Does she compensate the victim while he punishes the perpetrator? No gender differences in anonymous economic games across 11 nations

Dorrough, Angela R.; Olsson, Maria I.T.; Froehlich, Laura; Glöckner, Andreas; Martiny, Sarah E. Article — Published Version Does she compensate the victim while he punishes the perpetrator? No gender differences in anonymous economic games across 11 nations Journal of Behavioral Decision Making Provided in Cooperation with: John Wiley & Sons Suggested Citation: Dorrough, Angela R.; Olsson, Maria I.T.; Froehlich, Laura; Glöckner, Andreas; Martiny, Sarah E. (2021) : Does she compensate the victim while he punishes the perpetrator? No gender differences in anonymous economic games across 11 nations, Journal of Behavioral Decision Making, ISSN 1099-0771, Wiley, Hoboken, NJ, Vol. 34, Iss. 2, pp. 261-274, https://doi.org/10.1002/bdm.2208 This Version is available at: https://hdl.handle.net/10419/230289 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by-nc-nd/4.0/ RESEARCH ARTICLE Does she compensate the victim while he punishes the perpetrator? No gender differences in anonymous economic games across 11 nations Angela R. Dorrough 1 | Maria I.T. Olsson 2 | Laura Froehlich 3 | Andreas Glöckner 1,4 | Sarah E. Martiny 2 1 Social Cognition Center Cologne, Department of Psychology, University of Cologne, Cologne, Germany 2 Research Group Social Psychology, Department of Psychology, UiT The Arctic University of Norway, Tromsø, Norway 3 Department of Social Psychology, Faculty of Psychology, FernUniversität in Hagen, Hagen, Germany 4 Max Planck Institute for Research on Collective Goods, Bonn, Germany Correspondence Angela R. Dorrough, Social Cognition Center Cologne, Department of Psychology, University of Cologne, Richard-Strauss-Str. 2, Cologne 50931, Germany. Email: [email protected] Funding information University of Hagen Abstract Social role theory posits that occupational gender roles give rise to gender differences in behavior, such that men and women engage in qualitatively different prosocial behaviors. Therefore, we expected that women who observed an unfair situation (involving a victim and a perpetrator) would respond by demonstrating communal prosocial behavior (by compensating the victim), whereas men would respond with agentic prosocial behavior (by punishing the perpetrator). Furthermore, on the basis of social role theory, we expected that gender differences would be more pronounced in countries with a more unequal distribution of men and women in communal and agentic occupational roles. The current research tested the predictions using an economic game. Two studies consisting of samples from 10 countries (Study 1, N= 1,791) and a student sample from Germany (Study 2, N= 193) showed no support for the predicted gender differences in prosocial behavior and no systematic relationship between prosocial behavior and gender roles across countries. KEYWORDS agentic, communal, gender differences, helping, prosocial behavior, social role theory 1|INTRODUCTION Prosocial behavior is associated with helping, sharing, comforting, rescuing, and defending another individual, but it can also involve supporting the collective, group, or nation (Eagly, 2009). Previous research shows that men and women differ in the degree to which they engage in prosocial behavior. However, the direction of this gender difference is less clear. For instance, some research demonstrates that females are more prosocial than males (e.g., Carlo, Roesch, Knight, & Koller, 2001; Russell, Hart, Robinson, & Olsen, 2003; see also the meta-analysis by Fabes & Eisenberg, 1998), whereas other studies have shown that males are more prosocial than females (e.g., Eagly & Crowley, 1986). In the present research, we investigate whether women and men exhibit different kinds of prosocial behavior. Received: 26 February 2020 Revised: 10 August 2020 Accepted: 10 August 2020 DOI: 10.1002/bdm.2208 This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. © 2020 The Authors. Journal of Behavioral Decision Making published by John Wiley & Sons Ltd J Behav Dec Making. 2021;34:261–274. wileyonlinelibrary.com/journal/bdm 261 1.1 |Gender differences in prosocial behavior We argue that there is no simple answer to the question of which gender is more prosocial. Instead, on the basis of social role theory, we argue that men and women engage in different kinds of prosocial behavior (Eagly, 2009). According to the theory, behavioral gender differences are driven by an unequal distribution of men and women in different social roles (e.g., Eagly, Wood, & Diekman, 2000), which shape expectations regarding gender-typed behavior. Across the world, women more frequently occupy communal, caring-oriented social roles, whereas men more frequently occupy agentic, achievement-oriented social roles (Kan, Sullivan, & Gershuny, 2011). When observing men and women in these roles, people infer that they have specific traits (i.e., communal traits on the part of women vs. agentic traits on the part of men) that qualify them for these roles. As men and women are internally motivated to meet these expectations (Greenwald et al., 2002) and avoid backlash for engaging in gender stereotype-incongruent behavior, men are more likely to engage in agentic (prosocial) behaviors (e.g., behavior that can increase the status of the helper), whereas women are more likely to engage in communal (prosocial) behaviors (e.g., empathy-related behavior; Eagly, 2009). 1.1.1 |Gender differences using social psychological paradigms To investigate this assumption, Eagly (2009) reviewed research on gender differences in prosocial behavior in a range of social situations that involved interacting with strangers, romantic partners, family members, and work colleagues and concluded that men and women are not more or less prosocial. Instead, Eagly noted that women seem more likely to engage in communal (i.e., relational, friendly, compassionate, and comforting) prosocial behaviors, whereas men seem more likely to engage in agentic (i.e., dominant, assertive, aggressive, and risky) prosocial behaviors. For example, women show more prosocial behavior in dyadic relationships: they are more likely to empathize and engage in self-disclosure than men (e.g., Rose & Rudolph, 2006) and are more likely to provide emotional support in marriages (Neff & Karney, 2005). Men, on the other hand, are more likely to offer help to strangers (Johnson et al., 1989) and to help in the presence of others (e.g., De Caroli & Sagone, 2013). 1.1.2 |Gender differences using economic paradigms Reviewing the economic game literature, which uses money transfers as a measure of prosocial behavior, results in the same conclusion; that is, a more differentiated perspective on gender differences is necessary. Balliet, Li, Macfarlan, and Van Vugt (2011) conducted a metaanalysis of 272 effect sizes for gender differences in the amount men and women transferred to an interaction partner. They found no overall gender differences but concluded that gender differences in prosocial (transfer) behavior were game and context dependent. Indeed, a closer look into the literature reveals that gender differences in economic game behavior seem to hinge on a specific game feature: Women usually act more prosocially in games requiring an altruistic motive for transfer behavior, whereas men transfer more money to their interaction partners when the game involves an element of risk. For example, in dictator games, in which the player assigned to be a“dictator”can decide whether they wish to selflessly share their experimental endowment with a passive player, women often transfer more money to the interaction partner than men do (e.g., Eckel & Grossman, 1998; Engel, 2011). However, in economic games (e.g., trust games, Romano, Balliet, Yamagishi, & Liu, 2017; prisoner's dilemma games, Dorrough & Glöckner, 2019) that involve the risk of exploitation (where the player transfers money and the interaction partner does not), men typically transfer more money to interaction partners than women do (for an overview regarding the different motives in economic games, see Thielmann, Böhm, & Hilbig, 2015). In sum, previous research using social psychological as well as economic paradigms suggests that men and women are not more or less prosocial but rather prosocial in different ways. The assertion that men and women engage in different kinds of prosocial behavior (in line with gender role expectations), however, has only been inferred from comparing results between studies using different research designs (Eagly, 2009). A direct experimental test of this assertion within one study is still missing. The current project aims to fill this gap to get a better understanding of gender differences in prosocial behavior and to bring findings from previous research together. Taken together, based on the theoretical and empirical considerations above, it can be predicted that men are more likely to engage in agentic forms of prosocial behavior, whereas women are more likely to engage in communal forms of prosocial behavior. 1.1.3 |Gender differences in prosocial behavior across countries In addition to investigating whether men and women engage in different kinds of prosocial behavior, our research also tests whether gender differences in prosocial behavior vary between countries. According to social role theory, observing men and women in different (occupational) roles shapes gender stereotypes, which, when internalized, gives rise to gender differences in behavior in the direction of the stereotype (e.g., Eagly & Wood, 2012). As countries differ in occupational gender segregation, we should observe larger gender differences in behavior in countries with more occupational gender segregation (Eagly & Wood, 1999). Previous cross-national research on mating preferences and gender equality supports this assumption: Gender differences in preferences for stereotype-congruent partners were less pronounced in countries that ranked high (rather than low) on a gender equality index (Eagly & Wood, 1999). In line with social role theory and previous findings, it can be predicted that gender 262 DORROUGH ET AL. differences in prosocial behavior increase as gender segregation in occupational roles increases. However, it has to be noted that results from some previous multinational studies are contrary to this assumption. For example, gender differences in preferences (e.g., social and risk preferences; Falk & Hermle, 2018) and basic values (e.g., power, hedonism; Schwartz & Rubel-Lifschitz, 2009) were more pronounced in countries that ranked high (rather than low) on gender equality. Moreover, research has shown variation in the size of gender differences in prosocial behavior (in a prisoner's dilemma game) across different countries, without finding a systematic relationship between gender differences and gender equality (as measured by the Gender Inequality Index [GII]; http://hdr.undp.org/en/content/ gender-inequality-index-gii; Dorrough & Glöckner, 2019). Thus, it seems especially important to shed light on the question of whether or not gender segregation in the occupational domain influences gender differences in prosocial behavior. 1.2 |The present research The present research composes of two studies. In both Studies 1 and 2, we use an economic game to experimentally investigate whether men and women engage in different kinds of prosocial behavior. Furthermore, in Study 1, we test whether gender differences in prosociality are influenced by a country's level of gender equality. 1.2.1 |Compensation/punishment game One way to measure prosocial behavior in a standardized and controlled way is through economic games. Economic games are widely used by economists and psychologists in mononational as well as multinational studies (e.g., Dorrough & Glöckner, 2016; Romano et al., 2017). One advantage of using economic games over self-report is that economic games capture actual incentivized behaviors rather than mere intentions that might be biased by social desirability (see Camerer, 2003; Thielmann, Heck, & Hilbig, 2016). Economic games model complex social interactions by concentrating on a few central dependent variables and actual behavior. Games are conducted anonymously and provide a consistent study environment for all participants. Thus, in contrast to field studies or other types of experiments, economic games facilitate control over endogenous and exogenous factors (Murnighan & Wang, 2016), ensuring internal validity. Economic games are particularly suitable for multinational research as they provide only minimal context information, reducing the threat of poor cross-cultural validity. Furthermore, the anonymity of the game interactions rules out potential confounds, such as the extent to which participants perceive interaction partners to be attractive or similar to themselves. In economic games, participants receive an initial endowment and are required to decide how much of that endowment they would like to transfer to an interaction partner. The present research employs a third-party compensation/punishment game (e.g., Leliveld, Van Dijk, & Van Beest, 2012; Weng, Fox, Hessenthaler, Stodola, & Davidson, 2015), which allows for two potential prosocial behaviors. Specifically, participants observe that a person receives an unfair monetary transfer from a third player. They can then use their own monetary resources to compensate the victim (i.e., the player who has received an unfair transfer) in the form of a money transfer or to punish the perpetrator (i.e., the player who has made the unfair transfer) by reducing their payoff. As the participant has to sacrifice some of their own endowment without financial benefit to compensate and punish, both behaviors are forms of prosocial behavior (i.e., altruistic punishment and altruistic compensation; Fehr & Fischbacher, 2003). Communal behaviors involve empathy, caring for others, sociability, and affection (e.g., Abele, Uchronski, & Suitner, 2008). We consider compensation transfers to be a display of communal prosocial behavior as these transfers are relational and comfort the victim. In support of this claim, Weng et al. (2015) found in a compensation/punishment game that empathic concerns promoted compensation of a victim but not punishment of a perpetrator. Agency is commonly associated with aggression, protectiveness, and striving for power and status (e.g., Abele et al., 2008). Research suggests that punishment transfers are a display of agentic prosocial behavior. First, punishment is driven by spontaneous anger reactions (Mischkowski, Glöckner, & Lewisch, 2018) and is often perceived as an act of aggression by the punished person, leading to acts of revenge (e.g., counter-punishment; Nikiforakis, 2008). Second, punishment may result in behavioral change on the part of the perpetrator (e.g., Glöckner, Kube, & Nicklisch, 2018), which prevents the perpetrator from continuing to victimize others. As such, punishment may be an act of protection because it communicates to perpetrators that their behavior violates a norm, is not acceptable, and should be changed. Third, evidence indicates that punishment is related to power and dominance: Participants who see themselves as more powerful engage in more punishing transfers in different economic games (Chierchia, Parianen Lesemann, Snower, Vogel, & Singer, 2017). Further work demonstrates that participants make more punishment transfers in economic games with unstable power hierarchies (Dorrough, Glöckner, & Lee, 2017). These results show that the relevant dimensions of agency are activated not only in direct interpersonal settings with long-term interactions but also in anonymous economic games with short-term interactions. This might partially be due to individuals overgeneralizing from social contexts to such games (Rand et al., 2014). In summary, in the present research, we consider compensation a communal type of prosocial behavior and punishment an agentic type of prosocial behavior. Our paradigm minimized additional motives for third-party transfers such as demand characteristics and reputation by ensuring that interactions were anonymous and that participants interacted with their interaction partners only once and were not directly affected by the unfair transfer. To the best of our knowledge, our studies are the first to investigate gender differences in prosociality by providing male and female participants with a communal and an agentic behavioral option. Applying the above-mentioned reasoning to game behavior, we hypothesize that DORROUGH ET AL.263 men will make more punishment transfers (H1a), whereas women will make more compensation transfers (H1b). We additionally hypothesize that gender differences in compensation and punishment transfers will increase as gender segregation in occupational roles increases (H2). 2|OVERVIEW OF STUDIES The present research consists of two studies. In Study 1, participants from 10 countries took part in the compensation/punishment game. During the game, participants (who had been allocated to the role of “Player E”) indicated decisions for transfers of Talers, which was the experimental currency, to other players. Participants made a decision after being informed about a fair transfer (where a player had transferred 50 out of 100 Talers1 to another player) and an unfair transfer (where a player had transferred 0 out of 100 Talers to another player). For each transfer, participants indicated how they would like to invest their own endowment (50 Talers). Although participants were asked to indicate decisions for transfers in response to fair and unfair behavior, only the behavioral response for the unfair treatment is relevant to our research questions, as we were only interested in prosocial behavior (helping a person in need) elicited by the unfair condition. Thus, only data from this scenario will be analyzed.2 Participants could choose to spend their endowment by (a) keeping all of their Talers to themselves, (b) transferring some or all of their Talers in the form of punishment to the perpetrator (i.e., the player who behaved unfairly), and/or (c) transferring some or all of their Talers in the form of compensation to the victim (i.e., the player who was treated unfairly). Participants were informed that Talers transferred as punishment would reduce the perpetrator's experimental payoff, whereas Talers transferred as compensation would increase the victim's experimental payoff. To test our assumption that gender differences would increase with (perceived) occupational gender segregation, participants indicated to what extent a range of agentic and communal occupations were segregated by gender (see below). To ensure the generalizability of our findings, Study 1 was conducted not only in Western, educated, industrialized, rich, and democratic (WEIRD; Henrich, Heine, & Norenzayan, 2010) but also in nonWEIRD countries such as Chile and Indonesia. Study 2 serves as a conceptual replication of Study 1 but includes a more convenient student sample from a single country (Germany). Our hypotheses were preregistered on the Open Science Framework3 (Study 1, https://osf. io/q5a4v/ Study 2, https://osf.io/p5vyk/). All data, analysis scripts, and material files (English version) can be accessed at https://osf.io/ re7n3/. All translations of the scales can be accessed4 at https://osf. io/7ybns/. Study 1 was approved by the ethics committee of the University of Göttingen and was conducted in accordance with the ethical guidelines of the German Psychological Association (DGP) and the American Psychological Association (APA). Study 2 was a conceptual replication of Study 1 with only minor changes to the measures and procedure. 3|STUDY 1: SAMPLES FROM 10 COUNTRIES 3.1 |Method 3.1.1 |Participants and design Data were collected in the summer of 2018. We employed a 10 (country) ×2 (participant gender: male vs. female) ×2 (victim gender: male vs. female) mixed design with victim gender as a within-participant factor. Participants were recruited via an online panel provider (Toluna: https://de.toluna.com/). Our second hypothesis required sufficient variance in gender segregation in occupational roles, so we selected countries that varied in their rankings on the GII (http://hdr. undp.org/en/content/gender-inequality-index-gii), which is partly determined by the proportion of women in the labor force. We used data from 2017, which were the most recent data at the time of study planning. For pragmatic reasons, we did not consider countries with small Toluna panels (<50,000) or more than one official language (e.g., Paraguay and Belgium). This procedure resulted in the selection of the following countries (sorted by increasing gender equality): Indonesia (GII rank 104 of 160), Colombia (rank 87), Mexico (rank 76), Chile (rank 72), Russia (rank 53), the United States (rank 41), China (rank 36), Japan (rank 22), Spain (rank 15), and Sweden (rank 3). In each country, the sample was recruited to be representative of the general population in terms of age and gender. Study 1 is part of a larger project that collected data over two time points (the data analyzed here were measured at the second time point/Part 2).5 The a priori calculated sample size (n= 208 per country) was based on the design for the superordinate project using an effect size estimate of d= 0.14 for gender differences in game behavior (see Dorrough & Glöckner, 2019). To be able to detect small effects, the required sample size was estimated at 200 participants (for Players E, i.e., those who indicated decisions for compensation and punishment transfers) with a desired power of 80%. This power analysis was conducted with the use of G*Power (Faul, Erdfelder, Lang, & Buchner, 2007). For the analysis, we assumed a repeated measures mixed analysis of variance (ANOVA) as the closest pragmatic approximation for the cluster-corrected regression analysis we intended to run, as no convenient method for power estimation of the latter analysis was available. We aimed to achieve this sample size in each country. However, because of participant dropout, some countries have lower sample sizes. Thus, results for individual countries must be interpreted with caution. For the analyses, and as registered a priori on the Open Science Framework, we included only participants who (a) completed both Parts 1 and 2, (b) entered a valid participant code, and (c) indicated a country of origin that corresponded to their country selection when registering with the panel provider. From the final data set, we excluded participants who indicated different countries of origin in Parts 1 and 2 or who indicated a gender other than male or female (n= 3). In addition, we excluded participants (n= 112) who were (randomly) assigned to the role of “perpetrator”(Player C) or “victim” 264 DORROUGH ET AL. (Player D) rather than the role of interest (i.e., “observer”; Player E). The final Player E sample is listed in Table 1. It has to be noted that because of exclusion based on the above-mentioned criteria and because of dropouts between Parts 1 and 2, some national subsamples (for Players E) are smaller than originally planned. Sensitivity analyses assuming repeated measures mixed ANOVAs (see above) for the different countries show that the effects that can be detected with the available national samples (with a power of .80; alpha = .05) range from d= 0.14 (Russia; n= 216) to d= 0.21 (the United States; n= 114). It took participants approximately 45 min to complete Parts 1 and 2. In addition to receiving a basic payment, participants received a bonus payment that ranged from the equivalent of 0–4 US dollars and that was based on one randomly selected incentivized task from Part 1 or 2. For example, the bonus payment could be based on one decision in the compensation/punishment game by the participants themselves (if they had been allocated to the role of Player E or C) and/or their interaction partner(s). Participants' payment was first calculated in “Talers,”which was converted to the respective national currency for payout. The basic payment was credited to participants' Toluna accounts directly after they completed the study. The bonus payment, which depended on the decision of the participant and/or their interaction partner(s), was calculated after study completion in that we matched participants in groups of three according to their player roles (one Player C, one Player D, and one Player E). The bonus payment for these three players was determined by Player C's transfer and the compensation and punishment decision of Player E for this transfer. The bonus payment was credited to the participants' Toluna accounts by the Toluna project manager. The study was translated from English into the language of each subsample by a professional translation agency (https://www.e-kern. com/). Thereafter, researchers in psychology (who were fluent in the language of each translated survey) checked the translations against the English version. If necessary, the surveys were sent back to the translation company that considered the language edits made by the researchers. The final versions were subsequently checked by a native speaker; again, if necessary, minor changes were made following their feedback. 3.1.2 |Materials and procedure Participants first read information about data protection and details of the study (i.e., duration and payment) and then provided informed consent. At the beginning of the study (both in Parts 1 and 2), participants were informed about the calculation of their bonus payment. Participants learned that they would receive a bonus payment based on their answers and/or the answers of other participants in a randomly selected task. Gender segregation in occupational roles Participants reported the proportion of men across a range of communal and agentic occupations in their country: “In your country, what proportion (%) of individuals who work as [occupation] are male?”On the basis of the previous research (Koenig & Eagly, 2014), we assessed the perceived proportion of men across the following occupations: nursery teachers, geriatric aides, nurses, secretaries, and therapists (communal occupations) as well as police officers, attorneys, CEOs, rescue service workers, soldiers, politicians, and judges (agentic occupations). Compensation/punishment game Second, participants took part in the compensation/punishment game. Participants were randomly assigned to one of three different player roles—C, D, or E. Only the decisions of Player E were relevant to the hypothesis testing and analyzed. Even though data from Players C and D were not required to test our hypotheses, a minority of participants were assigned to those roles (n= 112). Doing this allowed us to calculate participants' bonus payments based on participants' actual behaviors and thereby avoid deception in terms of how the bonus payment was calculated.6 Player C had to decide how much of their endowment of 100 Talers they wanted to transfer to the recipient (Player D). The amount not transferred was kept by Player C. Player C was reminded that the amount they kept might form the basis of their bonus payment. Player C could transfer either 50 or 0 Talers to the recipient (Player D). Player D had a passive role and could not make a decision in the compensation/punishment game. Player E was informed that they could transfer Talers as compensation to Person D (which would increase Player D's payoff) and as punishment to Person C (which would reduce Player C's payoff). Player E was asked to make transfer decisions for two different scenarios (using a variant of the strategy method): In the first scenario, Player E was asked how to invest their Talers if Player C makes a fair transfer and sends 50 Talers to Player D (“How much of your Talers do you want to transfer to Person C and Person D if Person C keeps 50 Talers for herself [himself] and transfers 50 Talers to Person D?”). In the second scenario, Player E was asked how to invest their Talers if Player C makes an unfair transfer and sends 0 Talers to Player D (“How much of your Talers do you want to transfer to Person C and Person D if Person C TABLE 1 Subsamples in Study 1 Country NAge range in years % female Chile 142 18–82 46 China 163 20–75 45 Colombia 184 18–71 52 Indonesia 174 18–69 56 Japan 204 20–81 50 Mexico 194 18–75 51 Russia 216 19–77 58 Spain 198 18–78 49 Sweden 202 18–86 53 United States 114 19–86 54 Note: This table shows the subsamples that entered our analyses. That is, only Role E players are depicted in this table and considered in our power considerations and sensitivity analyses. DORROUGH ET AL.265 keeps 100 Talers for herself [himself] and transfers 0 Talers to Person D?”). Only the latter (unfair) transfer was examined in our analyses. For each of the two decisions, Player E received an endowment of 50 Talers and was asked to decide how to invest them. Player E could divide their Talers by (a) keeping all of their Talers, (b) transferring Talers in the form of punishment to Player C, and/or (c) transferring Talers in the form of compensation to Person D. To make prosocial behavior more profitable, Player E was informed that the Talers they transferred as either punishment or compensation would be tripled by the experimenter (for a similar approach, see Leliveld et al., 2012). To avoid confounds because of cultural differences in how much men and women interact with members of the opposite gender in their daily lives, participants were informed that they were interacting with players of the same gender. For exploratory analyses, we added an additional round of the game in which participants were informed that they were interacting with a victim (Player D) of the opposite gender. Following the procedure of previous multinational research (Romano et al., 2017), we limited the amount of written information and used pictorial instructions to illustrate the rules of the game (see Figure 1 for an example). Participants were able to start the game once they had correctly answered three comprehension questions (e.g., “Person C transfers 0 Talers to Person D. How many Talers do you have to transfer to Player D for them to receive 30 Talers as compensation?”). At the end of the study, participants were provided with contact information that they could use to request additional information about the study. 3.2 |Results As preregistered, for our main analyses, we only included data from same-gender interactions (in which the gender of Player D matched the gender of Player E; see reasoning for that decision above). Overall, the descriptive results did not indicate that men and women engage in different kinds of prosocial behavior (Figure 2). In fact, in all countries, men and women used punishment and compensation in a similar manner (Figure S1) in that both men and women invested more Talers in compensation (men: M= 14.15, SD = 12.33; women: M= 13.22, SD = 11.35) than in punishment (men: M= 9.05, SD = 10.34; women: M= 8.61, SD = 9.48; see Table S1 for country level results). These findings speak against our assumption of gender-typed prosociality. In order to determine the statistical significance of gender differences (or lack thereof) across different prosocial behaviors, we ran an ordinary least squares (OLS) regression analysis with cluster-corrected standard errors at the participant level to account for dependencies in error terms. Specifically, we predicted transfer behavior by prosocial dimension (punishment vs. compensation) and participant gender controlling for country differences by including country dummies. This analysis did not reveal the hypothesized interaction between participant gender and prosocial dimension (Table 2, Model 1).7 Predicting transfer behavior for the different dimensions separately (punishment in Model 2; compensation in Model 3) to test H1a and H1b more directly did not reveal significant effects of participant gender.8 When including data from interaction partners of the opposite gender and controlling for whether Player D had the same gender as Player E (yes vs. no) in an exploratory analysis, the interaction between participant gender and prosocial dimension remained nonsignificant (b=−0.49, t(1790) = −0.80, p= .423). The three-way interaction was also nonsignificant (b= 0.08, t(1790) = 0.11, p= .914). Finally, analyses per country showed no significant interaction effect between participant gender and prosocial dimension in any of the 10 countries (all p> .232). All these analyses speak for the robustness of the null findings. We predicted that gender differences in communal and agentic prosocial behavior would increase as people perceived more gender segregation in communal and agentic occupations (H2). We calculated an index of gender segregation of labor by subtracting the perceived proportion of men in communal occupations from the perceived proportion of men in agentic occupations. The index could vary from 100 (100% men in agentic professions and 0% men in communal professions) to −100 (0% men in agentic professions and 100% men in communal professions). In line with Koenig and Eagly (2014), the index of gender segregation of labor was positive (M= 24.73, SD = 23.37), suggesting that participants perceived men as more likely to hold agentic occupations than communal occupations. On average, participants indicated that 68% (43%) of people working in agentic (communal) professions were men. However, as expected, we found FIGURE 1 This figure presents an example illustration describing the punishment rule of the compensation/punishment game (i.e., if Player E transfers 10 Talers in the form of punishment to Player C, 30 Talers are deducted from Player C's account) 266 DORROUGH ET AL. that the perceived segregation of labor varied across countries, from M= 18.40 in Japan to M= 31.43 in Russia, which suggests that occupations had a more equal distribution of women and men in some countries than in others (for more details, see Froehlich, Olsson, Dorrough, & Martiny, 2020). To test H2, we aggregated the division of labor index at the country level. We then ran a multilevel regression analysis predicting prosocial behavior by dimension (L1), participant gender (L1), the interaction of dimension and participant gender, as well as the cross-level interaction of the division of labor index (L2) with the other predictors (group-mean centering used for Level 1 predictors, grand-mean centering used for the Level 2 predictor; see Enders & Tofighi, 2007). The degree of between-participant and between-country variability in the mean level of prosocial behavior is captured by the Level 1 (SD intercept = 1.32, 95% CI [0.28, 6.28]) and Level 2 (SD intercept = 0.28, 95% CI [0.03, 3.01]) random parameter estimates of the intercept, respectively. The Level 2 random parameter estimates of the division of labor index denote whether there is between-country variability in gender differences in compensation and punishment transfers (SD slope = 0.00, 95% CI [0.00, 2.83]). As with the results reported above, we did not observe a simple effect for participant gender, b= 0.43, z= 1.18, p= .238, but we did find a significant effect of prosocial dimension on transfer, b=−4.45, z=−12.57, p< .001. Contrary to H2, we did not observe the hypothesized cross-level interaction, b=−0.03, z=−0.15, p= .882. We ran an exploratory analysis where we replaced the gender segregation of labor index with other indicators of gender equality. However, the FIGURE 2 Prosocial behavior measured as amount transferred by dimension (punishment vs. compensation) and participant gender in Study 1. Spikes represent 95% confidence intervals based on cluster-corrected standard errors. The pvalue refers to the interaction between participant gender and dimension. A graphical illustration for each country can be found in Figure S1) [Colour figure can be viewed at wileyonlinelibrary.com] TABLE 2 Transfer behavior in Study 1 (1) (2) (3) Transfer 95% CI Transfer 95% CI Transfer 95% CI Participant gender (female = 0; male = 1) 0.650 (1.18) [−0.43, 1.73] 0.194 (0.42) [−0.71, 1.10] 0.662 (1.20) [−0.42, 1.75] Dimension (compensation = 0; punishment = 1) −4.236 *** (−9.12) [−5.15, −3.32] Participant gender * dimension −0.445 (−0.62) [−1.84, 0.95] Constant 13.19 *** (35.87) [12.47, 13.91] 8.975 *** (28.31) [8.35, 9.60] 13.17 *** (35.10) [12.43, 13.90] Observations 3,582 1,791 1,791 Participants/clusters 1,791 1,791 1,791 Adjusted R 2 0.042 0.005 0.000 Note: Transfer behavior is predicted by participant gender, dimension, and their interaction (Model 1) as well as by gender for both dimensions separately (punishment in Model 2 and compensation in Model 3). Country dummies are not reported. The table contains unstandardized coefficients. tstatistics are in parentheses. *** p< .001. DORROUGH ET AL.267 conclusions regarding the interaction term remain unchanged when using the 2018 GII (b=−4.66, z=−0.90, p= .368) and with women's labor force participation as part of the GII (b= 0.05, z= 0.39, p= .700). 3.3 |Discussion Study 1 investigated whether women and men engage in different kinds of prosocial behavior as a function of gender roles in the occupational domain. In sum, data from Study 1 supported neither the prediction of gender-differentiated prosociality (H1a and H1b) nor the prediction of larger gender differences in countries that have more (perceived) gender segregation in communal and agentic occupations (H2). However, we found a substantial simple effect of dimension (see Figure 2 and Table 2) in that participants, irrespective of their gender, preferred to compensate rather than punish. Exploratory analyses revealed that the null findings replicated when the victim was the opposite gender from the participant. One potential limitation of Study 1 is that participants might have perceived the punishment option as less attractive than the compensation option. We observed an unexpected significant effect of dimension, in that participants of both genders used an average of 4 Talers less (of an endowment of 50 Talers; ß=−0.19, p<.001) to punish (i.e., engage in agentic prosocial behavior) than to compensate (i.e., engage in communal prosocial behavior). Thus, we cannot exclude the possibility that a potential gender difference was masked by the fact that the punishment option was less attractive, and therefore, both women and men made little use of it. To rule out this possibility, we adapted our paradigm in two ways to make the punishment option more attractive in Study 2. 4|STUDY 2: STUDENT SAMPLE REPLICATION To test the robustness of the null effect concerning gender differences in communal and agentic prosocial behavior, we conducted a conceptual replication of Study 1 within a single cultural context (Germany). Study 2 included some alterations to the compensation/punishment game designed to increase the likelihood of punishment being used. To reduce comparisons between the compensation and the punishment transfer (which could lead to a preference for the compensation transfer as a more socially acceptable kind of prosocial behavior), participants in Study 2 were asked to make decisions about punishment and compensation transfers in separate rounds. We expected that this procedure would lead to a higher use of the punishment option (for a similar procedure, see Raihani & Bshary, 2015). In addition, the effectiveness of punishment (i.e., the ratio with which punishment points reduced other players' payoffs) was increased in one condition to make this behavioral option more attractive. 4.1 |Method 4.1.1 |Participants and design Data of N= 193 participants (18–64 years of age, 51% female)9 were collected in the on-site laboratory of the University of Cologne in February 2019 using a 2 (participant gender: male vs. female) ×2 (punishment effectiveness: medium vs. high) ×3 (behavioral option: compensation vs. punishment vs. both) ×2 (victim gender: male vs. female) between-within design (the last two factors were varied within participants). The sample size was based on the same power calculations that informed the sample size for Study 1. Participants received a basic payment of 3 Euros and a chocolate bar in exchange for taking part in the study. Participants' bonus payment for the compensation/punishment game was calculated in the same way as in Study 1. Participants could pick up their bonus payment on two predetermined dates at different university locations. The study took approximately 15 min to complete. After completing the study, the Taler income was converted into Euros using a conversion rate of 100 Talers = 6 Euros. On average, participants earned 8.70 Euros (approx. 9.80 US dollars) for participating in the study. 4.1.2 |Materials and procedure Study 2 followed the same procedure as Study 1. The compensation/punishment game in Study 2 was identical to the one in Study 1 except for the changes described below. These changes were implemented to allow for a strong test of gender-differentiated prosocial behavior by increasing the overall use of punishment behavior (in order to reduce potential bias toward compensation behavior). The first change we implemented was separating the punishment and compensation rounds (see Raihani & Bshary, 2015 for a similar procedure) to prevent participants from comparing the relative value of punishment and compensation (and therefore being more likely to opt for the more socially desirable option, which would be compensation). The second change we implemented was to add a second condition with a higher efficiency of punishment to make this behavioral option more attractive. In the medium punishment efficiency condition (Condition 1), participants first played separate compensation and punishment rounds (presented in randomized order). In the compensation round, participants could only keep their Talers and/or use them to compensate the victim. In the punishment round, participants could only keep their Talers and/or use them to punish the perpetrator. Afterwards, participants played a round in which they could use their Talers to keep, punish, and/or compensate (replicating the design of Study 1). Punishment efficiency in all three rounds was the same as in Study 1 (1:3). The second, high punishment efficiency condition (Condition 2),which was implemented to make the punishment transfer more attractive, followed the same procedure with the exception that punishment efficiency was increased to 1:5 (i.e., each punishment Taler reduced the outcome of the other player by a factor of 5). 268 DORROUGH ET AL.