Economic games, such as the Ultimatum Game (UG) and Prisoner’s Dilemma (PD) have become popular paradigms for exploring social decision-making. Indeed, the robust behavioral patterns observed across studies suggest that the UG and PD may provide important assessment tools for evaluating social cognition.

The UG, developed by Güth et al. (1982), involves two players who are asked to divide a given amount of money (e.g., £10). The proposer must decide how the money should be divided, whilst the receiver may choose to accept or reject the offer (e.g., the proposer could choose to keep £8 and offer the receiver £2). If the receiver accepts the offer, both players receive the agreed amount, but if the receiver rejects, neither receive anything. The striking finding observed in the UG is that players do not behave according to predictions made by classical economic theories of utility (Von Neumann and Morgenstern, 1947). Specifically, these theories assume that the receiver should accept any offer proposed in order to maximize their own financial gain since the alternative is to receive nothing. Instead, however, players show a consistent willingness to forfeit gains by rejecting offers that are deemed to be unfair. Previous studies suggest that at least half of receivers reject offers below 30% (Güth et al., 1982; Rubinstein, 1982; Thaler, 1988), a finding that has been replicated consistently, with some variations observed in gender (Eckel and Grossman, 2001), social distance (Page et al., 2000), hypothetical vs. real money rewards (Fantino et al., 2007; Gillis and Hettler, 2007) and cultures (Henrich et al., 2005). Acceptance rates can differ quite dramatically between societies, therefore any generalizing from behavior in the games to potential social cognition deficits may only be applicable within a particular culture.

The UG is typically played only once with each ‘opponent’ so that reciprocation cannot explain this robust behavioral pattern. Rejection of unfair offers has therefore been attributed to individuals’ preferences for fairness (Bolton, 1991), a desire to punish socially unacceptable behavior (Fehr and Schmidt, 1999) or an important adaptive mechanism which serves to maintain social reputation and group cooperation (Nowak et al., 2000). These accounts are not mutually exclusive.

In the PD, (Poundstone, 1992), two players can choose to either cooperate or defect in a social exchange situation. For example, in a monetary version of the PD, two players are asked independently whether they would like to cooperate or defect a given amount of money (e.g., £10). If both players choose to cooperate, the money is divided equally between them (£5 each). However, if player A chooses to cooperate whilst player B chooses to defect then player B receives all of the money and player A receives nothing. If both players choose to defect then neither receives any money. Therefore this game highlights the conflict between pure self-interest and mutual cooperation (Axelrod and Hamilton, 1981). Similarly, the Dictator Game (DG) allows a proposer or “dictator” to determine how to split a payoff with the receiver (Kahneman et al., 1986; Forsythe et al., 1994). However, unlike the UG, the recipient is passive and does not have a choice whether to accept or reject the offer. The iterated PD, which involves playing the same player repeatedly, also allows individuals to build strategic relationships by reacting to and punishing an opponent’s past behavior. For instance, players can use a tit-for-tat strategy whereby a player will respond to their opponent with, however, that opponent treated them most recently. Players form a reciprocating strategy where mutual cooperation in the iterated PD is usually higher than in the single shot PD (Axelrod and Hamilton, 1981).

Ultimatum Game and PD paradigms typically use a predetermined and often arbitrary sum of money, to which players have made no contribution. However, in real life settings, rewards are rarely obtained with little or no exertion. Usually, rewards are achieved through effort and are differentially satisfying depending on the relative effort applied and the relative rewards gained by others (Tabibnia and Lieberman, 2007). For example, consider that two colleagues complete a project and are rewarded with a share of the team bonus. If the bonus is evenly distributed between the colleagues, according to traditional UG findings, this would be considered a fair split and both would be satisfied with their equal share of the bonus. However, if one colleague had completed a larger share of the work whilst the other had exerted minimal effort, a 50/50 split might not be deemed fair, especially by the employee who contributed the most to the project. This more sophisticated assessment of fairness may be more typical of many real life social scenarios than the traditional UG and PD games. Indeed, it has been argued that participants in a typical UG may agree on an equal split simply because the stake is a “free gift” from the experimenter (Berger et al., 2012). Therefore taking into account the relative effort in relation to the reward obtained may have important implications for understanding cooperative behavior.

Recently, several investigators using social exchange paradigms have tried to move away from bargaining over a “free gift” from the experimenter. One approach is to delay the reward, with participants having to decide how to split the waiting time between them (Berger et al., 2012). In other studies, players had to decide how to share a “workload” such as solving mathematical questions (Ciampaglia et al., 2014). Studies have shown that individual contributions create “entitlements” which play an important role in offers proposed. For example, in the DG when people earn the right to take the role of dictator, they give less (Hoffman et al., 1994). Similarly dictators give less when they earn income themselves (in an academic test) compared to when income is determined by the experimenter (Cherry et al., 2002; Oxoby and Spraggon, 2008) or won on a coin toss (Ruffle, 1998). Carr and Mellizo (2013) reported an adapted UG where responders solve math problems to generate an endowment whereby responder-produced endowments increase offers while acceptance rates are higher suggesting that cooperative behavior is driven by the source of the entitlement. Further Cappelen et al. (2013) conducted a real-effort DG where students in two of the world’s richest countries, Norway and Germany, were matched directly with students in two of the world’s poorest countries, Uganda and Tanzania who generated money through a word entry task and found that participants assigned greater importance to individual contributions in their distributive choices than they did needs and nationality. Individual contributions, and thus entitlement therefore, appear to play a vital role in people’s distributive behavior (Hoffman et al., 1994; Konow, 1996, 2010; Cherry et al., 2002; Frohlich et al., 2004; Gächter and Riedl, 2006; Cappelen et al., 2007, 2010; Oxoby and Spraggon, 2008).

However, a major drawback of the UG and PD is that they can require a long and complex set-up with multiple players or confederates, which involves pairing participants either in person or more recently via the internet. In clinical trials and studies focusing on patient populations, this may not be possible given the logistics involved. Some studies match each player’s individual decision with previous real player responses. However, data can only be obtained in response to a previous player’s individual behavior, which may be aggressive or cooperative therefore not allowing each player to play under the same circumstances. Playing against a computer, however, allows the systematic manipulation of an opponent’s behavior to enable investigation of player behavior against different types of opponents. This allows the assessment of punishment of uncooperative behavior and negative reciprocity, (e.g., Fehr and Schmidt, 2006) in response to opponent defection. We sought to explore the effects of contribution and entitlement on reciprocity using an iterated PD so that we could establish whether participants who had contributed more to the pot of money were more likely to defect compared to when the opponent contributed more and when contributions were equal when an opponent has been aggressive or cooperative. We therefore aimed to develop versions of these economic tasks that are short, one player and easy to administer on a laptop or tablet, either in a laboratory setting or at home. Participants were not deceived in any way to believe their avatar opponent was a real person, nor were they led to believe that the responses of the avatar were based on previous participants’ responses. To our knowledge, there has been no empirical exploration of the impact of relative monetary contributions and entitlement in the UG and PD using computer avatar opponents.

The aim of this study was to evaluate how relative contribution to the common resource, and therefore a sense of entitlement, affects cooperative behavior. We predicted that varying levels of contribution would shift the distribution of offers, such that greater contribution would result in significantly less cooperative behavior. We predicted that this would be manifested in more rejection of offers in the UG and more defecting behavior in the PD.

Economic social exchange games are widely used to study cooperative behavior, yet two central caveats of these games is that first, players must divide an arbitrary sum of money to which they have not contributed and therefore may have little investment in. However, rarely in real life is bargaining behavior and cooperation devoid of motivating factors such as investment and contribution. In this study we found that contribution to the monetary stake robustly affects the propensity to engage in cooperative behavior. Second, economic games used to study social cognitive decision-making typically use a complex set-up of pairing participants, which is not practical. Here we observed behavior similar to previously reported real-time UG and PD therefore suggesting that games using computer avatars may be an effective alternative to a traditional set up.

As expected, we found that UG acceptance rates increased as avatar opponent offers become fairer (e.g., Güth et al., 1982). As we hypothesized, contribution had a substantial effect upon both acceptance rates and proposals. The highest acceptance rates were evident consistently when the avatar opponent contributed more, suggesting that people are more likely to consider an unequal offer as fair when they have made a lower contribution. Consistent with this, when participants acted as proposers, they made fewer 50% offers when their own contribution was greater, again indicating that relative contribution impacts perceptions of fairness. We also observed that when playing the role of the proposer, participants made more offers at the extremes, i.e., equal (50%) and very unequal offers (20%) compared with intermediate offers (40 and 30%). This is in line with the results of meta-analyses, suggesting that dictators’ offers in the DG follow a bimodal distribution, with a high concentration of purely selfish and perfectly fair offers (Engel, 2011; Tisserand et al., 2015). This suggests that perhaps the participants played more like a dictator against a passive opponent as opposed to a typical proposer in the ultimatum game that must have strategic considerations in order to avoid the rejection of the offer. It might therefore be interesting in future studies to manipulate an avatars acceptance threshold in order to ascertain how it might affect players’ proposer behavior.

We observed a small (r ∼-0.2), but significant, negative correlation between UG receiver and proposer behavior whereby participants who offered more also accepted fewer offers. This replicates previous findings suggesting that a participant with a high minimum acceptable offer will expect others to behave similarly and therefore increase their offer (Blanco et al., 2011). Additionally we detected a moderate (r ∼-0.5) negative correlation between defect rates in the PD and offers proposed in the UG in line with previous DG studies (Dreber et al., 2013; Capraro et al., 2014). However, despite good statistical power, we did not identify a significant correlation between PD and UG offer acceptance rates, suggesting that these games may measure dissimilar underlying processes. Previous studies have also observed a lack of correlation between UG acceptance and PD cooperation, which has been attributed to the PD reflecting cooperation and reciprocity, with the UG acceptance rates reflecting assertiveness and a tendency to avoid being dominated by the proposer (Yamagishi et al., 2012).

We further demonstrated that acceptance rates in the UG correlated positively with state anxiety, suggesting that negative affect is an important factor in cooperative behavior. Previous studies have reported that anxious traits lead to greater cooperation (Clark et al., 2013) and individuals with anxiety and depressive disorders are more likely to display cooperative behavior (McClure et al., 2007) and accept significantly more unfair offers than healthy controls (Grecucci et al., 2013).

As expected, our PD results showed that the highest defect rates occur when healthy volunteer participants contributed more to the financial pot, suggesting that they feel entitled to a larger share of the pot in this situation. However, defect rates were also higher when the computer opponent contributed more compared to when contributions were equal, which seems counterintuitive. A possible explanation is that participants may predict that their opponent is more likely to defect in this scenario, and therefore may also choose to defect in order to sabotage their opponent’s gains (i.e., in this case participants would rather see both players receive nothing than allow their opponent to take all the money; a form of punishment behavior). Previous studies have demonstrated that participants’ prediction of cooperation is extremely accurate (Frank et al., 1993; Brosig, 2002) such that they are inclined to play cooperatively against those who are expected to cooperate and to defect against those more likely to defect (Oberholzer-Gee et al., 2010).

We also observed that whilst overall cooperative behavior does not differ between males and females, the manipulation of contribution produces clear gender differences. Specifically, females defect more frequently when they have contributed more, whereas males’ defecting behavior is also relatively high when the opponent has contributed more. The defect response to predicted opponent defection hypothesized in the preceding paragraph would therefore seem to be particularly marked in male participants, potentially consistent with the idea that males are more prone to “altruistic punishment” of perceived unfairness (Singer et al., 2006). Whilst previous literature regarding gender differences in the PD is mixed (Rapoport and Chammah, 1965; Sibley et al., 1968; Tedeschi et al., 1969; Kahn et al., 1971; Mack et al., 1971; Dawes et al., 1977; Orbell et al., 1994), here we demonstrate the importance of considering contribution as a key factor influencing cooperative behavior between the sexes. Interestingly, the genders did not differ on our task when contributions were equal, most similar to typical implementations of the PD.

Many studies using the UG and PD employ time consuming and elaborate set-ups. However, we found that despite participants being fully aware that they were playing against a computer opponent, the degree of cooperative behavior when both players contributed equally was comparable to previous studies (Güth et al., 1982; Rubinstein, 1982; Thaler, 1988). However, some studies have found higher rejection rates for human compared to computer avatar opponents (Blount, 1995; Knoch et al., 2006). In the PD our rates of cooperation of 60% were slightly greater than the average figure of 47% obtained in a meta-analysis of 130 single shot PD experiments (Sally, 1995). Nevertheless, more recent real life studies also show slightly greater acceptance rates: van den Assem et al. (2011) analyzed 574 contestants’ behavior from the United Kingdom game show “Golden Balls,” adapted from the Prisoner’s Dilemma, finding that players cooperate on average 53% of the time. Moreover, during piloting of the tasks we found no significant differences between games played against the computer opponent and games played against a confederate human opponent. Therefore computer avatars may serve as a suitable and more practical alternative to the traditional setup. Indeed, it has been argued that human–computer interactions are evolving with computers increasingly being viewed as more human-like with autonomous interactions (Cassell and Bickmore, 2000). Remarkably, the substantial effects of contribution suggest that participants treat the computer avatar as though they have exerted effort much like a human opponent. It should be noted that we recruited relatively young participants who are more likely to be familiar with human–computer interactions such as gaming and the use of humanized computer avatars. We may have seen a different pattern in an older population.

A limitation of our findings is that money was accrued differently in the two games; “won” in the UG and “earned” in the PD. Recent experimental work on social preferences has shown that people are much more willing to accept inequalities due to effort than inequalities due to luck (Cappelen et al., 2007, 2013; Konow, 2010). However, in neither game was the pot of money a “free gift” from the experimenter. Evidence suggests that people who believe economic outcomes mainly depend on effort show less cooperative behavior whereas those who believe that other factors not under an individual’s control determine economic outcomes engage in more social cooperative behaviors such as wealth redistribution and charitable donations (Rey-Biel et al., 2011). Indeed, entitlements represent a strong motive over above consideration for needs and nationality that may contribute to explaining why people do not give a larger share of income to the more needy (Cappelen et al., 2013). In addition, potential framing of the tasks, which used phrases such as “win” and “steal,” is an important caveat that has been shown to influence cooperative behavior (e.g., Ellingsen et al., 2012). An interesting future direction would be to manipulate the framing of the tasks to ascertain how it affects cooperative behavior particularly in a human–computer interaction.

Another possible limitation of the study is that the offers in the UG had seven levels: 10, 20, 25, 30, 35, 40, or 50% whereas when players take the role of dictator they only have four offer options; 20, 30, 40, or 50%. In future versions of the tasks it would be valuable to match the offer levels in the two roles and to perhaps consider including opponent offers above 50%. For potential further studies, it would be interesting to evaluate whether behavior differs if the avatar of the computer opponent is replaced by an image of a computer screen, or a neutral stimulus such as a gray circle. This would allow investigation into the effect of using human-like avatars in these economic games. Further systematic exploration assessing human–human and human–computer interaction is needed to establish whether cooperation with an avatar provides real insight as to an individual’s cooperation with other human agents.

Finally, an important limitation of this study is that we have only used two games, which by no means comprehensively assess cooperative behavior. The PD and UG are very different games and are likely to be manipulating different aspects of behavior. Furthermore, in the versions we used here, the method of accruing endowments were not matched and the UG was played as a one shot game while the PD was repeated. Future studies should employ additional tasks and systematically manipulate game variables across these tasks in order to more fully characterize the behavior we report here.

Taken together, our results demonstrate that contribution is an important motivating factor that affects behavior in economic social exchange games. Traditional versions of these games involve the experimenter providing an arbitrary pot of money that must be divided. In the versions used here, participants either won or earned money, making either equal or unequal contributions to the communal pot. This version of the UG allowed us to show that perceptions of fairness depend on how potential rewards have been accrued. Participants were more accepting of unequal offers when they had made a smaller contribution, and less likely to make equal offers when they had made a larger contribution. For the PD we found that participants were more likely to split earnings when contributions were equal and females were more likely to defect when they had contributed more. Males were more likely to defect either when they had contributed more or when the opponent had and we suggest that the latter behavior may represent anticipatory punishment of a predicted defect by the opponent. Our findings have important implications for understanding perceptions of fairness and cooperative behavior in real-life situations where people have expended effort together to achieve gains that must then be divided.

Chief Investigator: RE. Principal Investigators: JR, MM, BS, and TR. AB and TS collected and analyzed data.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.