to more free riding behavior, because, to a certain extent, it vanishes out fairness and equity concerns. For that reason it is important that incentives are not based on non-random or discriminatory grounds because this could have negative effects on team performance.
4
Index
Abstract _____________________________________________________________________ 3 1. Introduction ______________________________________________________________ 5 2. Previous literature _________________________________________________________ 9 3. Models of preferences & hypotheses __________________________________________ 12 3.1 Models of preferences ____________________________________________________ 12 3.2 Hypotheses ____________________________________________________________ 16 4. Experimental design ______________________________________________________ 18 5. Results _________________________________________________________________ 21 5.1 Results on free riding ____________________________________________________ 21 5.2 Results on fairness and equity concerns ______________________________________ 24 5.3 Effect of treatments ______________________________________________________ 26 6. Discussion ______________________________________________________________ 30 7. Conclusion ______________________________________________________________ 32 References __________________________________________________________________ 34 Appendix ___________________________________________________________________ 37 AI. Logbook ______________________________________________________________ 37 AII. Instructions ___________________________________________________________ 38 General instructions _______________________________________________________ 38 Practice round ___________________________________________________________ 39 AIII. Decisions sheets and questionnaire ________________________________________ 40 Ronde [no differences] ____________________________________________________ 40 Ronde [random differences, good luck] _______________________________________ 41 Ronde [random differences, bad luck] ________________________________________ 42 Ronde [non-random differences, man] ________________________________________ 43 Ronde [non-random differences, woman] _____________________________________ 44 Anonieme vragenlijst ______________________________________________________ 45
5
1. Introduction
What is the fundamental base of an organization? As an answer to this question Alchian and Demsetz (1972) state that it is team production, in which a joint use of inputs yields a larger output than the sum of the products of the separately used inputs. When there are these synergetic advantages to achieve, the organizational firm prevails above the market. Rosen (1986) suggests, as another beneficial of team work, that it offers compensating differentials resulting from non-pecuniary benefits, such as more social interaction, less boring work and income smoothing. Nowadays, most organizations have implemented some form of team production for a wide range of productive activities (Hamilton et al., 2003; Van Vijfeijken et al., 2006). However, team production is amendable to free riding. According to the theory of the Homo Economicus, standard economic theory suggests that every agent is motivated by material self-interest only. He is rational and consistent in his choices, always prefers more to less and is able to efficiently allocate his limited income among the numerous things he desires to maximize his utility (Edgeworth, 1881). When teams are evaluated according to its joint output and the individual efforts are hard to observe, it is easy for employees to shirk. The one that shirks receives full benefits by reducing effort and only partially bears the costs of reduction in output, since this is divided among all team members (Osterloh, 2007). This makes free riding very attractive for individuals. Nevertheless, it has negative consequences for the collective interests, since team performance will be lower. However, it seems that there is a contradiction between this theory and the practice of team production and free riding. An increasing number of experimental studies in behavioral economics showed that standard economic theory is often rejected in practice (Charness and Rabin, 2002; Fehr and Schmidt, 1999; Dawes and Thaler, 1988). Traditional economics might not be fully capable of explaining the frequent existence of team production and a lack of free riding in reality. Economists consider important factors that may influence the decision-process of individuals. One of the most important ones is that people are not entirely selfish and have non-traditional preferences in which fairness and equality have a pivotal role.
Fairness is a relative concept that involves comparisons among people. The agents among whom comparisons are made are called the reference group. One item in which fairness is very visible is the difference in income between colleagues. Therefore agents’ effort choices have a lot to do with
6
the allocation of rewards within an organization. In the literature there seems to be a dilemma around this subject. From an economic point of view, the bonus system can be seen as a tournament, which suggests that rewarding agents may increase effort (Lazear and Rosen, 1981). But the payment of a bonus automatically results in income heterogeneity among the agents within the organization. Agents with approximately the same job description can have very different incomes, while it is likely that they value their rewards through a process of social comparison (Ockenfels et al., 2010). Most of the time, their direct colleague is their reference point and inequalities have possibly negative effects on morale and effort. How agents react to income heterogeneity depends on the perceived fairness of their salary.
Akerlof and Yellen (1990) emphasize the importance of fairness in developing their ‘fair wage-effort hypothesis’. Their core assumption is that agents compare their wage w to the fair wage w* and they theoretically show that employees tend to withdraw effort if the wage paid is smaller than the wage they perceive as the fair wage.
Individuals who are confronted with income heterogeneities within their team must determine how to interpret it, how the scenario relates to their own fairness values and finally how to respond to this specific case. Models of altruism, inequality aversion and reciprocity can help by predicting how inequality influences the effort decisions and offer different predictions of individual behavior, particularly with regards to absolute and relative contributions to the team project. The models expect ‘social preferences’, which assumes that agents behave perfectly rational. In the first two models the outcomes are purely outcome based and therefore the well-known concepts of traditional utility and game theory can be applied to analyze optimal behavior (Fehr and Schmidt, 2001). The last model focuses on reciprocity in public goods games and is not only outcome based but also depends on beliefs (Fehr and Schmidt, 2001).
The model of altruism is introduced by Becker (1974) assumes an increasing utility function that depends both on its own income as on the income of other individuals (see also Section 3). The two incomes together constitute the ‘social income’. An altruist will act to maximize this social income and therefore will refrain from all actions that lower the income of others by more than they increase his own, because this would reduce the social income and hence the utility of the altruist himself.
7
Utility can also be influenced by inequality aversion, which means that individuals dislike unequal incomes or consumption. The individual’s utility is not only derived from his own payoff, but also from his position with respect to the other members of his group (see also Section 3). People dislike obtaining either a higher or a lower payoff and, therefore, utility will increase in the equality of the payoffs of all players (Fehr & Schmidt, 1999).
Sugden (1984) suggests that individuals reciprocate by contributing the same fraction of their endowment as the other group members will do, because individuals bring notions of fairness to each interaction. This means that in situations with income inequality individuals contribute with different absolute amounts, but with the same percentage of the total endowment: they contribute a ‘fair share’ to the public good account, measured by the size of the contribution as proportion of their income (see also Section 3).
Since notions of fairness and equity concerns appears to have a great impact on free riding, it is very important to know when something is perceived as fair and when it is not. In this sense, Greenberg (1982) noted that the definition of fairness is subject to interpretation. Therefore, Konow (2000) comes with the accountability principle as rule of justice. It says that something is fair if it varies in proportion to relevant variables that someone can influence (e.g. effort), but not according to variables that someone cannot reasonably influence (e.g. gender). When someone is treated on those last variables, he will perceive this as discrimination: “an individual believes that he or she has been treated unfairly based on race, gender, age or any other such characteristics” (Harris, Lievens and Van Hoye, 2004, p.6). This is in line with the equity theory of Adams (1965), which predicts that inequality is only acceptable if it is justified by in differences in performance or merit.
Both experimental and theoretical research on social preferences show evidence that fairness concerns are strongly influenced by perceived intentions (Charness, 2004; Falk and Fischbacher, 2006; Falk, et al., 2008). These papers show that there is an aversion against intentional discrimination, which leads to effort-reducing behavior. Therefore, contribution to a team project might be different if income heterogeneity arises from ‘fair’ intentions than when they arise from ‘unfair’ intentions. If an agent feels discriminated and treated unfairly, this will have a negative effect on workers’ moral and therefore he will react with more free riding (Akerlof and Yellen, 1990).
8
In this thesis I want to find out to what extent the feeling of being treated unfairly has an influence on the behavior of people in a team project. I use a linear public goods experiment to make the individual inputs observable such that I can explicitly examine the impact of income heterogeneity on contributions to the public good. The heterogeneity is introduced both through random and non-random allocation of differing player endowments. The research question is:
‘Is there a differential impact of random and non-random income heterogeneity on free riding behavior within teams?’
It seems that non-random differences in income lead to more free riding behavior, because, to a certain extent, it vanishes out fairness and equity concerns. For that reason it is important that incentives are not based on non-random or discriminatory grounds because this could have negative effects on team performance. An important remark that should be made is that this most likely also depends on the gender distribution within the team.
The remainder of this thesis is structured as follows. I will start in Section 2 with providing an overview of the existing literature in field data studies and public goods experiments with income heterogeneity. Next, in Section 3, I will discuss some models of preferences and derive the hypotheses. Section 4 will describe the experimental parameters and implementation of the experiment. Next, I will display the results of the experiment and test the hypotheses in Section 5. In Section 6, I discuss the limitations and give some directions for future research. The last section summarizes some conclusions from the main findings and comes up with the answer of the research question of this thesis.
9
2. Previous literature
To answer the research question I have to analyze the effect of both random and non-random income heterogeneities within teams and look at how these heterogeneities are related to the team project contributions.
The relation between income and contribution to a collective account is evaluated in multiple field data studies (e.g. Andreoni, 2006; Vesterlund, 2006). Most of these studies show that income and contribution to the collective account are positively related and vice versa. The natural field experiments of Kube et al. (2013) and Cohn et al. (2011) give results in the same direction. Their experiments provide evidence that decreasing income has a significant and large negative impact on work moral and leads to a substantial decrease in performance. The work of Cohn et al. (2011) added evidence for social comparison among workers: when only one team member’s wage was cut, the performance decrease of this worker was more than twice as large as the individual performance decrease when both workers’ wages were cut.
I want to investigate the willingness of individuals to contribute to a team project under different degrees of heterogeneity. Since free riding occurs most often when teams are evaluated according to its joint output and individual efforts are hard to observe (Osterloh, 2007), I might not be able to easily measure free riding behavior as reaction to income heterogeneity in the field. Therefore, I will use a laboratory experiment, in the form of a simple public goods game, to measure the degree of free riding per individual. This is appropriate because I am able to manipulate the form and salience of inequality between subjects and measure the differences in behavior. Besides it allows for isolating the causal effect of inequality and to avoid the problem of endogeneity, which is needed for answering the research question (Croson and Gächter, 2010; Falk and Heckman, 2009).
Since Bohm (1972) provided the first study of a public goods game in an experimental setting, a multitude of these studies have been undertaken. While experimental designs may vary, the most common approaches is the Public Goods (PG) game. The PG experiment used in this study is a variation of the game first introduced by Ledyard (1995) and can be seen as a standard paper-and-pencil linear PG game. Each of the participants in a team of two members is given an endowment of tokens, which have to be allocated between their private account and a public account. This is
10
a fixed amount of 10 tokens per round. The tokens that are placed in the private account are kept by the individual for himself and gives a return in one token to the individual. Those tokens placed in the public account are summed up and multiplied by 1.5. The total amount in the public account is shared equally by all participants in the team, irrespective of whether the individual contributed to the public account or not. Hence, each team member earns from the public account and from his private account. Besides the tokens with which the game is played, each team member gets a personal fixed salary. This salary is different per round per person and depends on randomness or gender. The willingness to contribute or not and the utility individuals may get from contribution will be further elaborated in Section 3.
Variations on this standard PG game have been used extensively in psychology and economics experiments for almost three decades now. Several of these are reviewed by Ledyard (1995). While many experiments have been run using this pattern, the differences can be broadly classified along two dimensions: the provision of endowments – homogeneous or heterogeneous - and the design
of the game – a one-shot or repeated game, a linear or non-linear game, allowing for
communication or not, punishments for free riding etcetera.
In the provision of endowments the experimenters have to choose between homogenous or heterogeneous endowments. Contrary to my experiment, almost all experiments are involved with homogeneous endowments. However, some experiments did introduce income heterogeneity into PG games. They used various methods, like variable show-up fees (e.g. Anderson and Stafford, 2003) or variable endowments (e.g. Chan et al., 1999) for the participants, but all endowments where randomly assigned to them. The results from these studies are mixed, with some authors finding that income heterogeneity tends to reduce contributions to the public account (e.g. Isaac and Walker, 1988; Anderson, et al., 2003), while others report higher total contributions (Cherry, Kroll & Shrogen, 2005; Chan et al., 1996).
These mixed findings might be assigned to variations in the design of PG games. Some of those studies used a linear PG game (e.g. Isaac and Walker, 1988; Anderson, et al., 2003) while others used the non-linear (e.g. Chan et al., 1996) variant. In a linear PG game, the Marginal Per Capita Return (MPCR) stays constant and is the same at all levels of contributions, whereas in a non-linear PG game the MPCR varies as the level of contributions change. Therefore, the Nash equilibrium prediction differs between these two variations in the design, which could explain the mixed findings.
11
This is also the case when looking at repetition in PG games. In single round ‘one shot’ PG games, individuals do not play the Nash equilibrium, but generally contribute between 40 and 60 percent of their endowment to the public account (Dawes & Thaler, 1988). In multiple round PG games, repetition reduces contribution rates to the public account, but remain at between 15 to 25 percent of the endowment by the final round (Ostrom, 2014). The two most common explanations are, firstly, that individuals learn to play the optimal strategy and, secondly, that they strategically try to make others contribute.
More communication among participants generally increases contributions to the public account (Dawes, McTavish and Shaklee, 1977) and the effect of financial punishments for free riding (Anderson and et al., 2003) give mixed findings on contributions.
In the existing patterns and the dimensions mentioned above, one important aspect is missing. If endowments are heterogeneous, but each participant has the same chance of getting high or low endowments (= random), they may feel treated identical and there might be no effect of income inequality on worker morale. This would mean that the earlier experiments cannot display the relevant inequality of income and, consequently, not the effect on free riding behavior. Therefore, I introduce income heterogeneity in my game by providing participants with unequal token endowments on non-random arguments. The experimental design is based on ‘pure’ pay inequity situations. That is, inequitable situations that are not justified by differences in performance or merit, which equity theory (e.g. Adams, 1965; Konow, 2000) would predict are acceptable, but on differences that are perceived as ‘unfair’ and may lead to negative effects on morale and effort. I choose someone’s gender as non-random argument for inequality.
So, while other research has addressed the impact of heterogeneity of incomes on group contributions in one shot, linear PG games, I look further into this phenomenon by adding an extra dimension. I adopt the one shot, linear PG game framework without communication or financial punishments to keep the experiment as simple as possible and focus the attention on how
non-random income heterogeneity affects contributions on both the positive and the negative
discriminated members of the team. To measure this effect I let the subjects play three rounds were I compare income homogeneity (ND) with random income heterogeneity (RD) and non-random income heterogeneity (GD). For more details about the rounds, see Section 4.
12
3. Models of preferences & hypotheses
3.1 Models of preferences
The basic model is based on a standard linear PG game (Ledyard, 1995). The participants are assigned to teams of two people. Each participant is endowed with a personal fixed salary (FSi)
and an endowment of Z tokens. The personal fixed salary is given and cannot be used during the game. The endowment of Z tokens can either be kept for the private account, Pi, or contributed to
the public good account, Ti. The monetary payoff of each participant is given as:
𝑋𝑖 = Z − 𝑇𝑖 + MPCR ∗ (𝑇𝑖+ 𝑇𝑗) + 𝐹𝑆𝑖 (1),
where the public good is equal to the sum of the contributions of both group members denoted by (𝑇𝑖+ 𝑇𝑗). Contributing a token to the public good yields the Marginal Per Capital Return (MPCR): the ratio of the private value of one unit of the public good to the private value of the private good needed to provide that unit (Isaac and Walker, 1988). This means that for every token a participant spends on privately providing the public good, the MPCR measures how much the individual gets back.
When maximizing the monetary payoff of the team, we get:
𝜕(𝑋𝑖+𝑋𝑗)
𝜕𝑇𝑖 = −1 + 2 ∗ MPCR (2).
When the MPCR > 1/2, it would be Pareto efficient to contribute fully to the public good account such that Ti = Z. This would lead to the highest aggregate monetary payoff for the agents.
But when maximizing the individual monetary payoff of the agent, we get:
𝜕𝑋𝑖
𝜕𝑇𝑖 = −1 + 1 ∗ MPCR (3).
As predicted by economic theory, every agent is motivated by material self-interest only. Therefore I assume that the utility of an agent is based on monetary payoff only (Ui = Xi). Hence, when the
13
MPCR < 1, it is considered that Ti = 0 is the dominant strategy, which means that it is predicted
that both group members free ride completely.
The classical problem of public goods arises if ½ < MPCR < 1. This is where the amount that would be voluntarily contributed to maximize individual payoff is not equal to the amount that would maximize the monetary payoff of the team and consequently would be societally efficient. But it appears that individuals not always maximize their utility by maximizing only their own monetary payoff. Early experimental work on public good provision established that agents tend to provide public goods at higher rates than presumed by the theory described above (Smith, 1980). Ledyard (1995) defines this further with the finding that in one-shot PG games participants generally provide contributions halfway between the Pareto-efficient level and the free riding level. Researchers have been moving beyond simple characterizations of utility to include a variety of “other‐regarding preferences” on the part of participants (Sobel, 2005). One explanation provided for this deviation involves the non-traditional preferences of people, wherein fairness has a pivotal role. Numerous experiments suggest that many people are not purely selfish, but are also led by behavioral motivations like fairness or equity concerns.
In this experiment, the unfairness and inequality between participants will be larger than in a normal symmetric PG game with homogeneous incomes for both team members or heterogeneous incomes based on randomness. Therefore, the competing models of the pure self-interest standard model are even more interesting. The following models offer different predictions of individual behavior, particularly with regards to absolute and relative contributions to the public account, and help me to derive the hypotheses.
The first model is Becker’s (1974) model of altruism. He models a utility function that is comprised of two elements: the agent’s own income and the income of his team member. Hence:
𝑈𝑖(𝑋𝑖, 𝑋𝑗) = 𝑋𝑖 + 𝑋𝑗 (4),
where Xi and Xj are, respectively, the monetary payoff to the altruist and to the other member of
his team. The payoffs together constitute the ‘social income’. An altruist will act to maximize this social income and therefore will refrain from all actions that lower the payoff of others by more than they increase his own, because this would reduce the social income and hence the utility of the altruist himself.
14
When maximizing the utility of an altruist, we get:
𝜕𝑈𝑖
𝜕𝑇𝑖 = −1 + 2 ∗ MPCR (5),
whereas in all cases for MPCR > ½ it would be optimal to contribute fully to the public good account such that Ti = Z, because these would always led to the highest utility. There will be no
difference in contribution to the public account between the different rounds, since all
individuals simply have in each round the same endowment of Z tokens to play the game with. Next, there is the model of inequality aversion. This model states that aversion to inequalities drives cooperation and that the utility of an individual increases in the equality of the payoffs of all team members (Fehr and Schmidt, 1999). This implies that individuals derive utility not only from their own payoff, but also look at how their payoff compares to the payoff of the other team member. They dislike both inferiority and superiority, which means that they dislike obtaining either higher or a lower payoff than the other. Specifically, utility is assumed to take the following form:
𝑈𝑖(𝑋𝑖, 𝑋𝑗) = 𝑋𝑖− 𝛼𝑖𝑚𝑎𝑥{𝑋𝑗− 𝑋𝑖, 0} − 𝛽𝑖 𝑚𝑎𝑥 {𝑋𝑖 − 𝑋𝑗, 0} (6),
where Xi and Xj are the payoffs of the two team members. Parameter 𝛼𝑖 denotes the level of
inferiority aversion and 𝛽𝑖 is a parameter of superiority aversion. It is assumed that 𝛽𝑖 ≥ 0,
because individuals are not willing to ‘burn’ their money to eliminate advantageous inequality. Besides it is assumed that 𝛼𝑖 ≥ 𝛽𝑖; the disutility that comes from a position of disadvantage is higher than the disutility that comes from a position of advantage.
When maximizing the utility of an inequity averse individual, we get:
𝜕𝑈𝑖
𝜕𝑇𝑖 = −1 + 1 ∗ MPCR − 𝛼𝑖 + 𝛽𝑖 (7).
Hence, when the MPCR < 1 − 𝛽𝑖+ 𝛼𝑖, it is considered that Ti = 0 is the dominant strategy for
individual i. Since 𝛼𝑖 and 𝛽𝑖 are private information, it is unknown above which value of the MPCR an individual would contribute to the public account. But subsequently with the assumption that 𝛼𝑖 ≥ 𝛽𝑖, the MPCR should exceed 1 before an individual would contribute. If this is not the case, it is predicted that an individual would free ride completely.
15
In the PG game where income heterogeneity is introduced with differences in the FSi, the utility
functions of the two team members will vary from each other. Maximizing the utility of an inequity averse agent, who receives the higher endowment (advantaged treatment) gives:
𝜕𝑈𝑖
𝜕𝑇𝑖 = −1 + 1 ∗ MPCR + 𝛽𝑖 (8),
whereas maximizing the utility of an inequity averse agent in a disadvantaged treatment gives:
𝜕𝑈𝑖
𝜕𝑇𝑖 = −1 + 1 ∗ MPCR − 𝛼𝑖 (9).
Condition (8) and (9) require respectively that the MPCR should be above 1 − 𝛽𝑖 and 1 + 𝛼𝑖 before contributions will be made to the public account. Since a MPCR of 1 − 𝛽𝑖 is more easily to
achieve than a MPCR of 1 − 𝛽𝑖 + 𝛼𝑖 (required under homogeneity of income), this model predicts that an agent would rather contribute (part of) his endowment Z in an advantaged position than when treated the same as his team member, to reduce equalize earnings within the team (Buckley & Croson, 2006). The other way around, the low-endowed agent will reduce his contribution to the public account and wants to take some of the benefits of the high-endowed agent, since a MPCR of 1 + 𝛼𝑖 is harder to achieve than a MPCR of 1 − 𝛽𝑖 + 𝛼𝑖.
Last, the reciprocity model of Sudgen (1984) explains contributions to the public account based on the theory that every agent in the team will contribute his ‘fair share’. Every agent has the moral duty to contribute by at least this fair share. This can be seen as a social norm. Therefore, the agent does not systematically come to a choice through calculation, but conforming to an ethic rule how much he will contribute (Gauthier, 1986). Every agent chooses an percentage (p) of Xi he would
most prefer as contribution from every team member. The agent is under an commitment to the other team member to make a contribution of at least this most preferred level and thus a percentage of Xi, such that: 𝑇𝑖 = 𝑝𝑋𝑖.
This model assumes that in a PG game with income heterogeneity, both agents will tend to contribute the same fraction of their income to the public account to conform to a ‘fair-share’ threshold (Hofmeyr, Burns and Visser, 2006). This would predict that the absolute contribution of the high-endowment agent is larger than that of his low-endowment team member (T low FSi < Thigh FSi), since the monetary payoff of the high-endowment agent is larger than that of his
16
3.2 Hypotheses
To provide a clear answer to the research question I divide this question into two hypotheses. The first hypothesis is to test if there exists a difference in free riding between the three different rounds. The disparity between the rounds is made in the differences in the FSi that every subject received.
This FSi could not differ (ND: FS1 = FS2), differ at random (RD: FSgood luck > FSbad luck) of differ by
gender (GD: FSmale > FSfemale). The second hypothesis will test if these differences in FSi also cause
differences in fairness and equity concerns.
Above models about fairness preferences explain and predict why people may make contributions to the public account under either homogeneous incomes or heterogeneous incomes. But these models do not say anything about the intention behind the heterogeneity of the incomes. The equity theory of Adams (1965) says that high contributions and high levels of motivation can only be expected when agents perceive their treatment to be fair, which is strongly influences by perceived intentions (Charness, 2004; Falk and Fischbacher, 2006; Falk, et al., 2008). Differences in incomes on a non-random base that someone cannot influence, like gender, are not justified by differences in performance or merit and will be perceived as discrimination. Therefore this form of income heterogeneity can be perceived as unfair, whereas income heterogeneity based on randomness can be perceived as just unlucky instead of discrimination. In short, this means that the GD round is more unfair than the RD round than the ND round which may lead to different effects on morale and effort. That is why I predict that:
Hypothesis 1: Discrimination stimulates free riding: free riding will be strongest when incomes differ with gender (non-random) and weakest with equal incomes. The treatment with random income differences is in between (GD>RD>ND).
Free riding is defined as the absolute amount an agent will keep for himself, in his private account (i.e. Z - Ti).
Due to the missing of the intention-based element is the prediction I make in Hypothesis 1 contradictive with the theory of the models of altruism, inequity aversion and reciprocity:
17
The model of altruism explains why contributions are made to the public account, while standard economics states that this will never maximize someone’s own utility. Regarding to the experiment, this model predicts no changes between the various rounds if someone is a ‘real altruist’. A change between the rounds would suggest that someone’s fairness and equity concerns have changed due to the different intention behind the heterogeneity
The model of inequity aversion predicts a total of more free riding under income heterogeneity than under income homogeneity, where the individual with a high income free rides less to a smaller extent than the individual with the low income free rides more. But it does not predict a different outcome between the round with RD and the round with GD. A change between those two rounds would also suggest that someone’s fairness and equity concerns have changed due to the different intention behind the heterogeneity.
The same applies to reciprocity. It predicts differences between the rounds with homo- (ND) and heterogeneous (RD & GD) incomes, but no differences between the two different heterogeneous rounds. If there is a variation between those two, it would also suggest that someone’s fairness and equity concerns have changed due to the different intention behind the heterogeneity.
Since the GD round is perceived as unfair and the RD round only as unlucky, I predict that these rounds will show different outcomes based on differences in fairness and equity concerns due to the different intentions behind the heterogeneities. Therefore, I expect that:
Hypothesis 2: Discrimination vanishes out fairness and equity concerns: fairness and equity concerns will become less important when the incomes differ with gender (non-random) than when the incomes differ at random than when the incomes are equal (GD<RD<ND).
The fairness and equity concerns will be measured in terms of altruistic behavior, inequity aversion and reciprocity.
18
4. Experimental design
To test the hypotheses, I conducted a within subjects experiment. This experiments consisted of three rounds of the one shot linear PG game descripted in Section 2 and 3, where the endowment of Z tokens was 10 in every round for each subject and the MPCR is 0.75. The disparity between the rounds was made in the FSi that the two team members received. This FSi could not differ,
differ at random of differ by gender.
In the round where there were no differences between the subjects, everyone is treated the same and received a FSi of 15 tokens. I call this round the ‘no differences’ (ND) round. In the other two
round, the FSi was not homogeneous. One half of the subjects received a low FSi of 10, whereas
the other half received a high FSi of 20. In the round with ‘random differences’ (RD) there were
two different treatments: some had just good luck (FSi = 20) or bad luck (FSi = 10). In the round
with ‘gender differences’ (GD) the treatment was based on gender, where men received the high
FSi of 20 and women the low FSi of 10.
Table 1: Summary of rounds and treatments
The experiment was conducted on May 11th, 12th and 13th 2015 at the Willibrord Gymnasium in
Deurne in six different sessions. Each session had another sequence of the three different rounds (see Table 2) to control for order effects. The sample consists of students from the 2nd, 3rd and 4th grade with ages between 13 and 17 years. They participated in the experiment during three different days and only in one of the six sessions. Of the in total 137 participants, 81 were men and 56 were women. High school subjects were selected for this experiment because the incentives for their decisions where better payable than that of university students and at the same time these subjects would be old enough to understand the investment decision they had to make (Marwell and Ames, 1979). To test of all subjects understood the task, I introduced a pre-game survey with
Rounds Incentives per treatment
ND Everyone: FSi : 15 Z = 10
RD (‘unlucky’) Good luck: FSi : 20 Z = 10 Bad luck: FSi : 10 Z = 10
19
an example and some questions about it. Someone could not start the game before he or she answered all questions correctly.
Table 2: Summery of experiment per session
Session 1 Session 2 Session 3 Session 4 Session 5 Session 6 Round 1 ND ND RD RD GD GD Round 2 RD GD ND GD ND RD Round 3 GD RD GD ND RD ND # subjects 22 22 27 23 25 18 % male 0.73 0.64 0.56 0.43 0.60 0.56 % luck 0.45 0.41 0.56 0.52 0.52 0.56
Mean Free Riding (Std. Dev.) 5.02 (3.04) 6.44 (2.55) 4.67 (3.05) 5.61 (2.50) 5.17 (1.61) 4.54 (1.96) Grade 3rd 3rd 4th 4th 2nd 2nd
All participants were divided into two-person groups, varying between the rounds. How they were divided into couples differed per round and depended on the number on the card they randomly took when they came into the classroom. These cards were split in two stacks: one for men and one for women. The women got the odd number and the men the even numbers. In the round with no differences in income, the first half of the numbers was matched with the second half of the number (e.g. 1-13, 2-14, 3-15, etc.). The round were differences were made on a random base couples were made with opposite numbers (e.g. 1-25, 2-24, 3, 23, etc.) and in the round with differences between the genders the participants were matched with the previous/next number (1-2, 3-4, 5-6, etc.) where some were used twice as dummy for incomplete couples. The subjects did not know with whom they play together during the rounds.
All subjects knew that their performance in this experiment would only have consequences for their payment at the end of the experiment and not on school performance, since the scores will not be given to their teacher. Also, the subjects were aware that they were not allowed to communicate and did not get any feedback during the experiment.
20
After the three rounds of PG games I hand out an after game survey (see Appendix AIII). In this questionnaire I asked for their age and their experience with PG games. This information could be used when surprising results came up. Further I asked if they felt that they played against or together with their team member to measure to what extent they feel connect to their team member. I also asked whether they felt treated unfair during the different rounds to see if the free riding behavior is actually driven by this feeling. To make it more clear which of the fairness and equity concerns has the most impact on free riding, I measured altruism, inequality aversion and reciprocity separately and the changes during the different rounds.
The total tokens earned at the end of the game were paid in coins. On average, the participants earned 76.52 coins with a standard deviation of 8.92. With these coins the participants could buy something afterwards (in the break). With less than 74 coins, they could buy a small candy bar. When they had earned between 74 and 89 coins they could buy a little bag of chips and when the earnings were above 89, it was possible to buy a large candy bar.
21
5. Results
This section provides an overview of the results of the experiment and is divided into three basic subsections. First the results regarding to free riding will be discussed, which test hypothesis 1. Next, the results about fairness and equity concerns will test hypothesis 2. Finally, I will discuss the effect of being in the (dis)advantaged treatment and the difference in responding between men and women.
5.1 Results on free riding
The theory of Ledyard (1995) predicts that in one-shot games participants generally provide contributions halfway between the Pareto-efficient level and the free riding level, which is in this game equivalent to 5. The means of free riding in the ND round, 5.13, and the RD round, 4.98, are not significantly different from 5 (sign rank, n=137, ND: p=0.63, RD: p=0.98). In contrast, the mean of the GD round is with 5.62 and a p-value of 0.01 significantly higher than the predicted 5 (sign rank, n=137).
Table 3: Descriptive statistics for free riding behavior, by session
Random – gender Mean (min, max) Gender – random Mean (min, max)
Session 1 (n = 22) 5.02 (0, 10) Session 2 (n = 22) 6.44 (0, 10) - No - Random - Gender 5.36 (0, 10) 4.55 (0, 10) 5.14 (0, 10) - No - Gender - Random 6.22 (0, 10) 6.82 (0, 10) 6.27 (0, 10) Session 3 (n = 27) 4.68 (0, 10) Session 5 (n=25) 5.17 (0, 10) - Random - No - Gender 4.67 (0, 10) 3.93 (0, 10) 5.44 (0, 10) - Gender - No - Random 5.60 (0, 10) 4.68 (0, 10) 5.24 (0, 10) Session 4 (n = 23) 5.61 (0, 10) Session 6 (n = 18) 4.53 (0, 10) - Random - Gender - No 4.91 (0, 10) 6.13 (0, 10) 5.78 (0, 10) - Gender - Random - No 4.39 (0, 10) 4.11 (0, 10) 5.11 (0, 10)
The descriptive statistics in Table 3 and Graph 1 display the results about the free riding of the subjects during the experiment. It seems there exists a difference between the three rounds and
22
there arises some sort of pattern in line with Hypothesis 1. To statistically support this
observation I used a Friedman test to check if income differences in the various rounds influence the free riding behavior of the subjects. The Friedman test is a non-parametric statistical test that tests the differences between three or more treatments for studies using the same group of
participants in all treatments. When using this test the dependent variable, the level of free riding, is scaled from 0 to 10, where 0 is no free riding behavior at all and 10 is fully free riding. I will use this test to determine if there is a difference in free riding behavior when income differences vary. The null hypothesis is that the free riding behavior is the same in all rounds. The p-value (0.11) of the Friedman test is only marginally significant and therefore the null hypothesis cannot be rejected. Based on this test it cannot be said that three rounds are significant different from each other.
Graph 1: Descriptive statistics: free riding behavior per session by round
An alternative way to look at the differences between the rounds is in pairs. From the descriptive statistics it can be observed that free riding behavior is stronger in every session in the GD round than in the RD round (GD>RD). In 4 out of 6 sessions the GD round causes more free riding than the ND round (GD>ND). The differences in free riding behavior between the round with ND and RD are less obvious. To statistically test these differences, I used a Wilcoxon signed rank test. This test is a non-parametric statistical test that can be used comparing two repeated measurements on a single sample to assess whether their mean ranks differ.
5 ,3 6 6,2 2 3 ,9 3 5 ,7 8 4 ,6 8 5,11 4 ,5 5 6 ,2 7 4 ,6 7 4 ,9 1 5 ,2 4 4 ,1 1 5 ,1 4 6 ,8 2 5 ,4 4 6,1 3 5 ,6 4 ,3 9 N-R-G N-G-R R-N-G R-G-N G-N-R G-R-N
Free riding per round
23
This test gives the following p-values:
Table 4: P-value Wilcoxon signed rank test: differences in free riding between two rounds
Differences between rounds P-value
ND round – RD round 0.56
ND round – GD round 0.09*
RD round – GD round 0.02**
Note: Significance in indicated by ** for the 0.05 level and * for the 0.10 level.
These outcomes show some significant differences between the rounds. The differences between the ND and GD rounds are significant at the 10% level, whereas the differences between the RD and GD are even more clear with a 5% significance level.
In addition to the free riding aspect, Hypothesis 1 also talks about the aspect of discrimination. To see if it is really discrimination that stimulates the free riding in the GD round, I compare the statistics of Question 1 and 2 out the after-game survey (see Appendix AIII). It asks for pupils’ perceived level of unfairness during the RD and GD round and is measured on a Likert scale. It varies between 1 and 7, where 1 is no unfairness experienced at all and 7 means full experience of unfairness. As said before, if an individual believes that he or she has been treated unfairly based on race, gender, age or any other such characteristics, this is seen as discrimination (Harris et al., 2004). When looking at the difference between the RD and GD round a difference can be observed. In the RD round the unfairness is given a 3.04 out of 7 whereas in the GD round this is 3.38 on 7. A Wilcoxon sign-rank test gives a p-value of 0.01, which indicates that there is indeed a significant difference. This assumes that they experienced more unfairness under GD than under RD. To conclude it can be said that Hypothesis 1 ‘Discrimination stimulates free riding: free riding
will be strongest when incomes differ with gender (non-random) and weakest with equal incomes. The treatment with random income differences is in between (GD>RD>ND)’ is partially true.
When there is more perceived discrimination, there is also more free riding behavior. Since the Friedman test gives no significant result, it should be rejected that GD>RD>ND. But due to the Wilcoxon signed rank tests it can be said that free riding will be more when the incomes differ with gender than when incomes differ at random or when the incomes are equal (GD>RD=ND).
24
5.2 Results on fairness and equity concerns
The results descripted above in in the free riding subsection are, as foreseen, contradictive with the theory of the models of fairness and equity, since those models predict no differences in free riding between the RD and GD rounds.
In the RD round discrimination is added to a small extent, whereas in the GD round the degree of discrimination becomes heavier. I measured altruism, inequality aversion and reciprocity separately and per round in Question 6, 7 and 8 in the after-game survey (see Appendix AIII). To test if discrimination makes concerns about fairness and equity less important (Hypothesis 2), I look at changes in altruism, inequity aversion and reciprocity between the different rounds.
Graph 2: Descriptive statistics: altruism, inequity aversion and reciprocity by round
The descriptive statistics in Graph 2 display the difference in altruism, inequity aversion and reciprocity per round. For all models it seems there exists a small variation between the three rounds. A pattern in line with Hypothesis 2 can be observed: altruism, inequity aversion and reciprocity have a less pivotal role when there is more discrimination. To statistically support the observation that the rounds differ from each other, I again use a Friedman test. I look into the three concerns one by one. This time the dependent variable is the level of fairness and equity concerns (altruism, inequity aversion and reciprocity). Because these variables are measured on a Likert scale, they vary between 1 and 7, where 1 is no concerns about fairness and equity at all and 7 means full care about fairness and equity (see appendix AIII). The null hypothesis is that all concerns - altruism, inequity aversion and reciprocity - are the same in all rounds.
4 ,5 5 3 ,7 8 4,2 4 ,2 9 3 ,5 9 3 ,9 9 4 ,1 2 3 ,4 8 3,88
Altruism Inequity averion Reciprocity
Fairness and equity concerns per round
25
The results are summarized in Table 5. The Friedman tests cannot reject one of the null hypotheses, which means that based on these tests it cannot be said that someone’s fairness and equity concerns have changed during the experiment. Therefore, I zoomed in again with a Wilcoxon signed rank test to look at the changes between two of the rounds. These tests give some interesting significances. The p-values from all three concerns show significant differences between ND and GD, which indeed show that fairness and equity concerns will become less important in the GD round.
Table 5: P-values of statistical tests: differences in fairness and equity concerns between rounds
Altruism Inequity aversion Reciprocity
Friedman test: 0.12 0.59 0.22
Wilcoxon signed rank:
ND round – RD round 0.14 0.22 0.25
ND round – GD round 0.00*** 0.06** 0.01***
RD round – GD round 0.14 0.41 0.11
Note: Significance in indicated by *** for the 0.01 level and ** for the 0.05 level.
As with the first hypothesis, Hypothesis 2 also talks about the aspect of discrimination. This time the ND and GD round should be compared with each other. I haven’t measured the perceived level of unfairness in the ND round of the experiment, but this should be for all subjects at the lowest rate, since there are no differences made between them. Therefore, when looking at the difference between these two rounds a difference can be observed as well. In the ND round the unfairness should be 1out of 7 whereas in the GD round this is 3.38 on 7. A Wilcoxon sign-rank test gives a p-value of 0.00 which indicates that there is indeed a very significant difference between perceived levels of fairness between those two rounds.
Consequently, it can be said that hypothesis 2 ‘Discrimination vanishes out fairness and equity
concerns: fairness and equity concerns will become less important when the incomes differ with gender(non-random) than when the incomes differ at random than when the incomes are equal (GD<RD<ND)’ is partly true as well. It cannot be fully confirmed, since the Friedman test gives
no significant result. But due to the Wilcoxon signed rank tests it can be said that concerns will be more when the incomes differ with gender than when incomes are homogeneous (GD<ND).
26
5.3 Effect of treatments
When having a closer look at the results, it seems that men and women respond in a different way on being in the (dis)advantaged treatment. This leads to differences in free riding behavior and equity concerns between the genders.
Table 6: Descriptive statistics: mean (std. err.) of free riding per round by treatment
ND
RD
GD
Bad luck Good luck
Man 5.04 (0.38) 4.87 (0.63) 4.81 (0.50) 5.56 (0.39)
Woman 5.27 (0.35) 5.93 (0.43) 4.31 (0.55) 5.71 (0.33)
Table 6 zooms in on the descriptive results of free riding behavior by looking at the difference between the treatments. In all cases, the subjects with a disadvantage treatment (bad luck/woman) show on average more free riding than the subjects with an advantage treatment (good luck/man). That is in line with expectations, since the disutility that comes from a disadvantaged treatment is higher than the disutility that comes from a treatment with advantage (Fehr and Schmidt, 1999). To statistically support this observation I used a Wilcoxon rank sum (or Mann-Whitney) test to check if the amount of free riding significantly differ between the treatments. This test is a non-parametric statistical test can be used comparing two samples hat come from the same population, but with different treatments, to assess whether one of the samples differ from the other due to the treatment. First, I zoom in on the RD round. In the second part of this subdivision I look further into the GD round.
As can be observed in Table 6 men differ in their free riding behavior in de RD round, but only to a very small extent (0.07). Women change their free riding behavior much more (1.62) in this round. Hence, the Wilcoxon rank sum test gives no significant difference in free riding for men (n=81, p=0.89) and a significant difference of free riding for women on 5% level (n=56, p=0.02). This can be interpreted as: in the RD round women let their free riding behavior depend on their treatment, whereas men do not.
27
An explanation might be found when zooming in at the fairness and equity concerns of the subjects by looking at the differences per treatment. Overall, fairness and equity concerns differ between the good luck and bad luck treatment. In Table 7, 8 and 9 can be observed that these differences in treatment have another effect on men than on women.
Where altruism is more important for men with bad luck than men with good luck, this is the other way around for women.
In all cases, inequity aversion is seems to be more important for women than for men. But the treatments ‘good luck’ and ‘bad luck’ show similar patterns for men and women.
This is definitely not the case when looking at reciprocity. Men are very constant in their opinion about reciprocity: there are only very small differences between the rounds and the treatments. In contrast, women do change their opinion about reciprocity according to their treatment.
Table 7: Descriptive results: mean (std. err.) of altruism per round by treatment
ND
RD
GD
Bad luck Good luck
Man 4.47 (0.23) 4.79 (0.35) 3.98 (0.30) 4.20 (0.24)
Woman 4.68 (0.21) 3.83 (0.26) 4.62 (0.32) 4.02 (0.21)
Table 8: Descriptive results: mean (std. err.) of inequity aversion per round by treatment
ND
RD
GD
Bad luck Good luck
Man 3.36 (0.21) 3.55 (0.35) 3.05 (0.24) 3.01 (0.20)
Woman 4.39 (0.21) 4.27 (0.32) 3.77 (0.24) 4.16 (0.21)
Table 9: Descriptive results: mean (std. err.) of reciprocity per round by treatment
ND
RD
GD
Bad luck Good luck
Man 3.99 (0.23) 3.82 (0.36) 3.91 (0.32) 3.90 (0.23)
28
To see if these observations can be statistically proven I compare the treatments of good luck and bad luck with each other in a Wilcoxon rank sum test for both men and women. This gives the following p-values as outcome:
Table 10: P-values Wilcoxon rank sum test: difference between good luck and bad luck
Altruism Inequity aversion Reciprocity
Men 0.063* 0.33 0.88
Women 0.038** 0.24 0.03**
Note: Significance in indicated by ** for the 0.05 level and * for the 0.10 level.
These p-values show that women actually lower their altruism and reciprocity significantly by their treatment. This is not the case for men as the results for inequity aversion and reciprocity are far from significant. The change in altruism, significant at the 10% level, show an opposite effect between men and women towards this concern. Therefore, it seems that men and women value their preferences in different ways, which leads to different free riding behavior.
One reason of the difference between the sexes could be that men really perceive the RD round as fair while women do not. The findings of Question 1 of the after-game survey (see Appendix AIII) show that this could indeed be the case, since a Wilcoxon rank sum test gives a significant difference of the perceived fairness between men and women (n=137, p=0.01).
But it could also be the case that this result is caused by the gender differences in social preferences. Previous studies by Eckel and Grossman (1998) and Andreoni and Vesterlund (2001) have already shown that women exhibit greater degree of reciprocity than men do. Croson and Gneezy (2009) compared several studies and add to this that men choose for efficient allocations (altruism), while women are more inequality averse.
Now, I zoom in at the GD round. In this round the differences in free riding between the two treatments is 0.15, which is with a p-value of 0.92 far from significantly different from each other and that bother treatments (men and women) free ride more in the GD round than in the ND round. This is contradicting with the equity theory of Adams (1965), since this theory expects less free riding behavior from the subjects in the advantaged treatment than from the subjects in the disadvantaged treatment.
29
Also for this finding I am looking for an explanation by zooming in at the fairness and equity concerns. If looking at the last columns of Table 7, 8 and 9, it can be observed that the differences between the treatments are 0.18, 1.15 and 0.06 for respectively altruism, inequity aversion and reciprocity. To see if these differences are significant I use again a Wilcoxon rank sum test, which gives the following p-values:
Table 11: P-values Wilcoxon rank sum test: differences between men and women in the GD round
P-value
Altruism 0.56
Inequity aversion 0.00***
Reciprocity 0.92
Note: Significance in indicated by *** for the 0.01 level.
These values show that there is a major difference in inequity aversion between men and women, which could explain the free riding behavior in the GD round. Where women find it important that the earnings are equal for everyone, men do significantly care less about the equality. This is not very surprising, since it is assumed that 𝛼𝑖 ≥ 𝛽𝑖; the men are the advantaged group and will always
be better off and disutility that comes from a position of disadvantage is higher than the disutility that comes from a position of advantage (Fehr and Schmidt, 1999).
However, it is remarkable that the men react different to their advantage in this round than they did in de RD round. This seems to indicate that they have a different 𝛼𝑖 in the RD round than in the GD round. Searching for explanation lead me to Question 4 and 5 of the after-game survey (see Appendix AIII) about the connection they felt with their team member. Individuals may be more prone to cooperate when others in their group are similar to them, since this fosters a strong group identity (Kramer and Brewer, 1984; Kollock, 1998). The difference in connection men tend to feel between the RD and GD round indeed differ from each other, albeit only marginally significant (n=81, p=0.11).
Also the studies of Ben-Ner et al. (2004) and Houser and Schunk (2009) can be interesting in this case. These papers state that women’s decisions are more context dependent and sensitive to the gender of their counterpart while men’s are not. Since the subjects in this round know that their team member is from the opposite sex, this might be an explanation for the difference in free riding.
30
6. Discussion
The results of this study show that discrimination on non-random grounds has a negative effect on team performance. This can contribute to decisions companies have to take regarding to salaries to minimize free riding as much as possible. This study is, however, subject to some limitations and all conclusions must be drawn with great care.
First, the external validity of this study is questionable since the sample is not fully representative and randomly selected from the entire population. Although all subjects are old enough to understand the investment decision they had to make (Marwell and Ames, 1979) the age of 14.7 on average is not close to the average age of employees within a (normal) company. Besides, this experiment contained a subject pool consisting of students with a relatively high intelligence (gymnasium-only). This could have affected the choices they made. Chen et al. (2013) investigate the relationship between cognitive abilities and economic behavior. They report a positive relationship between GPA and selfish behavior and a negative relationship between GPA and social preferences. This would imply that the experiment is biased with more free riding and less fairness and equity concerns. It is also possible that people with a lower education, experience more discrimination, since jobs where no high level of education is needed are more male and female typed, which can cause more discrimination (Davison & Burke, 2000). Therefore, the experiment should also be done with a more diverse group of respondents, to check if adults and lower educated people react in the same way on (non-)random income heterogeneity before conclusions should be drawn.
Another item is the transparency of wages. In this experiment it is made very clear to the participants that there exists of a variety in the income between the two team members. Within a lot of organizations, information about each other’s wage is not publicly available. This should mean that the negative effects of social comparisons are of a lower extent than this experiment tells us.
A third limitation comes from an inaccuracy in the after-game survey. In Question 1 & 2 of this questionnaire (see Appendix AIII), I asked for perceived unfairness during the experiment. I used the outcomes to test the degree of perceived discrimination, while it was not a scale that was
31
developed to measure perception of discrimination. As a result it may be that not everyone has interpreted these questions in the same way and that these results are biased.
To study the influence of discrimination, I used ‘gender’ as non-random variable to measure discrimination. Since the last part of the Results section shows divergent effects between men and women, it may well be that there is some omitted variable included in this experiment. I cannot control for this since I only test the non-randomness of income heterogeneity by differences based on gender. It is questionable if other forms of discrimination, such as differences based on appearance, race or religion, would give the same outcomes. Therefore, the conclusions must be drawn carefully and cannot be easily generalized for all forms of discrimination. For further research, it would be very interesting to investigate if the results of this study are generalizable, and if not, to study in depth where the differences between gender discrimination and other forms of discrimination are coming from.
32
7. Conclusion
In this thesis I wanted to investigate the influence of discrimination on team performance due to measure the willingness of individuals to contribute to a team project under different degrees of income heterogeneity. Therefore, I set up a one shot linear PG game that could help me to analyze the differences between random (RD) and non-random (GD) income heterogeneities and income homogeneity (ND) within teams.
I predicted that discrimination would stimulate free riding behavior (Hypothesis 1) and would vanish out fairness and equity concerns (Hypothesis 2) and that these effects would be most visible under non-random (GD) income heterogeneity and least under no differences (ND) in income. The treatment with random income differences would be in between since the GD round is perceived as more unfair than the RD round. It can be said that Hypothesis 1 and 2 are both partially confirmed.
When there is more perceived discrimination, there is also more free riding behavior. Since the Friedman test gives only a marginally significant result, it should be rejected that GD>RD>ND. But due to the Wilcoxon signed rank tests it can be said that free riding will be more when the income differ with gender than when income differ at random or when the incomes are equal (GD>RD=ND).
It is also the case that there are less fairness and equity concerns if the perceived discrimination is higher. It cannot be fully confirmed that GD<RD<ND in one of the three fairness and equity concerns, since the Friedman tests cannot reject one of the null hypotheses. But due to the Wilcoxon signed rank tests it can be said that all concerns – altruism, inequity aversion and reciprocity - will be lower when the incomes differ with gender than when incomes are homogeneous (GD<ND).
In addition, it seems that men and women value a (dis)advantage in their treatment in a different way, which leads to different free riding behavior in the RD round. The free riding behavior in the GD round can also be explained by differences in social preferences between the genders: results show an opposite reaction on altruism and a major difference in inequity aversion between men and women.
33
Therefore, an answer to the research question ‘What is the difference between random and
non-random income heterogeneity within teams on free riding behavior?’ could be that non-non-random
differences in income lead presumably to more free riding behavior, because, in a certain amount, it vanishes out fairness and equity concerns. An important remark that should be made is that this most likely depends on the gender distribution within the team as well. For that reason it is important that incentives are not based on non-random or discriminatory grounds because this could have negative effects on team performance.
34
References
Adams, J. S. (1965). Inequity in social exchange. Advances in experimental social psychology, 2(267-299). Akerlof, G. A., & Yellen, J. L. (1990). The fair wage-effort hypothesis and unemployment. The Quarterly
Journal of Economics, 105(2), 255-283.
Alchian, A. A., & Demsetz, H. (1972). Production, information costs, and economic organization. The
American economic review, 62(5), 777-795.
Anderson, L. R., & Stafford, S. L. (2003). An experimental analysis of rent seeking under varying competitive conditions. Public Choice, 115(1-2), 199-216.
Andreoni, J., & Vesterlund, L. (2001). Which is the fair sex? Gender differences in altruism. Quarterly
Journal of Economics, 116(1), 293-312.
Andreoni, J. (2006). Philanthropy. Handbook of the economics of giving, altruism and reciprocity, 2, 1201-1269.
Becker, G. S. (1974). A Theory of Social Interactions. The Journal of Political Economy, 82(61),
1063-1093.
Ben-Ner, A., Putterman, L., Kong, F., & Magan, D. (2004). Reciprocity in a two-part dictator game. Journal
of Economic Behavior & Organization, 53(3), 333-352.
Bohm, P. (1972). Estimating demand for public goods: An experiment. European Economic Review, 3(2), 111-130.
Buckley, E., & Croson, R. (2006). Income and wealth heterogeneity in the voluntary provision of linear public goods. Journal of Public Economics, 90(4), 935-955.
Chan, K. S., Mestelman, S., Moir, R., & Muller, R. A. (1996). The voluntary provision of public goods under varying income distributions. Canadian Journal of Economics, 29(1), 54-69.
Chan, K. S., Mestelman, S., Moir, R., & Muller, R. A. (1999). Heterogeneity and the voluntary provision
of public goods. Experimental Economics, 2(1), 5-30.
Charness, G., & Rabin, M. (2002). Understanding social preferences with simple tests. Quarterly journal
of Economics, 117(3), 817-869.
Charness, G. (2004). Attribution and reciprocity in an experimental labor market. Journal of Labor
Economics, 22(3), 665-688.
Cherry, T. L., Kroll, S., & Shogren, J. F. (2005). The impact of endowment heterogeneity and origin on public good contributions: evidence from the lab. Journal of Economic Behavior & Organization, 57(3), 357-365.
Cohn, A., Fehr, E., Herrmann, B., & Schneider, F. (2011). Social comparison in the workplace: Evidence from a field experiment (No. 7). Working Paper Series, Department of Economics, University of Zurich. Croson, R., & Gächter, S. (2010). The science of experimental economics. Journal of Economic Behavior
