Fairness in a Game Setting
BACHELOR THESIS THESIS WITHIN: Economics NUMBER OF CREDITS: 15 Credits
PROGRAMME OF STUDY: International Economics AUTHOR: Mariam Soumi and Viktor Gustafsson JÖNKÖPING May 2020
i
Bachelor Thesis in Economics
Title: The Effects of Gender and Culture on Fairness in the Ultimatum Game Authors: Mariam Soumi and Viktor Gustafsson
Tutor: Anna Nordén Date: 2020-05-18
Key terms: Ultimatum game, fairness, gender differences, individualism, collectivism.
Abstract
Fairness is an important topic that has captured the interest of researchers in many fields. Looking at behavioural and experimental economics, various methods have been used to shed the light on fairness. One of the most recognizable ways is through the use of the ultimatum game. In this paper, we aimed to look at fairness considerations by utilizing this game, while also highlighting gender and culture as factors of importance. Two models, specifically Rabin’s reciprocity model and Fehr and Schmidt’s inequity aversion model, in addition to various research papers on the topics of gender and culture, were used as the cornerstones for this paper. The experimental design was a replication of the ultimatum game, which was carried out online. Results showed that there were noticeable effects from both gender and culture on the behaviour of the participants in the game. However, the results were not statistically significant enough to be considered as determining factors in regard to why individuals behave in a specific manner in the game. Since research from the field has reached mixed conclusions, we can say that making inferences about human behaviour is harder than one might think.
ii
Table of Contents
1.
Introduction ...1
2.
Theoretical Framework and Literature Review ... 4
2.1. Theories of Fairness and Social preferences ... 4
2.1.1. Rabin’s reciprocity model ... 4
2.1.2. Fehr and Schmidt’s inequity aversion model ... 5
2.2. Research on the ultimatum game ... 6
2.2.1. Gender ... 6
2.2.2. Culture... 8
3.
Hypotheses... 10
4.
Experiment Design and Procedures ... 12
4.1. Experimental Setup ... 12
4.2. Econometric models... 13
4.3. Experimental limitations ... 15
5.
Results ... 16
5.1. Descriptive statistics - participant characteristics ... 16
5.2. Ultimatum game results ... 17
5.3. Regression analysis ... 20 5.3.1. Proposer group... 20 5.3.2. Responder group ... 21
6.
Discussion ... 22
7.
Conclusion ... 26
8.
References ... 27
iii
Tables
Table 1: Behavior of proposers ... 11
Table 2: Behavior of responders ... 11
Table 3: Experimental treatments ... 12
Table 4: First set of hypotheses – proposer behavior ... 14
Table 5: Second set of hypotheses – responder behavior ... 15
Table 6: Descriptive statistics – Participants’ characteristics ... 17
Table 7: Ultimatum game - average offers ... 18
Table 8: Ultimatum game - general averages ... 18
Table 9: Ultimatum game - rejection rates ... 19
Table 10: OLS Regression for the first set of hypotheses ... 20
Table 11: Logit regression for the second set of hypotheses ... 21
Appendices
Appendix A - Scripts ... 31Appendix B – Proposer form ... 32
Appendix C – Responder form ... 34
Appendix D – Consent form ... 36
1
1. Introduction
Standard economic theory has long assumed that individuals are rational utility or profit maximizers. Yet, such an assumption does not necessarily always hold. We tend to observe individuals who forgo their own self-interest and instead choose fairness over rationality. By looking at the evolutionary side of things, natural selection has been in favor of fairness as an outcome (Rand et al., 2013). Higher paying strategies were typically in line with fairer behaviors, which made the act of being fair highly coveted in early human interactions. Undoubtedly, this meant that equitable strategies were more likely to be imitated allowing them to stand the test of time. According to Fehr and Schmidt (1999, p. 817) “the economic environment determines whether the fair type or the selfish type dominate equilibrium behavior”. What Fehr and Schmidt were hinting at is that there are certain factors that come to surface and determine when fairness actually prevails. So, whether Rand et. al (2013, p. 2585) were true when they said: “in this unfair world, myopic self-interest is vanquished whereas fairness triumphs” or not, that is a matter left for discussion.
Gender, as a distinguishing factor, is of great importance since it is one of the demographic variables that have been studied by many researchers. Gender differences are evident in a vast majority of studies from economical decision making (Eckel & Grossman, 2008a) to cooperation (Molina et al., 2013). Based on literature, it is found that there are differences between men and women especially in terms of risk, social preferences, as well as general preferences for competition (Croson & Gneezy, 2009). In addition, Croson and Gneezy’s (2009) research shows that women are more risk averse than men, due to several suggested reasons such as emotions, overconfidence, and framing effects. Another paper written by Chen et al. (2013) suggests that females’ biological state, particularly while on contraceptives, leads to noticeable changes in their bidding behavior for example. The above-mentioned collection of research is a small sample that simply shows how gender differences are evident in many ways. Such differences could be associated with various factors that ultimately affect the behavior of both genders. But in the context of fairness, especially from the viewpoint of both behavioral as well as experimental economics, gender has only had the attention of a few researchers. In contrast to gender, culture as a factor of interest has held the attention of many researchers who have incorporated the topic in a vast collection of studies. Numerous research papers have been aimed towards studying the behaviors of individuals from different cultural backgrounds and the notion that culture can affect behavior has been long established by researchers (Camerer, 2003). For example, when investigating whether there were differences between levels of cooperation among individuals
2
from different cultural backgrounds1, Gächter et al. (2010) were able to find substantial and significant
differences in the behavior of the participants. Furthermore, in various experimental game setups, differences on the national and ethnic levels have been found in different samples of participants (Chuah et al., 2009). This goes to show that research on culture has been able to show tangible results. However, what cultural research in the field of behavioral economics has lacked, is an accurate classification of the participants’ cultural background that is based on established findings. Cultural classification systems, such as the cultural dimensions created by Hofstede (1984) and the global cultural maps by Inglehart and Baker (2000), have been widely used for the purpose of highlighting the different characteristics observed in various populations. But such systems have been used by a minority of behavioral researchers when trying to account for culture as a factor of interest.
Thus, this research paper aims at filling a gap that has been persistently seen in past papers. To be able to look at fairness considerations in decision-making, we will be emphasizing gender as a factor while also including appropriate cultural classifications. With this focus in mind, we intend on finding the answer to the following research question: How does gender affect fairness in the context of the ultimatum game? And what are the effects of culture in that same context?
Taking inspiration from two research papers, one by Sara Solnick (2001) and the other being by Eckel and Grossman (2001), we will be studying gender effects through the use of the ultimatum game. By observing the behaviors and choices of participants, the ultimatum game in particular allows us to check whether there are any noticeable differences in the levels of fairness of both females and males. In addition, we will be investigating the cultural factor by relying on the Individualism-Collectivism (I/C) cultural dimension from the classification system established by Hofstede. Since the use of the I/C dimension is lacking in the field, we will be incorporating it in our research as a means to split participants into categories in order to check for the presence of any behavioral differences. Our intentions with this paper are not to find groundbreaking evidence nor are they to emphasize any differences between genders or cultures, instead we intend on adding our findings to the existing pool of research in the hopes of helping increase our understanding of overall human interactions.
One of the main limitations of our replication is that we are not able to utilize the possibility of performing our experiment at Jönköping University. Due to the outbreak of COVID-19, we instead resorted to using an online social media platform (Instagram) to gather our participants. This in turn restricted the amount of people that could offer us help while gathering our data. In addition, due to our limited experience in performing experiments, another limitation that we face is the lack of
1 The participants were grouped into five cultural categories namely: English speaking, Protestant Europe, Orthodox, Southern Europe, Arabic speaking and Confucian (Gächter et al., 2010)
3
reliability of our approach. Since this experiment is not as well-planned as some of the previous work done in the field, our execution process cannot be compared to the extensive lab experiments that have been previously performed. Nonetheless, our goal with this paper is to expand the discussion on gender and culture hoping to spark interest in further research to be performed in the future. The remainder of this paper will be divided as follows. In Section 2, we will discuss the theoretical foundation of this research paper by looking at two models, while also diving deeper into the two pillars of the paper which are: gender and culture. Section 3 will have a rundown of our hypotheses. Section 4 contains the outline of the experimental design, econometric models as well as the limitations of our experimental setup. We then present our results and findings in Section 5. The last two sections of the paper, Section 6 and 7, will be a discussion and a conclusive summary of our findings, respectively.
4
2. Theoretical Framework and Literature Review
For us to understand how fairness prevails as an outcome, whether in games or other circumstances, we must look at an aspect besides individuals’ material self-interest, in particular social preferences. According to Fehr and Fischbacher (2002, p. C2) “a person exhibits social preferences if the person not only cares about the material resources allocated to her but also cares about the material resources allocated to relevant reference agents”. Since we tend to find people who are not only motivated by personal self-interest but care about obtaining fair outcomes too, this shows that social preferences are also involved in the decision-making process. Such preferences can be divided into different types including reciprocity, inequity aversion, altruism, and spiteful preferences. Even though the different types of social preferences all go hand in hand and are a gateway to understanding individuals’ behavior, for the purpose of this paper, only two types will be highlighted. The first type discussed is reciprocity which is expressed by Rabin’s reciprocity model and the second is inequity aversion explored by Fehr and Schmidt’s inequity aversion model.
From there on, we present research on the topic of gender in the ultimatum game to expand our understanding of the game setting we have at hand. In addition to research on gender, we will also highlight research that looks at culture, in a broad context as well as in the ultimatum game, because it plays a role in understanding why certain behaviors are observed.
2.1. Theories of Fairness and Social preferences
2.1.1. Rabin’s reciprocity model
In his 1993 paper, Rabin was able to observe that an individual’s behavior is often the result of a reaction to other individuals’ intentions. Such a reaction usually comes in the form of reciprocity. Individuals match received kindness with kind behavior and respond to hostile actions with hurtful demeanor (Rabin, 1993; Fehr & Fischbacher, 2002). In his model, utilities are not only driven by future materialistic benefit, but instead they also rely on each player’s beliefs in a given game. Such beliefs are translated into “kindness functions” which measure how kind the players are towards each other. These functions are then incorporated into defining the players’ utility functions which are dependent upon: “the individual’s strategy, their belief about the other player’s strategy choice, and their belief about the other individual’s belief about their strategy” (Rabin, 1993 p. 1286). When behavior and beliefs get combined, a ‘fairness equilibrium’ is achieved (Rabin, 1993). Each individual’s actions depend on understanding the intentions of those that they interact with in order to choose what type
5
of behavior they should reciprocate. Thus, a fair division in an ultimatum game would be expected to prevail if players believe that there are fair intentions at hand, otherwise an unfair division would take place when players suspect dealing with unjust behavior.
A shortcoming of this model is the possibility of obtaining more than one equilibrium. Such a scenario arises since it is not possible to identify whether two individuals will behave ‘nicely’ or ‘rudely’ given the circumstances that they are presented with (Fehr & Schmidt, 2001). In addition, the concern with fairness becomes eroded the higher payoff is (Rabin, 1993). Nonetheless, the model showcases how intentions can lead to the occurrence of certain behaviors like reciprocity, especially when looking at the fairness of offers in an ultimatum game.
2.1.2. Fehr and Schmidt’s inequity aversion model
Fehr and Schmidt (1999) model fairness as a form of self-centered inequity aversion because their belief is that besides the presence of purely self-centered individuals, there is a fraction of the population that is motivated by fairness consideration. The researchers assume that individuals behave in a more sympathetic manner when the other individual’s material payoff is below fair standards. But when the payoff is above the point of fairness, jealousy is awoken and as a result, people would rather achieve a more equitable outcome even if it means incurring a price while doing so (Fehr & Schmidt, 1999; 2001). In the model, the researchers use basic utility functions which also capture individuals’ disutilities when faced with inequality (Fehr & Schmidt, 2001). Another assumption present in this model, is the heterogeneity of individuals because heterogeneous preferences are capable of revealing a new dimension in the economic environment (Fehr & Schmidt, 1999). Therefore, the model shows that when people with different preferences for fairness are mixed, one of two possible outcomes is expected to hold: 1- the selfish players are capable of inducing those who are inequity-averse to behave selfishly or 2- the players who are inequity-averse are able to exert enough influence on the selfish players leading to fairer outcomes.
However, the model can be seen as a model on the weaker side due to its simplicity when compared to models about reciprocity for example (Fehr & Fischbacher, 2002). In addition, the similarity between reciprocity and inequity-aversion, in terms of individuals’ desire to reduce others’ payoffs in some form, is backed by more evidence in favor of reciprocity than inequity-aversion (Falk, Fehr & Fischbacher, 2008). Regardless of the fact that there is a fine line between the two, the inequity-aversion model is still helpful because it shows that the diverse nature of people can be an associated explanation when looking at how individuals behave. This model is particularly helpful since it could help showcase why offers for example, get accepted or rejected in a given ultimatum game.
6
As presented above, the two models complete each other in the sense that both lead to a fairer outcome than the initial starting point. We believe that throughout the process of investigating our research question, the behaviors that we will observe will be a combination of the expected behaviors from both models. In other words, our belief is that the proposers will behave along the lines of the expectations from the inequity aversion model, where they will give the less well-off responder a somewhat equal division. As for the responders, they will behave along the lines of the reciprocity model. This would be the case, since responders are expected to accept the offer due to their assumption that the proposers’ intentions are in line with having a kindness equilibrium prevail. This belief of ours will be accentuated when we present and investigate our hypotheses in section 3.
2.2. Research on the ultimatum game
Before diving into research regarding the ultimatum game, it is important to understand how the game itself is played, to allow for a better understanding from a contextual point of view. The classic ultimatum game consists of two players set to interact with each other in a simple manner that mimics a typical bargaining situation. Each participant gets randomly assigned the role of proposer or responder. Being the proposer means that the player is endowed with a certain amount of a resource, which is typically money, that they have to split between themselves and the second player - the responder. The latter has the power to decide whether they accept or reject the suggested offer. If the offer is accepted by the responder, then the agreed division takes place. But if it is rejected, both players receive nothing. The typical assumption in the game is the presence of standard income-maximizing behavior, in which proposers are expected to offer the least amount possible. As for the responders, they will accept such an unfair division because it provides them with a positive gain making them better off than before. Now that the rules and expectations of the game are clear, we can move on to presenting research from the field of experimental economics.
2.2.1. Gender
When looking at gender in the field of economics, Catherine Eckel and Philip Grossman can be attributed to most of what we know about how participants, of both genders, behave in experimental circumstances. Their work with gender differences can be found in various setups including: the punishment (1996), dictator (1998) and ultimatum game (2001). But for the purpose of this research paper, their most relevant work, is the 2001 paper where their findings were strictly from an ultimatum game in a lab experiment setup.
7
In their paper, Eckel and Grossman (2001) conducted the first experiment that explicitly tested for gender differences by the use of the ultimatum game. Eckel and Grossman had a unique game design that had not been utilized before in which they placed participants in groups of eight, where the two standard roles of the game (proposer and responder) were split among the participants at random. Instead of hiding the identity of the participants, which is a standard procedure in the ultimatum game, they allowed them to sit in groups while playing. What each participant knew, was only the gender of the player that they were paired with from that group but not the actual identity of the person they were playing against. By mixing the genders of the pairs in each round, they were able to observe interesting behaviors. In general, women gave offers that were closer to half of the stake size and were less generous specifically when paired with other females (Eckel & Grossman, 2001). Looking at yet another paper that also studies the behavior of genders in the ultimatum game, Solnick in her 2001 paper was able to find behaviors that could also be traced in Eckel and Grossman (2001). In both papers, an observed feature was that the average offer presented to females was less than that made towards males (Croson & Gneezy, 2009; McGee & Constantinides, 2013).
Although both aforementioned papers share important similarities, the differences in the designs are said to lead to striking differences in the observed results. One of the most important differences is that in rejection rates. Solnick (2001) reported that offers made by women were faced with rejection more often than those made by men while in Eckel and Grossman’s (2001) paper, they found the complete opposite. The dissimilarities in results, have been attributed to the different design approaches that each paper utilized. On one hand, Solnick used a one-shot ultimatum game following the “strategy method”. In the strategy method, proposers place their offers and the responders place their minimum acceptable offers (MAO’s), which is the least amount the responder is willing to accept, simultaneously. The offers and the MAO’s are then matched by the experimenters to check whether a deal would in fact take place. On the other hand, Eckel and Grossman used eight rounds of the “game method”, which is the conventional form of the game, that was presented at the start of section 2.2. Such design differences were said to lead to differences in the perceived levels of risk associated with the two roles in each version of the game (Eckel & Grossman, 2008a).
But to the extent that the design of the game does in fact lead to such differences has been questioned by McGee and Constantinides (2013). They investigated whether playing repeated games leads to the presence of a learning effect, which could potentially be associated with gender differences in the results. They were able to conclude that even though they could detect some differences in the behavior of both genders throughout the first rounds, the difference was not significant enough to be considered and instead ruled it out as ‘noise’. In addition, Oxoby and McLeish (2004) performed a
8
similar study where they employed both the one-shot version of the ultimatum game and the strategy vector method. They were also unable to find any significant or meaningful differences which further discredits the idea that the experimental design used leads to general alterations in the participants’ behaviors.
Despite all the behaviors traced in the Solnick as well as the Grossman and Eckel paper, both research papers reached similar conclusions that there were actually no significant differences between the genders when making an offer (Solnick, 2001; Eckel & Grossman, 2001). What the two papers concluded was that even though they were able to pick up some variation in the behavior of their participants, the differences between the two genders were small enough to be ruled as insignificant. This could hint that behavioral differences observed are not as substantial as one would think. However, having such conclusions formulated leaves the floor open for research to continue to question whether there are in fact differences between the behavior of females and males. For this particular reason, we intend on replicating and incorporating gender into our paper, allowing us to explore the question regarding the existence of differences between the behavior of both genders.
2.2.2. Culture
One of the most famous pieces of research on behavior in the ultimatum game is the paper by Henrich et al. (2001) where they investigate the presence of the Homo Economicus model among 15 small-scale societies2. The studied societies were spread across twelve countries on five continents and the
participants were recruited from societies with various economic and cultural conditions. The findings show that there is no evidence supporting the presence of an individual whose only interest is to maximize their personal material payoff thus discrediting the selfish Homo Economicus model (Henrich et al., 2001). The findings also stressed the fact that there is visible influence from the different cultures of the participants specifically when some participants rejected hyperfair offers3 due
to what their culture dictates while others did not reject offers that might have been perceived as unfair4 also due to established norms (Henrich et al., 2001). On the contrary, we see that findings from
yet another paper by Roth et al. (1991) showed the opposite in terms of the nature of proposed offers. The participants were students from four different cities namely: Jerusalem, Ljubljana, Pittsburgh, and Tokyo, hence representing four different cultures. In the paper, the researchers were able to find patterns of consistent income-maximization in line with the Homo Economicus model in their
2 The sample of small-scale societies included: three foraging societies, six slash-and-burn horticultures, four nomadic herding groups, and three sedentary agriculturalist societies (Henrich et al., 2001) 3 Like the Ache in Paraguay and Lamelara in Indonesia that rejected offers higher than 50% of the stake 4 Like the Au and Gnau in Papua New Guinea accepting offers below 40% of the stake size
9
respective pool of participants. But in the conclusion, they infer that the observable variations in behavior could in fact be attributed to the interference of culture (Roth et al., 1991; Camerer, 2003). Another classical way to look at the effects of culture, not necessarily in the context of the ultimatum game but by looking at life in general, is through Geert Hofstede’s (1984) cultural dimensions. Research regarding the five dimensions stretches across many fields, but one dimension in particular is of importance to this paper, which is the dimension of Individualism-Collectivism. Individualism according to Hofstede (1984, p. 148) refers to: “the relationship between the individual and the collectivity which prevails in a given society”. Research has been able to show that individuals from collectivistic cultures are more cooperative and care about the collective well-being of their community when compared to those from individualistic cultures who instead prioritize their individual well-being (Triandis, 2001). In addition, individuals who grow up in individualistic cultures are brought up with the spirit of competition, which tends to be missing in collectivistic cultures because ideals of fairness among members of the community are fostered instead (Rochat et al., 2009). This dimension is of great importance because it could help pinpoint some of the cultural differences that are found in the field of experimental economics and particularly within the ultimatum game.
A research paper that checks for such cultural differences, through the I/C dimension, is one by Chuah et al. (2009). In the research paper, they looked at the behavior of Malaysian and British participants in the ultimatum game. They stated that: “Individualist cultures emphasize individual interests and freedom at the expense of collective ones” (Chuah et al., 2009, p. 741). The researchers attributed the individualistic nature of British participants to the lower observed offers in the game when compared to Malaysian participants who were considered more collectivistic and thus offered more. Even though in their paper they were able to identify other factors that add to the cultural argument, it is consistently observed that culture does interfere with the outcomes of the game since participants have shown parts of their own cultures through various behaviors while playing. With this in mind, we were also interested in finding whether peoples’ cultural background would affect their behavior in our setup. Therefore, we followed in the footsteps of Chuah et al. and incorporated the I/C cultural dimension into our own research topic to test for the presence of any cultural differences.
10
3. Hypotheses
In order to investigate the presence of any effects from both gender and culture on fairness, a series of hypotheses, based on the theoretical and empirical foundations, is designed. Starting with the ultimatum game itself, it has two basic roles which are: proposers and responders. Since we have this natural split in the game, two sets of hypotheses are created where one tests the behavior of the proposer group while the second tests that of responders. The behavior of the proposers is observed in terms of the offered amount where an offer close to 40% of the total endowment is classified as ‘fair’ based on reviewed literature5. As for the responder group, their observed behavior is dependent
upon rejection rates to the proposed offers, regardless of whether the offer itself is deemed as ‘fair’ or ‘unfair’. The second classification used in creating the hypotheses, comes in terms of the different treatments that are used in the experiment. We have a total of four treatments prepared for our participants: Female-Female, Male-Male, Female-Male and Male-Female. The makeup of the treatments is explained in section 4, alongside the rest of the experimental setup. A plan for having an additional control group was also in place but could not be executed due to a shortage in participants. In addition, the cultural factor is involved in the hypotheses-creation process by yet another natural split between the participants. This split depends on whether the participants’ cultural background is collectivistic or individualistic. The three classifications are used to help us create a total of six hypotheses which are presented in Tables 1 and 2. The reason why we have such a great number of hypotheses is tied to the fact that each hypothesis allows for a much more focused view of what is being observed rather than trying to create a hypothesis that includes more than one factor at a time. The three hypotheses in the first set have an alternative hypothesis stating that there are no differences in the offers made. Thus, the first set, which tests the behavior of proposers, is as follows:
Hypothesis 1:
H0: There is no difference in the amount offered by male and female proposers when paired with a
responder from the same gender Hypothesis 2:
H0: There is no difference in the amount offered by male and female proposers when paired with a
responder from the opposite gender
11 Hypothesis 3:
H0: There is no difference in the amount offered by proposers from an individualistic and
collectivistic culture
Table 1: Behavior of proposers
BEHAVIOR OF PROPOSERS
Hypothesis 1 Hypothesis 2 Hypothesis 3
COMMON FACTOR measuring amount offered by proposers to respective responders
SPECIALIZED FACTOR gender: same (FF, MM) gender: opposite (FM, MF) culture
H0 no difference no difference no difference
H1 difference is present difference is present difference is present IMPORTANT DUMMIES
IN REGRESSION dummy 2: MM dummy 1: FF dummy 1: FM dummy 2: MF P-individualistic
As for the second set, the three hypotheses have a shared alternative hypothesis stating that there are no differences between rejection rates. This makes the second set, which tests the responders’ behavior, the following:
Hypothesis 4:
H0: There is no difference in the rejection rate between male and female responders when paired
with a proposer from the same gender Hypothesis 5:
H0: There is no difference in the rejection rate between male and female responders when paired
with a proposer from the opposite gender Hypothesis 6:
H0: There is no difference in the rejection rate between responders from an individualistic and
collectivistic culture
Table 2: Behavior of responders
BEHAVIOR OF RESPONDERS
Hypothesis 4 Hypothesis 5 Hypothesis 6
COMMON FACTOR measuring rejection rates of responders to offers from proposers
SPECIALIZED FACTOR gender: same (FF, MM) gender: opposite (FM, MF) culture
H0 no difference no difference no difference
H1 difference is present difference is present difference is present IMPORTANT DUMMIES
12
4. Experiment Design and Procedures
4.1. Experimental Setup
The design of our experiment is inspired from three papers that could be thought of as the foundation for this experimental setup. For the gender aspect of the design, the inspiration comes from Solnick’s 2001 experiment as well as Eckel and Grossman’s 2001 setup in which both papers looked at gender differences in the ultimatum game. In her experiment, Solnick divided the participants into five experimental categories which are: Female-Female, Male-Male, Female-Male, Male-Female, and a randomized control group. The categories denote the genders of the proposers and responders interacting together, in which a Female-Male pairing for example represents a female proposer playing against a male responder respectively. This denotation system applies to the rest of the categories mentioned. The same system of categorization is applied in our own setup, showed in Table 3, with the exception of the control group since we do not have such a group in our sample. Another similarity shared with Solnick’s paper is the number of rounds used. We use a one-shot ultimatum game in which participants essentially play for one round only. A feature of difference however is the game design chosen. We use the “game method” which is inspired by Grossman and Eckel’s approach where participants play sequentially. In comparison to Solnick, she used the “strategy method” which meant that participants had to play simultaneously. The reason why the “game method” is decided upon, is solely based on our setup which is a field study rather than a typical laboratory experiment. Having a field study, makes it more difficult to play simultaneously which results in having to use sequential play instead.
Table 3: Experimental treatments
PROPOSER
RESPONDER
Female Male
Female Female-Female (FF) Male-Female (MF) Male Female-Male (FM) Male-Male (MM)
Due to the outbreak of Covid-19, our participants are recruited with the help of social media instead of more traditional methods such as recruiting university students through posters on campus. Posts on our ‘Instagram story feed’ are posted at the same time and individuals that interact with the post, are then counted as participants. In contrast to how experimental games are typically played out, our participants are not paid, due to financial limitations, and do not have any monetary gains from their participation. To perform the game itself, the platform Google Forms is utilized. Six separate forms are created, in which every two forms correspond to a specific currency namely: Swedish crowns, US
13
dollar and Euro, to mimic the currencies used by our participants. The two forms within each currency group are split into a form directed towards proposers, found in Appendix B (page 32) and another towards responders, found in Appendix C (page 34). The form explains the rules of the game while also exposing only the gender of the person they are matched with.
The first number of participants are given the role of proposer to ensure that subsequent participants, those given the role of responders, would in fact have an offer to respond to. From there on, participants are given their respective role at random accompanied by the relevant instructions form. Proposers are informed through the instructions that they are endowed with 500 sek, or the equivalent6, while responders are informed of their offer before they start playing. The participants
are each also given an ID to preserve their identity and allow for a less biased experience. After the game, a section in each form is dedicated to some questions regarding the characteristics of each individual. Participants are asked to provide us with their: gender, age and nationality. The third characteristic, nationality, would act as a gateway to allow for the incorporation of the second factor of importance, which is culture. Chuah et al. (2009), which is the third and last pillar in the process of creating the experimental setup, integrated the I/C cultural dimension into their research by having participants from both Britain and Malaysia. Taking inspiration from their paper and placing it into our experiment, asking for the participants’ nationality allows us to classify them as either an individual from a collectivistic or individualistic background. Afterwards, consent forms to use personal data, which are found in Appendix D (page 36), are sent out to the participants to ensure that GDPR guidelines are met to the best of our abilities.
4.2. Econometric models
For us to be able to find answers to the proposed hypotheses by the means of regression analysis, two sets of econometric models are created, where one corresponds to testing the behavior of the proposers and the second tests that of responders. The models highlight some key factors that will allow us to either accept or reject our null hypotheses. The sets of models are as follows:
First set: proposer behavior
Model testing for gender differences: 𝑂𝑓𝑓𝑒𝑟 = 𝛼0+ 𝛼1𝐹𝐹 + 𝛼2𝑀𝑀 + 𝛼3𝐹𝑀 + 𝛼4𝑀𝐹 + 𝑒
Model testing for cultural differences: 𝑂𝑓𝑓𝑒𝑟 = 𝛾0 + 𝛾1𝑃𝑖𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙𝑖𝑠𝑡𝑖𝑐+ 𝑒
6 To make the endowment as equal as possible in all three currencies, it is decided upon that the following conversion system is applied: 500 sek= 50$ = 50€
14
The first set focuses on offers coming from the proposers in the sample which is indicated by our dependent variable Offer. Dummy variables are used to help distinguish between when the pairings are of the same gender, opposite gender, or if a proposer is from a specific cultural background. For same-gender pairings, the dummy variable Female-Female (FF) takes on the value of 1 when two females are paired together, and Male-Male (MM) takes on the value of 1 if the pairing is between two males. As for the opposite-gender pairings, we also have two dummy variables where Female-Male (FM) takes on the value of 1 when the proposer is a female and the responder is a male, and Male-Female (MF) takes on the value of 1 when the proposer is a male against a female responder. For the same-gender treatment, the dummy MM will be excluded where it will act as the benchmark when comparing offers from same gender pairings. The same application is used for opposite gender pairings, where MF will be excluded and used as the benchmark for the regression. The final dummy variable accounts for the culture of the proposer in model 3, where Pindividualistic is 1 if the proposer is
from an individualistic background and 0 if from a collectivistic one. All models have constants and error terms. A summary of the regression models, that are run to test the behavior of proposers, is found in Table 4.
Table 4: First set of hypotheses – proposer behavior
HYPOTHESIS EXCLUDED TREATMENT OLS MODEL
HYPOTHESIS 1: SAME GENDER Male-Male Model 1: 𝑂𝑓𝑓𝑒𝑟 = 𝛼0+ 𝛼1𝐹𝐹 + 𝛼3𝐹𝑀 + 𝛼4𝑀𝐹 + 𝑒
HYPOTHESIS 2: OPPOSITE GENDER Male-Female Model 2: 𝑂𝑓𝑓𝑒𝑟 = 𝛼0+ 𝛼1𝐹𝐹 + 𝛼2𝑀𝑀 + 𝛼3𝐹𝑀 + 𝑒
HYPOTHESIS 3: CULTURE P-Collectivistic Model 3: 𝑂𝑓𝑓𝑒𝑟 = 𝛾0+ 𝛾1𝑃𝑖𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙𝑖𝑠𝑡𝑖𝑐+ 𝑒
Second set: responder behavior
Model testing for gender differences: 𝑅𝑒𝑗𝑒𝑐𝑡𝑖𝑜𝑛 = 𝛿0+ 𝛿1𝐹𝐹 + 𝛿2𝑀𝑀 + 𝛿3𝐹𝑀 + 𝛿4𝑀𝐹 + 𝑒 Model testing for cultural differences: 𝑅𝑒𝑗𝑒𝑐𝑡𝑖𝑜𝑛 = 𝜆0+ 𝜆1𝑅𝑖𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙𝑖𝑠𝑡𝑖𝑐+ 𝑒
This second set is focused on observing the behavior of the responders. The dependent variable of interest is Rejection which highlights rejections from responders to offers proposed by the respective proposers. Yet again, dummy variables are utilized to allow us to test for gender and cultural effects. The first four gender-related dummy variables are exactly the same as the ones presented in the previous set and are used in the same exact manner. However, in the model testing for cultural effects, we have a new dummy. Rindividualistic takes on the value of 1 if the responder is from an individualistic
background and 0 if otherwise i.e. from a collectivistic one. The models in this set also have constants and error terms. The only difference, between the two sets of models, lies in the fact that the second set will be run by using a Logit regression since the dependent variable is binary. The models used to test for responders’ behaviors are summarized in Table 5.
15 Table 5: Second set of hypotheses – responder behavior
HYPOTHESIS EXCLUDED TREATMENT LOGIT MODEL
HYPOTHESIS 4: SAME GENDER Male-Male Model 4: 𝑅𝑒𝑗𝑒𝑐𝑡𝑖𝑜𝑛 = 𝛿0+ 𝛿1𝐹𝐹 + 𝛿3𝐹𝑀 + 𝛿4𝑀𝐹 + 𝑒
HYPOTHESIS 5: OPPOSITE GENDER Male-Female Model 5: 𝑅𝑒𝑗𝑒𝑐𝑡𝑖𝑜𝑛 = 𝛿0+ 𝛿1𝐹𝐹 + 𝛿2𝑀𝑀 + 𝛿3𝐹𝑀 + 𝑒
HYPOTHESIS 6: CULTURE R-collectivistic Model 6: 𝑅𝑒𝑗𝑒𝑐𝑡𝑖𝑜𝑛 = 𝜆0+ 𝜆1𝑅𝑖𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙𝑖𝑠𝑡𝑖𝑐+ 𝑒
4.3. Experimental limitations
Within our experimental setup, we find two major limitations. The first limitation is what is termed as hypothetical bias. According to Muñoz-García & Li (2018, p. 207) this bias is defined as: “subjects reporting a higher willingness-to-pay in the ‘hypothetical’ treatment of the experiment (when they do not have to pay for the good) than in the ‘real’ treatment (in which subjects may actually pay for the good)”. A research paper by Ajzen et al. (2004) shows that the bias occurs due to the beliefs and attitudes of the participants depending on the context that they are presented with. In our experiment, and even though on the initial page of the instructions in every form sent out, we have a short text that urges students to behave as though the experiment mimics reality, the extent to which individuals place themselves in such a mindset remains unknown. The implications of this bias could eventually lead to somewhat ‘inflated’ results in the ultimatum game, where offers and acceptance rates could be too high. But in the case of our experiment, the bias cannot be avoided due to two reasons. The first reason is the lack of financial resources, which would otherwise allow for a more realistic feel to the experiment, and the second is due to the nature of the research in which the experiment is conducted online rather than in a controlled setting.
In addition, the phrasing of the sentences and the words used while explaining the game might lead to another issue which is the anchoring effect. The effect arises when individuals unconsciously place too much focus on the initial piece of information they receive, without giving as much attention to any subsequent information (Xifen & Yong, 2018). For the case of our online game, the effect might be present in the sense that the participants put focus on the monetary sum more than the gender of the opponent or vice versa. Furthermore, research shows that various factors such as ambiguity and lack of familiarity on the decision-maker’s part, could lead to elevated levels of anchoring to occur (Galinsky & Mussweiler, 2001). In the case of our participants, the concept of the game itself might be completely new which could lead to uncertainty on how to behave. Regardless of the extent at which our participants mentally anchor some of the information available to them, it can be argued that the effect is present and in fact affecting the offers of our proposers and the acceptances/rejections of our responders thus leading to alterations in the outcomes of the game.
16
5. Results
5.1. Descriptive statistics - participant characteristics
With our recruitment process we, were able to obtain 60 participants in total. One of the major classifications that we have in our game is the split between the two roles in which 50% i.e. 30 participants, were assigned the role of proposers and the rest were responders. The split had to be equal since every proposer is matched with a respective responder and vice versa. Since we are also interested in studying gender as a factor, this means that our sample has to also have representations of both females and males in an equal manner. Consequently, 15 female participants were assigned the role of proposers and 15 were responders. The same split was used to divide the male participants in their roles.
The third major classification of interest is cultural background. Since the participants were asked to provide us with their nationality, this information is used to help us when applying the I/C cultural dimension. By looking online to check for the countries included in the scores for this dimension7,
some of the cultures that we have in our data did not have an I/C score associated with them8. This
led us to our first assumption, which is that any country without an I/C score would be categorized based on the scores of their neighboring countries. An example would be Palestine9, which neighbors
Lebanon and since Lebanon is considered a collectivistic country, the same classification is applied to the participant from Palestine. The same system of classifications is applied to the participants from Uganda, Eritrea and Mongolia. In addition, some of our participants have dual citizenships10, where
all the cases are as follows: one of the held citizenships is considered collectivistic while the other is individualistic. As a result, this made the process of classifying them according to the appropriate cultural background slightly more difficult. Ultimately, this led to our second assumption which is to count individuals with dual citizenships as ones with collectivistic backgrounds. The reason for this decision is because we believe that those participants have pride in their mix of culture and probably incorporate collectivistic beliefs in their day-to-day life so much so that it would lead them to mentioning both cultures when asked about their nationality. In addition, our strong assumptions come as a result of not wanting to exclude any observations from our dataset since it is quite limited.
7 The scores are obtained from the website dedicated to Hofstede’s work on the dimensions: https://www.hofstede-insights.com/country-comparison/
8 In our sample the countries are Palestine, Uganda, Eritrea, and Mongolia.
9 A noteworthy point here is that we considered Palestine a state of its own and did not use the I/C score of Israel as a substitute
10 The combinations at hand are: 2 individuals with a Swedish/Iraqi background, Syrian/Swedish, Japanese/Swedish, Syrian/Canadian and Dutch/Syrian.
17
This eventually leads us to the following cultural breakdown based on our sample: in the proposer group, 63.3% of the participants come from an individualistic background compared to 66.7% in the responder group. The rest of the percentages from both groups are considered as individuals from a collectivistic background i.e. 36.7% in the proposer group and 33.3% in the responder group. A summary of our participants’ characteristics is presented in Table 6.
Table 6: Descriptive statistics – Participants’ characteristics
PROPOSERS OBSERVATIONS PERCENTAGE
Males 15 50.0% Females 15 50.0% Individualistic 19 63.3% Collectivistic 11 36.7% RESPONDERS Males 15 50.0% Females 15 50.0% Individualistic 20 66.7% Collectivistic 10 33.3%
5.2. Ultimatum game results
After looking at our participants’ characteristics in a more in-depth manner, we can take a closer look at how the game itself was played out. Table 7 has a summary of the average offers based on treatments as well as cultural background. Starting with the behavior of all the proposers combined, we observed an average offer equal to 216 sek and a median amount of 250 sek. Diving into the four treatments we have, the following results are obtained. With a Female-Female pairing, the average offer made was 231.3 sek compared to average Male-Male offerings of only 168.8 sek making the difference in the offers equal to 62.5 sek. As for the mixed gender groups, the average offer when the proposer was a female, hence the pairing was Female-Male, was half of the endowment equivalent to 250 sek. In comparison, we observed an average of 218.6 sek in Male-Female pairings. The difference in the average offers of Male-Female and Female-Male offers is equal to 31.4 sek. Median offers were 250 sek with the exception of the Male-Male group which had a median of 200 sek. The least amount offered, equivalent to zero, was by a pairing from the Male-Male treatment and the highest offer was 330 sek from a pairing in the Male-Female treatment. Additionally, proposers from an individualistic background gave on average more than those from collectivistic backgrounds where the latter only offered 204.5 sek compared to 222.6 sek with the difference between the two being 18.1 sek. The median offer based on the cultural breakdown is 250 sek, which is the same irrespective of the cultural background of the proposer.
18 Table 7: Ultimatum game - average offers
GENERAL FF MM FM MF INDIVIDUALISTIC COLLECTIVISTIC
MEAN 216 231.3 168.8 250.0 218.6 222.6 204.5 MEDIAN 250 250 200 250 250 250 250 MINIMUM 0 150 0 200 100 0 50 MAXIMUM 330 250 250 300 330 330 250 STANDARD DEVIATION 71.1 37.2 99.8 28.9 75.8 71.6 72.3 OBSERVATIONS 30 8 8 7 7 19 11
Some noteworthy observations come from the general averages in the game presented in Table 8. Out of the total amounts offered, female proposers gave 240 sek on average compared to 192 sek by male participants. Looking at the offers from the responders’ perspective, female responders, in both Female-Female and Male-Female groupings, received an average offer equivalent to 225.3 sek. On the other hand, the male responders from Male-Male and Female-Male groups, received an average offer that was much less and equal to 206.7 sek.
Table 8: Ultimatum game - general averages
TREATMENTS AVERAGE OFFER
PROPOSERS offer when the proposer is a Female (FF and FM) 240.0
offer when the proposer is a Male (MM and MF) 192.0 RESPONDERS
offer to Female responders (FF and MF) 225.3 offer to Male responders (MM and FM) 206.7
To check for the significance of the differences, the Wilcoxon-Mann-Whitney U test is utilized to compare the averages we have. The test is a well-known nonparametric test, designed to allow for the comparison of the means of unpaired samples (Wilcoxon, 1945; Mann & Whitney, 1947). Even though our samples do not follow normal distribution, the test’s ranking system allows for such a comparison to happen in order to check for the samples’ significance (Wilcoxon, 1945). The standard null hypothesis of the test states that there is no difference between the two samples meaning that the distributions are essentially identical. All the results from the test are found in Appendix E (page 38). Starting with the gender averages, and in particular from same gender groups, we notice a difference of 62.5 sek between the offers of Female-Female and Male-Male pairings. However, this difference is not significant. This means that the offers do not differ when a female or male proposer is paired with a responder from the same gander. The same is found when checking the offers in Female-Male and Male-Female groupings where the difference in average offer, equivalent to 31.4 sek, was not significant. This in turn allows us to reach a similar conclusion. Offers made by female and male proposers did not differ much when paired with responders from the opposite gender. Moving on to the effect of culture on the offers made by the proposers, the difference is 18.1 sek. Yet, the test shows
19
that no significant differences are found when the proposers are from a collectivistic background compared to being from an individualistic one. This leads us to the general conclusion that even though we do notice some variation in the behavior of our proposers, from both the gender and cultural perspective, the differences are not significant enough to be considered.
Continuing onto behavior in the responder group, we only observed 5 cases of rejection. The majority of rejections observed, specifically 3 out of 5, were made by female responders towards offers made by male proposers (Male-Female pairing). The final two rejections came from a Female-Female and a Male-Male pairing. No rejections were noted in the Female-Male treatment. The percentages of rejection, per treatment as well as from a cultural standpoint, are showcased in Table 9. For the cultural breakdown of the rejections, out of the 19 individualistic responders, 10.5% of them rejected offers proposed to them. In the collectivistic sample of responders, consisting of 11 participants, 27.3% of them rejected the offers. The 5 cases of rejections observed, did not allow for the non-parametric test to be performed since the number of rejections is far too little to be used. To be able to analyze data using this test, a sample that is slightly larger than what we have is required11. This deems our obtained
data, regarding rejections, unworkable since the data does not comply with the minimum requirement to be able to conduct the test.
Table 9: Ultimatum game - rejection rates
OVERALL FF MM FM MF INDIVIDUALISTIC COLLECTIVISTIC
REJECTIONS 16.7% 12.5% 12.5% 0.0% 42.9% 10.5% 27.3%
ACCEPTANCE 83.3% 87.5% 87.5% 100.0% 57.1% 89.5% 72.7%
OBSERVATIONS 30 8 8 7 7 19 11
Moving away from the averages within the game, we observed notable behavior by some of our participants. In two instances, proposers offered what could be deemed as extremely unfair offers by proposing 50 sek and zero. What is interesting is how the responders behaved towards such offers. The participant that was offered 50 sek rejected the offer compared to the participant that was offered zero who chose to accept it. Additionally, two of our proposers offered more than 50% of the stake size and instead gave 300 and 330 sek. As for the observed rejections by our female participants, two of them rejected offers of 250 sek, which is an amount conventionally deemed as fair. But to check whether the results that we obtained are significant or not, the following subsection presents the results from the regression analysis run.
11 It is difficult to determine what the smallest possible sample size is, but based on the table of critical values used, any sample smaller than 5 observations does not have an associated critical value
20
5.3. Regression analysis
As previously mentioned, six hypotheses were created to help understand the behavior of our participants in the game. To test these hypotheses, the models discussed in Tables 4 and 5, are run on the software EViews11. Additionally, the hypotheses are tested at the 5% level of significance. All regression results are presented in Tables 10 and 11.
5.3.1. Proposer group
Starting with the proposers, a standard simple linear regression using the Least Square (LS) method is used to compute the models with the dependent variable Offer. This method is used for the regressions since it is one of the simplest, most used, and is typically the standard method when using EViews. The first 3 models, presented in section 4.2, are all regressed and analyzed in the same manner where the constant in the model, which is a representation of the excluded dummy variable, is used to make comparisons. By looking at the results based on our female proposers, we notice that the offer increases by 12.5 sek in Female-Female pairings in comparison to the average offer of 187.5 sek in Male-Male groups. A similar pattern is observed in the opposite gender groups. The average offer in Male-Female groups is 221.4 sek and that offer increases by 40 sek when we have a female proposer offering a male responder (FM pairing). Nonetheless, the differences observed are not significant at the 5% level of significance. Even though both results have no statistical support, we can still note that females are slightly more generous and fairer in their offers, regardless of the gender of the person they are playing against, when compared to their counterpart male proposers.
Table 10: OLS Regression for the first set of hypotheses
OLS regression
Model Variable of interest Constant Coefficient Std. Error t-statistic p-value R-squared
Model 1 FF 187.500 12.500 34.463 0.363 0.720 0.158
Model 2 FM 221.429 40.000 36.843 1.086 0.288 0.158
Model 3 P-individualistic 204.546 18.086 27.211 0.665 0.512 0.016
As for the results from the cultural perspective in model 3, what we observe is an increase to the average offer by an amount equivalent to 18.1 sek when the proposer is from an individualistic background compared to when they’re from a collectivistic one. Yet again, this difference is also not supported statistically, but it could hint that our proposers from individualistic backgrounds are slightly fairer than those from collectivistic ones.
21 5.3.2. Responder group
The regression analysis is slightly different for models 4 through 6, since the dependent variable Rejection, is binary. Having a dummy variable on the left-hand side, leads us to using a Logit model instead of the standard LS method. The reason for using such a model is simply because it provides a more logical approach to analyzing the models at hand. For Logit models, we can calculate the marginal effect which highlights the relationship between the dependent and independent variables. Looking at both same and opposite gender pairings, the dataset at hand proved to be extremely difficult to work with. No results are obtained when we compare Female-Female to Male-Male pairings and Male-Female to Female-Male pairings due to statistical errors. One of the main reasons for such an error lies in the extremely small number of rejections obtained alongside the levels of similarities shared between the dependent and independent variables from models 4 and 5. The only solution to such a problem would be to eliminate the explanatory variable causing the regression method to breakdown. However, in our case, removing such variables means that the model would cease to exist. As a result, we reach no conclusions regarding gender differences in rejection rates. This conclusion, regarding gender differences in rejection rates, is also matched by the lack of findings that we faced when performing the non-parametric test on the same treatments. Moving on to the variable R-individualistic in model 6, we notice a decrease of 18.7% to the overall rejection rate in the model. The results also lack statistical support at the 5% level of significance but could hint towards the fact that having a responder from an individualistic background decreases the chance of obtaining rejections to the offers made in the game.
Table 11: Logit regression for the second set of hypotheses
Logit regression
Model Variable of interest Constant Coefficient Marginal effect Std. Error z-statistic p-value McFadden R-squared
Model 4 FF - - - -
Model 5 FM - - - -
Model 6 R-individualistic -0.847 -1.350 -18.749% 1.016 -1.329 0.184 0.067
A noticeable feature of the presented analysis is that no signs of autocorrelation were present in the Durbin-Watson statistic for the LS regression. Furthermore, the residuals for the models were checked and showed no signs of compliance with the normality assumption. Lastly, both the R-squared and Mc-Fadden R-squared, the alternative measure for Logit regression, of all of the models created were not high, indicating that the calculations obtained were in fact not significant. As a conclusive summary to our results, we are led to believe that the results at hand are not reliable, which is a property that is often associated with analyses of significantly small sample sizes of the likes of our case.
22
6. Discussion
Starting with the gender aspect, the results obtained share both similarities and dissimilarities to previous papers on the same topic. Looking at how the endowments are split by our proposers, females consistently offered a bigger stake, irrespective of the responder's gender, when compared to males. Eckel and Grossman (2001) have similar results, in which the behavior of female proposers is much closer to the point of fairness when compared to that of males. Research typically suggests that females have higher tendencies to avoid risk, regardless of the setting, and that this aversion eventually affects their decision-making abilities (Eckel & Grossman, 2008b; Croson & Gneezy, 2009). Having such a natural inclination by our female participants ultimately leads them to also behave in line with Fehr and Schmidt’s (2001) inequity aversion model presented in section 2.1.2. But whether the decision to offer fairer splits is due to wanting to avoid risk or achieving an overall fair outcome is hard to distinguish, since there’s only a fine line between the two. In addition, the nature of the setup entailed that the game would be hypothetical in its entirety. Research has been able to establish links between an alteration in participants’ willingness to pay, in such setups, and the presence of hypothetical bias (Pesheva et al., 2011). Thus, we can expect that the willingness to pay would be affected when inequity aversion, fairness considerations, as well as the hypothetical bias are all combined together. The interaction between the three factors could potentially provide an explanation as to why our female proposers appear to be much fairer than our male proposers. Another possible explanation would be that male proposers were simply showing signs of standard income-maximizing behavior. Under such an explanation, our female participants could be considered as divergents since their preferences lie outside of wanting to increase their outcome and are actually seeking fairness instead. However, and since the differences in offered amounts are too little, for both same- and opposite-gender groups, this could hint that the behavior of our participants is simply random. This is matched by the findings in both Eckel and Grossman (2001) as well as Solnick (2001). Both papers agree that the observed gender differences are minor enough to be brushed off, which could also be applied to our case as well.
However, a distinction in our results that is not observed in previous research, is that females in our sample were offered more than our male responders. The average offer received by female responders was 225.3 sek compared to only 206.7 sek received by males. A possibility for such a behavior might be due to the low rejection rates of females (Solnick, 2001) which could be associated with their inherent preference for risk aversion. Since female responders have lower tendencies to reject a division of money, proposers would be tempted to offer females slightly higher amounts than what
23
they would offer males. Eventually, this tactic could be considered as a guarantee that the proposers would receive their share of the split more frequently when paired with a female responder compared to a male one. Furthermore, research shows that the behavior of male participants can be influenced by certain situational factors. A study by Sarlo et al. (2013) shows that males are much more affected by framing effects when certain words are used in an experimental setup. For our case, specific words within the sets of instructions for example, might have influenced the way our male proposers made their offers. Regardless, the afore-mentioned explanations are mere possibilities of what could be the potential cause of gender differences. Further research on this specific matter is needed in order to provide much more concrete justifications as to why females were offered more.
Moving the analysis onto a cultural perspective, our findings do not match the conventional results. In our data sample, proposers with an individualistic background, offered on average 222.6 sek, which is higher and closer to a fair split than what proposers from a collectivistic background offered. Even though the difference between the two cultural backgrounds did not have support statistically, it is a noteworthy feature of our data since it is not as common to find such results. When cultural differences are observed, the majority of involved participants are typically from collectivistic backgrounds, such as the case in the paper by Henrich et al. (2001), further highlighting the impact of culture in the prevalence of such differences. As previously discussed, individualistic cultures cultivate the spirit of competition rather than cooperation (Rochat et al., 2009), which would be translated into a pronounced presence of income-maximizing behavior rather than concerns for fairness within the game setting. The same pattern is also observed by Chuah et al. (2009) when they analyzed the behavior of British and Malaysian participants. However, having the opposite present in our data, defies the findings by other researchers. As a result, this opens up the possibilities for a different perspective on how individualism might affect fairness considerations in decision-making where it can instead be attributed to fairer outcomes than what has been commonly seen. Nevertheless, in order to be able to truly understand how individualism and collectivism affect the behavior of individuals in a game setup, a more focused research design is required in order to hone the focus on the cultural aspect alone without having interventions from other competing factors like gender.
Continuing the discussion onto rejection rates, it has proved to be slightly trickier to analyze the gathered data. In general, we had only 5 cases of rejection in a sample containing 30 responders. This makes it harder to distinguish whether both gender and culture did in fact play a role in terms of rejection rates or if the occurrence could be simply deemed as random behavior. Previous literature for example, shows that rejection rates vary between males and females. Some researchers found support to the claim that offers from males are rejected more often than those made by females
24
(McGee & Constantinides, 2013; Eckel & Grossman, 2001) while others find the complete opposite (Solnick, 2001). A similar pattern to what Eckel and Grossman found could be observed in our data, where we noted that 4 out of the 5 rejections were made against offers made by male proposers and that the majority of these rejections were made by females. This could potentially mean that there is a possibility for our data to lean in one direction more than the other, but in general it is an inconclusive result. Moreover, applying cultural analysis for our dataset proved to also be challenging. Researchers have been able to find cultural differences in various experimental setups (Camerer, 2003) and have consistently highlighted their presence, but for us to also make such inferences is not possible. Even though we do find that the majority of rejection cases were made by a pairing between two individuals from a collectivistic background, this cannot provide solid evidence deeming collectivism a factor in the occurrence of rejection.
Reverting back to the reciprocity model from section 2.1.1, we can find evidence of Rabin’s model in our sample of responders while also having some outliers. The model relies on the responders’ ability to reciprocate proposers’ ‘good’ behavior. Hence, any split that would bring potential gains would be matched with an acceptance. Since the majority of the offers were met with acceptances, we can be slightly assured that the model is present in the gathered observations. But in our case, we also had three cases that defy the assumptions of the model itself. An example is having two responders rejecting a split of 250 sek with another accepting a proposer’s offer of zero. The occurrence of such results does not seem to comply with the logical assumptions, which makes it challenging to truly interpret the behavior of our participants. But this opens up the possibility that not every individual is interested in reciprocating good nor bad behavior but instead choose to make circumstantial decisions based on what suits them.
Another possible explanation to the encountered behaviors, would be that everything noted was simply the product of noise in the data. For instance, in the paper by Roth et al. (1991, page 1088), they found evidence of cultural variations in the behavior of their participants but go on to say that by the 10th round all behavior converged towards income-maximization. Such a convergence can be attributed to the learning effect. This effect was first explored by Roth and Erev in 1995 and later by Slonim and Roth in 1998, where they noticed that repeated play had a strong effect on how participants were behaving in the ultimatum game. What they reach as a conclusion is that when the same stake size is used to play multiple rounds of the game, proposers have more time to learn the behavior of the responders allowing for lower offers to appear (Dickinson, 2000). It can be argued that the same behavior would be present in the responder group, where they also eventually learn to accept any offer that provides them with a positive gain. Since our setup did not have neither a test run nor multiple
