آیا طرفدار اجتماعی رفتاری بازی خاص است؟ ترجیح طرفدار اجتماعی و انتظارات طرفدار اجتماعی

Abstract We observed the behaviors of the same people across five games – two prisoner’s dilemma games, a trust game (in which the subject took on the role of both truster and trustee), a dictator game, and a faith game – any pair of which was separated by an interval of several months to reduce potential carry-over effects, and found strong consistency in behaviors among these games. We also found consistency between the expectations of other players’ behaviors and the player’s own behavior across games. We further found that the consistent behavioral pro-sociality observed across different games was related to the general measures of pro-social value orientation and perceiving the game situations. These findings suggest that individual and cultural differences in game behaviors can reflect both the ways in which people perceive game situations and their general social preferences.

Introduction The goal of our study is to determine whether people behave consistently across different types of experimental games and, if so, what drives this behavior. To investigate this question, we observed individuals’ behaviors across different games and found that these behaviors were fairly consistent. We first present the theoretical foundations used to predict why we should find such behavioral consistency and then present our research methods. Finally, we discuss the implications of our findings. Social value orientation One factor for which we expect behavioral consistency across different experimental games is the participant’s social value orientation (SVO). In social psychology, efforts to identify the motivational bases of cooperation can be traced to earlier studies of decomposed games (Messick & McClintock, 1968). Over time, the decomposition methodology of identifying players’ motivational bases developed into more comprehensive measures of SVO (Kuhlman and Marshello, 1975, Liebrand, 1984 and Van Lange et al., 1997). Simply put, an individual’s social value orientation (SVO) is defined by a combination of weights given to his/her own and his/her interaction partner’s welfare in evaluating the desirability of a particular outcome in an interdependent relationship. Although nine prototypical SVOs can be produced by combining the positive, negative, and zero weights for one’s own and one’s partner’s welfare, three major types of orientations have been studied extensively: cooperative, individualistic, and competitive. Cooperators (sometimes called joint gainers) assign positive weights to both their own welfare and their partner’s welfare when evaluating the desirability of a particular outcome. Individualists assign a positive weight to their own welfare and disregard their partner’s welfare. Competitors give a positive weight to their own welfare and a negative weight to their partner’s welfare. Because the proportions of competitors and altruists (who assign a positive weight to their partner’s outcome but no weight to their own) are found to be relatively small, individualists and competitors are often categorized together as pro-selves, and cooperators and altruists are often grouped as pro-socials (Liebrand, 1984 and Van Lange and Kuhlman, 1994). While the distinction between pro-socials and pro-selves was originally proposed based on the relative weights they assign to the player’s own outcome and his/her partner’s outcome, more recent characterizations of these two types of players involves preferences for the equality of the two players’ outcomes. In other words, pro-socials prefer equality of outcomes; they place a positive value on their partner’s welfare (van Lange, 1999) and often prefer equality more than joint gain (Eek & Gärling, 2006). Previous studies have repeatedly demonstrated that a player’s SVO is moderately correlated with his/her cooperative choices in social dilemmas. For example, Balliet, Parks, and Joireman (2009) conducted a meta-analysis of social dilemma studies and found that the effect size of the relationship between cooperation and SVO in those studies was r = .30. Efforts to relate an individual’s cooperative choices in social dilemmas and related games to his or her motivational bases are not unique to social psychologists. Some economists (e.g., Andreoni, 1990, Bolton and Ockenfels, 2000, Fehr and Gächter, 2002, Fehr et al., 2008, Fehr and Schmidt, 1999 and Rabin, 1993) use the concept of social preferences or non-self-regarding preferences in evaluating the outcome of choices of interdependent relationships. These social preferences include altruism, fairness, inequity aversion, reciprocity, strong reciprocity, and spite. Despite certain differences in the specific types of preferences or value orientations, the same basic idea is shared by economists and social psychologists: people evaluate the desirability of the outcomes of their own actions (in conjunction with the actions of their interaction partner) using both their own and their partner’s welfare as criteria. Furthermore, both economists and social psychologists assume that the behavioral choices made in experimental games reflect the social preference and social value orientation in the minds of the game players. Because game players are driven by internal traits, such as social value orientation and social preference, some level of consistency is to be expected across different game behaviors. Specifically, it is expected that those who have a pro-social value orientation will consistently behave in a pro-social manner in games in which social value orientation affects player choices. For example, players are expected to cooperate in the prisoner’s dilemma game, provide a fair share in the dictator game, return a fair share of the entrusted money as trustees, and entrust some of their own money as trusters in the trust game. The consistency prediction based on the player’s SVO is straightforward for games in which players must share their own resources with their interaction partners. For example, the dictator in the dictator game gives some portion of the endowment he/she controls to his/her interaction partner. The prisoner’s dilemma game can also be expressed in the exchange format (Yamagishi et al., 2007 and Yamagishi et al., 2005) in which each player provides some portion of his/her resources to the partner. Similarly, the trustee in the trust game gives some portion of the entrusted money to the truster. In these games, giving can be regarded as a form of (unconditional or conditional) altruistic behavior. Trusting behavior in the trust game can also be regarded as a form of altruism because the trusting choice increases the total resources allocated to the two players. However, the trusting choice in the faith game (FG; Kiyonari and Yamagishi, 1999 and Kiyonari et al., 2006) involves no increase in the joint outcome or the partner’s outcome. The faith game (FG) is played by two players – a dictator and truster. The dictator plays a dictator game in which he/she divides a sum of money between him/herself and a recipient. The dictator then receives the portion of the money he/she kept regardless of the choice of the recipient/truster. The truster knows that the dictator believes that he/she totally controls the resource allocated between the two, and decides whether to receive the unknown amount of money a randomly matched dictator allocated to a recipient or to receive directly from the experimenter a fixed amount of money that is known to be less than half of the amount that the dictator divided. The truster in the FG therefore has no control over the amount that the matched dictator receives. Unlike in the TG, even pro-social trusters in the FG who want to promote the dictator’s outcome cannot do so by choosing to trust because the trusting choice might reduce the truster’s (and the joint) outcome in cases where the matched dictator has chosen to take the full endowment. He/she does not know if the choice of trust (i.e., choosing to receive whatever the dictator allocated to a recipient) increases or decreases the inequality between the two. Thus, the truster’s type of SVO or social preference cannot be used to predict his/her choice in the FG. Prediction of the trusting choice in the FG based on the player’s SVO requires the assumption that players project their own SVO onto the dictator, such that pro-socials expect the matched dictator to be similar to themselves and behave pro-socially, whereas the pro-selves expect the dictator to behave in a purely self-promoting manner and take all of the endowment he/she controls. This “projection hypothesis” seems reasonable considering the strong correlation between players’ own choices and the expectations of the partner’s choice often observed even in one-shot games (Acevedo and Krueger, 2005, Krueger et al., 2012, Watabe et al., 1996 and Yamagishi and Sato, 1986). Post-decisional and pre-decisional projection Although the projection hypothesis appears to be straightforward and reasonable, what are actually projected is not as clear. One interpretation is that the player uses his/her own behavior as a sample of size one drawn from the population of game players (Dawes, 1989 and Krueger et al., 2012). This type of “post-decisional” projection occurs after the player has made his/her choice and thus cannot explain the correlation between pre-decisional expectations and the actual choices (Messick et al., 1983). Furthermore, this type of prediction is expected to occur only within the same game. More specifically, the range of this type of prediction is limited to the choices in games that are similar to the one in which the player has already made his/her choice (i.e., the population corresponding to the sample of size one). Thus, we cannot use this type of post-decisional projection as a supplemental assumption to explain the consistency of behavior in the FG and other games. For example, this type of projection cannot explain the behavioral consistency between the trust choice in the FG and reciprocation of trust by the trustee in TG because there is no behavioral counterpart in the trustee’s choice in the TG corresponding to the choice of trust in the FG. One implication of the above argument is that a different type of projection is needed as a supplemental assumption to allow for the prediction of consistency between the FG and other games. What is needed is a “pre-decisional” projection, or an inference of other people’s internal state (including their perception of the game situation) on which their behavior is based. The player thinks, “how would I feel and act in that person’s position?” As discussed in the empathy literature, this inference of another player’s choice can be based on emotional simulations or the application of the naïve theory of humans (Gallese and Goldman, 1998, Stueber, 2006 and Zahavi, 2008). The player’s own SVO may be projected onto another player via emotional simulation (the player thinks, “how would I feel if I made a given choice in that situation?”). A pro-social player would think, “I’d feel happy if I make a choice that makes my partner happy, and I believe that other people feel the same way.” The inference of the partner’s internal state according to simulated emotions allows us to predict behavioral consistency between the FG and other games. Pro-social players who actually feel happy when they behave in a pro-social manner would expect that other people would also feel that way when they behave in a pro-social manner, regardless of the types of game that they are playing. In addition to using simulations of one’s own feelings, the other player’s choice may be predicted based on naïve theories that the player has about human interactions in general. One model of human behavior accepted by many economists is that humans are rational pursuers of self-interest; this is an example of a naïve theory. According to the naïve theory of humans, the player’s actions may partly reflect his/her own SVO. For example, pro-social players may share the view that humans are generally pro-social, and pro-selves may share the view that humans generally care most about their own welfare. However, the outcome of applying such a theory can be affected by both internal factors (personal experiences) and external factors (the historical time or the culture/society) in which the player lives, as demonstrated by Henrich et al. (2005), who conducted ultimatum and other game experiments in industrial and primitive societies and found that the level of market integration can greatly affect societal differences in game behavior. One interpretation is that different societies simply have different social preferences. As Yamagishi (2005) noted, however, this observation may actually be based on differences in the naïve theory of how people would behave in the games. Perception of the game situation The way in which the game situation is perceived and interpreted by the game players may provide another source for the behavioral consistency observed across different games. Some people may perceive the game situation as a means for mutual cooperation, while others may instead view it as a means to compete with others to secure their own resources. Kelley and Stahelski (1970) first suggested attributing differences in game behavior to both player type and the player’s perception of the game situation, and proposed a triangle model of behavior and expectation. Their study demonstrated that cooperators tend to believe that some of the people they encounter in experimental games are cooperative and others are not, whereas defectors tend to believe that people are generally uncooperative. This finding suggests that cooperators and defectors have different types of naïve theory in terms of humans and the game situation. The difference in the ways that cooperators and defectors perceive humans (termed “Kelley and Stahelski’s triangle model”) was further elaborated upon and expanded into the might-over-morality hypothesis proposed by Liebrand et al., 1986, Beggan et al., 1988 and Van Lange and Kuhlman, 1994). According to this hypothesis, cooperators and defectors interpret other people’s cooperative behaviors differently: the former individuals regard cooperative behavior as a reflection of the actor’s good intentions, and the latter regard it as a reflection of their weaknesses. The former perceive the game situation as a situation for moral behavior, whereas the latter perceive the same game situation as a power struggle. The aforementioned might-over-morality studies suggest the possibility that the players’ perceptions of human nature and the game situation provide the basis for their behavioral consistency across different games that is independent of the consistency produced by their social values and preferences. As Bogaert, Boone, and Declerck (2008) argue, these two sources of consistency – players’ SVO and their beliefs and perceptions of humanity at large – may be tightly intertwined. They argue that social value orientation is an amalgam of an individual’s self- and other-regarding preferences and their views of humans and the social world. In other words, pro-socials are characterized by their care for the welfare of others and by a socially oriented rationality in which cooperation could be considered an intelligent behavior. In contrast, pro-selves are characterized a lack of care for the welfare of others and by the belief that all other people are similarly self-serving. This assumption may in fact be true, such that these two factors inseparably constitute the pro-social and pro-self syndromes. However, these two factors may have relatively independent influences on individual behavior in experimental games and, in particular, on the behavioral consistency within those games. Assuming that this study’s first goal (to demonstrate behavioral consistency between different types of experimental games) has been achieved, the second goal of this study is to determine whether players’ beliefs about human nature and their perceptions of the game situation can predict their behavioral consistency more effectively than can their social value orientation alone. The third purpose of this study is to explore how the players’ social value orientation and their perceptions of the game situation interact and jointly affect their behavioral consistency in experimental games. We have not formulated specific hypotheses about the relationship between players’ SVO, their perceptions of the game situation, or their behavior in different games. The findings from this study are expected to provide valuable information that helps us clarify the relationship between the two.

Findings Basic game behaviors Table 2 shows the means and standard deviations of the game behaviors. In this table, we report the mean of the cooperation levels collected across three trials as the measure of PDG2 because the correlations among the levels of cooperation in the three trials in the PDG2 were high (r = .88, .89, and .92). Some of the sample sizes in this table are smaller than those listed in Table 2 because we excluded participants who, based on their post-experimental questionnaire responses, were found to not understand the basic game instructions. Table 2. Means and standard deviations of behaviors in experimental games. Game behavior n Yen SD Ratio to the endowment PDG1 96 720.83 592.75 .360 PDG2 76 166.67 132.66 .333 tTG 93 298.92 205.10 .498 rTG 93 .398 .214 DG 97 1088.67 522.84 .435 tFG 85 244.71 164.39 .408 Table options Two PDGs We conducted the one-shot PDG twice: once as a pure one-shot game (PDG1) and once as a repeated one-shot game (PDG2). Despite differences in the instructions and the long time intervals between the two games, the correlations between the two one-shot PDGs were fairly high (r = .68, p < .001). 6 This finding marks the level of correlation that can exist between any of the same two games and against which the levels of correlation between different games should be compared. The lower-left triangle of Table 3 shows the correlations between expectations for these games. Again, expectations in the PDG1 were strongly correlated with those in the PDG2 (r = .68, p < .001). Table 3. Correlations among game behaviors and expectations across games. PDG1 PDG2 tTG rTG DG tFG PDG1 96 .68⁎⁎⁎,a .50⁎⁎⁎ .50⁎⁎⁎ .05 .53⁎⁎⁎ PDG2 .68⁎⁎⁎,b 76 .50⁎⁎⁎ .57⁎⁎⁎ −.05 .76⁎⁎⁎ tTG .42⁎⁎⁎ .49⁎⁎⁎ 93 .52⁎⁎⁎ .08 .59⁎⁎⁎ rTG – – – 93 −.00 .36⁎⁎ DG .26⁎ .17 .17 – 97 .09 tFG .53⁎⁎⁎ .50⁎⁎⁎ .38⁎⁎⁎ – .35⁎⁎ 85 Note. Expectations of the allocator about how much the matched truster entrusted were not assessed. The upper right triangle is for behaviors, and the lower left triangle is for expectations. The diagonal entries indicate the number of participants used for the analysis. ⁎ p < .05. ⁎⁎ p < .01. ⁎⁎⁎ p < .001. a .70 In Sample 2. b .61 In Sample 2. Table options Correlations among game behaviors The across-game correlations (the upper-right triangle of Table 3) are fairly high, especially in comparison to the correlation of .68 (p < .001) between the two PDGs. However, we noted a lack of correlation in DG allocation choice that will be discussed later in this paper. Of particular interest is the fact that the trusting choice in FG was correlated strongly with the other game behaviors (except for the DG). Principal factor analysis of the game behaviors excluding the DG yielded a single factor solution. The factor loadings were .69, .92, .70, .64, and .79 for the PDG1, PDG2, tTG, rTG, and tFG, respectively. We decided to use this factor score as an indicator of each participant’s general behavioral pro-sociality (gPS). 7 Correlations among expectations The correlations among expectations for different games shown in the lower-left triangle of Table 3 were also generally high compared to the correlation between expectations in the PDG1 and PDG2 (r = .68, p < .001), although the dictator game was again an exception. We also conducted a principal factor analysis of the expectations and obtained a single factor, the loadings of which were .74, .87, .50, and .60 for the PDG1, PDG2, tTG, and tFG, respectively. We decided to use this factor score as an indicator of the general expectation of other players’ pro-social behavior (gEXP). Correlations between game behavior and participant expectation Table 4 shows the correlations between behaviors (columns) and expectations (rows) within and across games. The correlations between the allocation choice in the DG and expectations in the other games were all low and non-significant, as were the correlations between the expectations of the dictator’s choice in the DG and the actual behaviors in the other games. We will return to a discussion of this finding concerning the actual choice and expectations of the DG. Table 4. Correlations between behaviors (columns) and expectations (rows) within and across game behaviors. Expectations Behaviors PDG1 PDG2 tTG rTG DG tFG gPS Pro-social PDG1 .80⁎⁎⁎ .65⁎⁎⁎ .42⁎⁎⁎ .39⁎⁎⁎ .08 .45⁎⁎⁎ .64⁎⁎⁎ .29⁎⁎ PDG2 .71⁎⁎⁎ .76⁎⁎⁎ .34⁎⁎ .44⁎⁎⁎ −.01 .49⁎⁎⁎ .72⁎⁎⁎ .22⁎ tTGa .52⁎⁎⁎ .52⁎⁎⁎ .45⁎⁎⁎ .74⁎⁎⁎ .03 .23⁎ .50⁎⁎⁎ .26⁎⁎ DG .17⁎ .06 −.02 .14 .33⁎⁎⁎ .14 .05 .01 tFG .50⁎⁎⁎ .52⁎⁎⁎ .40⁎⁎⁎ .44⁎⁎⁎ .03 .50⁎⁎⁎ .53⁎⁎⁎ .29⁎ gEXP .78⁎⁎⁎ .79⁎⁎⁎ .46⁎⁎⁎ .50⁎⁎⁎ .02 .54⁎⁎⁎ .78⁎⁎⁎ .32⁎⁎ Pro-social .34⁎⁎ .35⁎⁎ .30⁎⁎ .33⁎⁎ .08 .31⁎⁎ .39⁎⁎⁎ Pro-social controlling for gEXPb .21 .15 .20 .23 .19 .21 .24 ⁎ p < .05. ⁎⁎ p < .01. ⁎⁎⁎ p < .001. a Expectation of the trustee’s choice by the truster on a 7-point scale. b The general measure of Pro-social SVO after controlling for gEXP. Table options Except the DG, two prominent patterns emerged (Table 4). First, the behavior–expectation correlation is generally high within a single game, especially in each of the two PD games, and the correlation between expectations of the trustee’s behavior by the truster and the trusting choice of the truster are also high.8 Second, the behavior–expectation correlations across different games are similarly high, though not as high as those found within the same game. The latter finding is of particular interest because such behavior–expectation correlation cannot be explained using post-decisional projection of the player’s own choices. Behavioral pro-sociality and general expectations Table 4 also reports correlations between the general behavioral pro-sociality index and expectations in different games (the last column) and between correlations of the general expectation index and different game behaviors (second to last row). Each of these correlations was relatively high, except those involving expectations and behavior in the DG. The correlation between behavioral pro-sociality and general expectations was strong (r = .78, p < .001), exceeding the strength of the inter-behavior correlations found between the PDG1 and PDG2. Such a strong correlation is hard to interpret as a strict consequence of the SVO, in which pro-socials would be expected to engage in pro-social behaviors in various games and assume that the other players would behave in the same fashion in a variety of games. If the SVO was the cause of this consistency, the correlation between the general behavioral pro-sociality and general measure of expectations should be substantially reduced by controlling for player’s SVO. The general measure of SVO was positively correlated with both the general pro-sociality (r = .39, p < .001) and general expectation (r = .32, p < .01). 9 Interestingly, the correlation between the two was reduced only slightly (from .78 to .75) after controlling for SVO, suggesting that the expectations, especially the generalized expectations, are derived from the model of humans (how ordinary people would behave in this type of situation), which in turn reflect the personal, historical, social, and cultural experiences of the players in addition to their projections of their own pro-sociality. The trusting choice in the faith game One of the central concerns in this study is the relationship between the trusting behavior in the FG with the other game behaviors because such a relationship is not predicted purely using the player’s social value orientation. This prediction requires an auxiliary assumption, as discussed earlier. The correlations shown in Table 3 indicate that the trusting behavior in the FG was strongly correlated with the other game behaviors, with the exception of allocation choice in the DG. The correlation between the PDG2 and tFG (r = .76, p < .001) was even stronger than that between the two PDGs (r = .69, p < .001). Social value orientation Table 5 lists the distributions of the SVO types according to each of the four SVO measures. Because the numbers of altruists and competitors were small, we classified our participants into pro-socials (altruists and cooperators) and pro-selves (individualists and competitors). As shown in the table, the percentage of participants classified as pro-socials actually varied between the two methods. We first classified the participants according to each measure by combining the cooperators and altruists as pro-socials and combining the individualists and competitors as pro-selves.10 The agreement rate within these categories between the three choice measures taken at intervals of 26 months was 78.6%, and the agreement between the two ring measures (taken at an interval of 31 months) was 57.7%. We constructed the overall triple dominance measure of pro-sociality by first classifying each triple dominance measure into either 0 for pro-selves and 1 for pro-socials and then taking the average of the two measures. Similarly, we constructed the overall ring measure of pro-sociality by taking the average of the two ring measures. The correlation between these two overall measures of pro-social SVO (r = .38, p < .001) was reasonably high. Finally, we constructed a general measure of pro-social SVO by averaging the two overall measures. Table 5. Frequencies of cooperators, individualists and competitors. Three-choice Ring measure Wave 1 Wave 4/7 Wave 2 Wave 8 Altruists 0 0 5 7 Cooperators 74 151 34 51 Individualists 19 41 30 88 Competitors 7 2 3 3 Unclassified 5 10 28 34 % Pro-socials 70.5% 74.0% 34.0% 31.7% Sample S1/n = 105 S1 + S2/n = 204 S1/n = 100 S1 + S2/n = 183 Note. The numbers reported in this table for Waves 4/7 and 8 are based on the combined sample (Samples 1 and 2) to give readers more information. The numbers reported in the analysis section are based on Sample 1 because the other variables used for analysis are only from that sample. Table options The mean of this general measure of SVO (gSVO) was 0.60 (SD = 0.30). All of the game behaviors were found to be moderately correlated with this general measure of SVO, except the DG (see the second to last row of Table 4). Moreover, the gSVO was moderately correlated with expectations in all games except the DG (see the last column of Table 4). In the earlier analysis (section on general behavioral pro-sociality and the general expectation), we learned that the strong behavior–expectation correlation cannot be reduced to the common cause of social value orientation. Furthermore, it was suggested that the generalized expectations reflect factors other than the players’ own pro-sociality. We further found that the observed correlations between game behavior and gSVO could be substantially reduced by controlling for generalized expectations (see the last row of Table 4). The correlation between the gSVO and generalized behavioral pro-sociality was also reduced to .24 (ns.) from .39 (p < .001). This finding is consistent with Pruitt and Kimmel’s (1977) goal/expectation theory of cooperation, which asserts that cooperation is a joint function of the players’ preference for cooperation and the expectation that their partners will also cooperate. Perception of the game situation In the post-experimental questionnaires following the PDG1, PDG2, TG and FG, participants rated the extent to which they believed that the experimental situation resembled specific real-life situations. The players were shown nine characteristics of each situation and subsequently evaluated how similar they thought the experiment was to each specific situation. The participants’ responses to these questions in the four games (a total of 36 items) were subjected to a principal factor analysis using the varimax rotation. After eliminating three characteristics that did not load strongly on any meaningful factors, two useful factors were obtained from this analysis. Table 6 shows the two factor loadings after removing the three characteristics. Impressively, the factor loadings are similar across the four games that were conducted over a three-and-a-half-year period. This finding suggests that participants evaluated the game situations using the same frame regardless of the differences in the incentive structure between games. For example, they perceived the FG with the same framework (or dimension) that they used to perceive the prisoner’s dilemma game. We constructed two situational characteristic scales by taking the respective means of the first three characteristics and second three characteristics of the situation across the four games. We named the first and second scales the “collaborative situation scale” and “business transaction scale,” respectively. Those who scored high on the collaborative situation scale consistently perceived the four games as instances of social exchange, in which both they and the other players provided and received help. Those who scored high on the business transaction scale consistently perceived the four games as instances of business transactions, in which individuals pursued their own interest by taking advantage of others. Table 6. Factor loadings of the situational characteristics. Situation In Factor 1 social exchange Factor 2 business transaction Mutual help PDG1 .71 .17 PDG2 .69 .13 TG .82 .11 FG .74 .02 Collaborative work PDG1 .75 .04 PDG2 .73 .14 TG .84 .08 FG .75 −.03 Give-and-take PDG1 .69 .01 PDG2 .73 .11 TG .79 .04 FG .75 −.05 Money-making PDG1 −.17 .56 PDG2 .17 .70 TG −.09 .74 FG .10 .70 Business PDG1 .27 .52 PDG2 .19 .58 TG −.01 .61 FG .20 .61 Taking advantage of others PDG1 −.04 .37 PDG2 .13 .53 TG −.01 .65 FG .05 .68 Communality 6.98 4.59 Table options Table 7 shows the correlations of these two types of perception with the participants’ game behaviors. The collaborative situation scale was found to be correlated with game behaviors, but the business transaction scale was not. Table 7 also shows the correlation between these two types of perception with the general measure of SVO, which were weak and/or non-significant. This lack of correlation suggests that perceptions of the game situation are relatively independent of each participant’s social value orientation. Table 7. Correlations between the two scales of the perception of the game situation and game behaviors. PDG1 PDG2 tTG rTG DG tFG gPS gEXP gSVO Collaborative situation .30⁎⁎ .33⁎⁎ .28⁎⁎ .23⁎ .01 .27⁎ .39⁎⁎ .31⁎⁎ .17 Business transaction −.16 −.17 −.06 −.19 −.12 −.05 −.13 −.11 −.02 Note. Only those who preferred to be an allocator. ⁎ p < .05. ⁎⁎ p < .01. Table options Perception of the situation and SVO Next, we conducted a regression analysis on each game behavior to examine the relative power of situational perception and social value orientation as determinants of game behavior. We did not include perception of the situation as a business transaction in this analysis because it was previously found not to be correlated with game behavior. The standardized regression coefficients are shown in Table 8. The effect of having a pro-social SVO was found to be approximately as strong as that of perceiving the game as a collaborative situation, except in the DG. Given this lack of correlation, an individual’s social value orientation and game perceptions appear to exert independent and roughly equal influences on his/her game behavior, at least with respect to the games assessed in the present study. Table 8. The effects (standardized regression coefficients) of the two scales of perception of the game situations. DV: Behavior Game behavior PDG1 PDG2 tTG rTG DG tFG gPS gEXP Collaborative situation 0.25⁎ 0.27⁎ 0.33⁎⁎ 0.17 0.02 0.24⁎ 0.33⁎⁎ 0.26⁎ Pro-social SVO 0.29⁎⁎ 0.30⁎⁎ 0.25⁎ 0.31⁎⁎ 0.07 0.27⁎ 0.33⁎⁎ 0.27⁎ R2 0.18 0.19 0.20 0.14 0.01 0.15 0.26 0.17 ***p < .001. ⁎ p < .05. ⁎⁎ p < .01. Table options Allocation behavior in the dictator game One of the most surprising findings shown in Table 3 and Table 4 is the lack of correlation between behavior and expectation in the DG with those in the other games studied. As discussed earlier, the dictators can freely determine the game’s outcome, such that their decisions should strongly reflect their outcome preferences. This fact implies that the dictator’s decision should be correlated at least moderately with their behavior in other games that reflect the player’s pro-social preference, such as cooperation in the PDG and reciprocation of trust in the TG. Strangely, this prediction was not supported. Furthermore, neither the dictator’s behavior nor his/her expectation of the other dictator’s behavior was correlated with the general measure of each individual’s SVO. Because this finding was unexpected, we performed an additional analysis to determine how this pattern emerged. In this post hoc analysis, we found that the correlation between allocation choice in the DG and other game behaviors varied dramatically depending on the participant’s response to one of the post-experimental questions. This question read: “The role assignment into allocator and recipient was actually determined by a lottery. If you had a choice, would you want to play the role of an allocator or the role of a recipient?” Sixty-four percent (61/95) of the participants preferred the role of the allocator, while the remaining 36% (34/95) preferred the role of the recipient. Table 9 shows the correlations of the allocation choice in the DG with other game behaviors among those who preferred to be allocators and those who preferred to be recipients. Among those who preferred to be allocators, DG behavior was positively correlated with other game behaviors. Furthermore, DG behavior was positively correlated with gSVO among those who preferred to be dictators. These results indicate that dictators’ choices reflect their pro-social SVO in the expected manner. The more unexpected result was the significant and negative correlations among those who preferred to be recipients. The correlations were all in the opposite direction of the theoretical predictions – those who gave more to their partners in the DG actually cooperated less in the PDG, trusted less in the TG and FG, and returned less of the entrusted money in the TG. Furthermore, the correlation with the gSVO was also negative. Interestingly, the differences between these two types of participants did not strongly affect the observed correlations between the other game behaviors. Table 9. Correlations of DG behavior with the other game behaviors among those who preferred to be allocators or recipients. Preference to be PDG1 PDG2 tTG rTG tFG gPS gEXP gSVO Allocator (n = 61) .43⁎⁎⁎ .34⁎ .34⁎⁎ .30⁎ .46⁎⁎⁎ .39⁎⁎ .32⁎ .46⁎⁎⁎ Recipient (n = 34) −.45⁎⁎ −.64⁎⁎⁎ −.28 −.40⁎ −.47⁎ −.70⁎⁎ −.62⁎⁎ −.45⁎ Recipients excluding the extra-fair (n = 28) −.05 −.39 .31 .28 −.02 −.39 −.20 −.07 Note. The last row provides the correlations among those who preferred to be recipients after excluding ultra-fair participants, who gave more than half of their endowment of 2500 yen to the recipient (n = 4 among those who wanted to be a dictator, and n = 6 among those who wanted to be a recipient). ⁎ p < .05. ⁎⁎ p < .01. ⁎⁎⁎ p < .001. Table options The above findings suggest that the allocation decisions of those who preferred to be recipients were not consistent with their social value orientations. While those who preferred recipient status allocated more to their partnered recipients (M = 1230.88 yen, SD = 615.34) than those who preferred dictator status (M = 1020.51 yen, SD = 455.01; t(53.4) = 1.75, p < .09), they did not do so according to their preferences. Conformity to the norm may be one possible explanation for such allocation choices by those who wanted to be recipients. Those who felt a strong normative pressure to behave in a fair manner (or even an ultra-fair manner, giving more than half to their paired recipient) wanted to be a recipient because they disliked feeling as though they were the target of such normative pressure. Actually, six of 34 participants who preferred to be recipients allocated more than half of the 2500 yen to their recipients, and removing those from the analysis substantially reduced the correlations with other game behaviors (see the last row of Table 9). Susceptibility of DG allocation behavior to normative considerations has been previously suggested by List (2007), who demonstrated the effects of framing on this game. Similarly, Haley and Fessler, 2005 and Rigdon et al., 2009 and Mifune, Hashimoto, and Yamagishi (2010) demonstrated that the DG allocation amount is affected by the simple presence of an eye-like picture, suggesting that the idea of being ‘watched’ for normative behavior could effectively prevent participants from taking on the role of norm violator. The finding described above suggests that this susceptibility to normative concerns is more pronounced in the dictator game than in the other games because the allocator is solely responsible for the outcome (cf., Kiyonari et al., 2006).