Does stress make us more—or less—prosocial? A systematic review and meta-analysis of the effects of acute stress on prosocial behaviours using economic games

Prosocial behaviour is fundamental for our social togetherness. Yet, how acute stress, a common everyday occurrence, influences our behaviours towards one another is still unclear. In this systematic review and meta-analysis, we aimed to quantitatively investigate the effect of experimentally induced acute stress on prosocial behaviours in economic games. We also probed possible moderators to explain differences in findings. We included 23 studies, 77 individual effects, and 2197 participants in the meta-analysis and found no overall differences between stress and control groups in prosocial behaviours (SMD=-0.06), or costly punishment (SMD=-0.11). There were no moderating effects of stressor type, participants' gender/sex, or the delay from the stressor to the task. However, the potential recipient of the donated money (person vs. charity) and the complexity of the decision did reveal some differences under stress. The results of this meta-analysis suggest that there is currently no clear answer to the question of whether or not stress increases or decreases prosociality. We highlight important open questions and suggest where the field should go next.


Introduction
Humans are a cooperative species, we rely on one another for our health and well-being (Rand and Nowak, 2013;Tomasello and Vaish, 2013). The question of how the experience of acute stress-a ubiquitous experience in daily life-influences our sociality has therefore attracted a great deal of attention. In particular, the last decade has seen an increase in research investigating whether acute stress changes the way we interact with one another, and whether or not it makes us behave more-or less-prosocial. By and large, research findings are mixed, with studies showing strong enhancing as well as detrimental effects of acute stress on prosocial behaviours, while others do not find an effect at all (for two recent discussions see: Faber and Häusser, 2021;von Dawans et al., 2020). In this regard, Faber and Häusser (2021) argue that when under threat (such as under acute stress) conflicting behaviours might be observed. On one hand, stress-induced increases in selfish behaviour (to protect available resources) and on the other hand stress-induced increases in prosocial behaviour as a means for resource-building by affiliating with others (at least in the long-term). While several possible factors have been proposed to explain these inconsistencies, such as differences between men and women (e.g., Prasad et al., 2017;Youssef et al., 2018), or differences in stressor-types (e.g., the difference between physical or social stressors; von Dawans et al., 2018), there has not been a comprehensive review of the state of research, and an attempt to more formally quantify the research findings. Here, we set-out to do both, using a meta-analytic approach. We subsequently investigated possible moderators that could explain the divergent effects and the heterogeneity in findings across studies. Finally, we outline open questions and ideas for future research that can further our understanding of how acute stress influences prosocial behaviours.
Acute stress is a momentary response to a threatened homeostatic balance (Chrousos, 2009). This response results in an orchestrated cascade of actions, engaging both psychological and biological processes to allocate available resources to deal with the situation at hand. This stress induced shift in resource management brings about cognitive and behavioural changes, such as increased physiological arousal, increased vigilance and attention for salient stimuli, as well as less deliberate and more automatic behaviours and decisions at the cost of cognitively demanding control processes (de Kloet, 2014;Hermans et al., 2014;Joëls and Baram, 2009;Wirz et al., 2018). However, it is still unclear how acute stress might impact prosocial behaviours. Two contrasting theoretical accounts propose that acute stress might result in either an increase (e.g., Tend-and-Befriend; Taylor et al., 2000) or a decrease (e. g., Fight-or-Flight;Cannon, 1939) in prosociality. Many research questions have been framed through this dichotomy in behavioural changes (either an increase or a decrease in sociality). However, given that studies have found facilitating effects of acute stress on social behaviour (e.g., Berger et al., 2016;Roelofs et al., 2005;Singer et al., 2017), as well as stress-induced decreases in social behaviours (e.g., Bodenmann et al., 2015;Nitschke et al., 2020b), and some studies even show no effects of acute stress (e.g., Starcke et al., 2011;Tomova et al., 2020), it is unlikely that this dichotomy for either more prosocial or more selfish behaviours-as measured with economic games-holds true in every stressful situation and in all types of interactions. Instead, acute stress will likely result in more idiosyncratic behaviours based on a host of interacting contextual and individual factors (cf. Faber and Häusser, 2021;Thielmann et al., 2020). However, it is still unclear which contextual or individual factors contribute to these changes under stress, and how these may interact.
Here, we aimed to quantify the effects of acute stress on prosociality. Crucially, prosociality can be defined in several ways, and consequently, prosocial behaviour has been measured in multiple ways in the laboratory (Penner et al., 2004;Schroeder and Graziano, 2015). Broadly defined, prosocial behaviours are actions that benefit others, often at a cost to the actor (Thielmann et al., 2020). We focused on economic games, a common method of measuring prosociality in behavioural economics and experimental psychology. By focusing on economic games, we can investigate the effects of acute stress on prosocial behaviours using a relatively homogeneous and standardised set-of experimental paradigms (Thielmann et al., 2021), namely: the Dictator Game, the Trust Game, the Ultimatum Game, and the Prisoner's Dilemma. Importantly, these economic games are the most common ways researchers have investigated the effects of acute stress on prosocial decisions. Furthermore, one of the key advantages of economic games is that they can measure actual behaviour by implementing incentives-such as an additional monetary payout at the end of the experiments. In these paradigms, participants typically play a series of one-shot games-rounds with independent, usually anonymous other players-meant to measure participants' prosocial behaviours. Importantly, economic games based on game theory have optimal decisions, i. e., decisions that maximize the monetary gains of the player. Thus, depending on the type of game there are (monetary) incentives to act more (i.e., non-cooperative) or less (i.e., cooperative) selfish (Rilling and Sanfey, 2011). Contrary to classic game theory (von Neumann, 1928;von Neumann and Morgenstern, 2007), however, it has been suggested that humans are intrinsically prosocial, meaning that they routinely (and automatically) make decisions to forego personal gains in order to benefit others (Engel, 2011;Lengersdorff et al., 2020;Rilling and Sanfey, 2011;Zaki and Mitchell, 2013). In the context of economic games, by deviating from a monetary, and self-benefiting, optimal decision, individuals may act prosocially and thus not maximize their own gains. Importantly, what constitutes an optimal decision varies between the types of games played. See Fig. 1 for an overview of the economic games discussed.
In the Dictator Game (Kahneman et al., 1986), the participant (i.e., the dictator) is endowed with an amount of money and can decide (without fearing any repercussions, due to the one-shot, non-interactive and anonymous set-up of the game) how much to share with another player (recipient), or a cause (e.g., a charity). Since the optimal monetary choice is to keep the entire endowment for oneself, any deviation can be seen as a prosocial act. Typically, the amount of money the participants share is seen as a (somewhat) continuous measure of prosociality, with higher amounts (more giving) equating to more prosociality (see Supplemental Materials for an overview of average behaviours across all studies reported here, for both control and stress conditions).
In the Trust Game (Berg et al., 1995), the first player (i.e., the trustor; Player A) can decide how much of an endowment they want to invest in another player (i.e., the trustee; Player B). The investment is then multiplied by a factor y (where y > 1), meaning the amount the trustee has at their disposal increases from the initial endowment. The trustee can then decide whether or not they want to return the trust offered and share the multiplied investment, or keep everything for themselves. Note that in this game, the investment and reciprocation matrices are often preset, meaning that participants have the option to either trust or not trust (decisions for Player A), or to reciprocate or exploit (decisions for Player B). Importantly, the trustor can decide not to share anything at all, which will result in them securing the initial endowment for themselves. Consequently, trusting another player is risky, because defection by the trustee can result in losses, up to a complete loss of the endowment. However, if the trust is reciprocated, trusting can be worth the risk (Rand, 2016;Rilling and Sanfey, 2011). Furthermore, in a one-shot game, there is no incentive for the trustee to reciprocate and send money back. Any amount returned, can thus be seen as a prosocial can decide (without fearing any repercussions) how much of an endowment they want to share with another player (Player B). B) Trust Game: Player A (investor, trustor) can decide how much of an endowment they want to share with another player (Player B; recipient, trustee). This amount is then multiplied by a factor of y (y > 1). Player B can then decide if they want to return the trust and share a given amount with player A. C) Ultimatum Game: Player A (proposer) can decide to split an endowment x ways with Player B. Player B can then decide to accept or reject that offer; if rejected, neither player receives a payout (i.e., costly punishment). action (and similar to the Dictator Game).
As with the previous two games, the Ultimatum Game (Güth et al., 1982) is a resource allocation game. In the Ultimatum Game the first player (i.e., the proposer; Player A) can decide to split an endowment with another player as they see fit, and offer that amount to the second player. Player two (i.e., the recipient of this offer; Player B) can then decide to accept or reject the proposed offer. By accepting the offer, both players will receive a split of the endowment as proposed by the first player. If rejected, however, neither of the two players will receive any money. Here, rejection is often referred to as costly punishment, since a rejection comes at no monetary benefit to the second player. Hence, the proposer is incentivised to offer a fair split of the endowment, rather than proposing an unfair split, that might be seen as selfish and thus rejected by the second player. Exploitative behaviour is thus limited to less fair offers, rather than plainly unfair offers, as the recipient can reject the offer, and consequently punish the proposer.
Lastly, the classic Prisoner's Dilemma game (Tucker and Straffin, 1983) pits two individuals against each other in a social dilemma; here, individuals have to make a decision whether to cooperate (e.g. stay silent) or 'betray' the other player. If both players 'betray' each other (i.e., defect) then they will receive a greater punishment than if both players do not confess (i.e., stay silent). However, if one player confesses (i.e., implicates the other) and the other cooperates (i.e., stays silent), then the former will be rewarded (e.g., walk free), whereas the latter will be (severely) punished (i.e., an even greater punishment than if both had betrayed each other). Hence, defection behaviour is incentivised as betraying the other player offers a greater reward than cooperating with them-thus, any deviation from this can be seen as prosocial behaviour.
These economic games are intended to model more complex real-life prosocial behaviours in highly standardardized and controlled experimental settings. They are meant to test (private) preferences in behaviours (such as the inclination to act prosocially) (Pisor et al., 2020); however, since these tasks only measure observable behaviours, the underlying cognitive and motivational processes (e.g., why and how people make decisions) are largely unknown, requiring some interpretative leaps when making claims about the selfishness or prosociality of participants. Despite these limitations of economic games, so far they have been the mostly widely used approach for measuring prosocial tendencies under acute stress. Therefore, we focused on these in the current study (see 'Questions for future research' for alternative ways of measuring prosocial behaviour under stress).

Moderators
In the context of acute stress, both enhancing as well as detrimental effects on prosocial behaviours have been reported. Given this divergence in findings, several moderating factors have been proposed and investigated. Where possible, we aimed to systematically investigate these potential moderators in this meta-analysis.

Gender/Sex
Theoretical accounts have suggested that men and women might differ in their social-behavioural responses to an acute stressor. Specifically, Taylor et al. (2000 however, Geary andFlinn, 2001;Taylor, 2006) proposed in their "tend-and-befriend" hypothesis that females' response to stress involves seeking out others as a means to cope with the stressor, in contrast to a more confrontational or avoidant male response. While some studies have reported gender/sex differences (e.g., Prasad et al., 2017;Youssef et al., 2018), most studies do not report findings that would speak to systematic differences in social decision making following acute stress, compared to control. Both men and women have been shown to display increased affiliative behaviours (in the broadest definition), as well as reductions in social behaviours, and importantly, also no effects of acute stress, compared to controls. Given these mixed findings, we aimed to systematically explore potential gender/sex 1 differences across each economic game.

Stressor-type
Acute stress can be elicited and measured in many different ways. Consequently, acute stress has not always been induced as well as experienced in the same way across studies and designs, since different stressors typically result in different stress-responses. Commonly used indicators of an acute stress response include increased levels of glucocorticoids (i.e., elevated salivary cortisol), due to increased hypothalamic-pituitary-adrenal (HPA) axis activation; and increased activation of the sympathetic nervous system markers (heart rate, electrodermal activity, salivary alpha amylase, among others), and markers of emotional arousal (e.g., subjective stress, psychological distress). Psychosocial stressors, like the Trier Social Stress Test (Kirschbaum et al., 1993), result in a robust activation of these systems and markers. Physical stressors typically rely on an aversive physical experience (i.e., pain), such as in the Cold Pressor Task (Hines and Brown, 1936), and cause relatively short-term activations of the sympathetic nervous system and no reliable activation of the HPA axis (Schwabe et al., 2008). Furthermore, in contrast to physical stressors, psychosocial stressors are designed to highlight aversive interpersonal interactions as the source of distress (Dickerson and Kemeny, 2004), and as a consequence might affect social behaviours that follow. In this regard, a few studies have experimentally manipulated the stressor type and it has been proposed that social and physical stress may interact to influence prosocial behaviour (cf. von Dawans et al., 2018). Given this, we aimed to systematically explore stressor-type as a potential moderator.

Delay from stressor to task
The acute stress response unfolds dynamically and in stages (Joëls and Baram, 2009) with potentially differing effects on cognition and behaviour at each stage . In the immediate aftermath of the stress onset, we experience heightened vigilance due to activation of the brain's salience network at the cost of compromised executive functioning, with these effects reversing approximately one hour after the termination of stress (Hermans et al., 2014). We would therefore expect to find different effects of stress depending on the time-delay between stress exposure and the prosocial behaviour task, specifically if the effects are tied to HPA axis activation.

Decision complexity
We were interested in whether variations in how the dependent variable was measured influenced the effect of stress on prosocial behaviours. Specifically, we investigated: a) the number of decisions per task, and b) decision complexity, i.e., whether participants had to make a binary decision (i.e., share vs. not share in the Dictator Game) or whether the decision was more complex (i.e., how much money to share). Automatic decisions in economic games have been associated with increased prosociality compared to those made after more deliberation (Rand, 2016). Additionally, acute stress can compromise working memory processes (Shields et al., 2016;Wirz et al., 2018) resulting in more habitual and automatic responding (Hermans et al., 2014;Starcke and Brand, 2012) with potential consequences for social behaviour (Frith and Frith, 2012).

Present study
The aim of this meta-analysis was to 1) quantify the effects of acute stress on prosocial behaviour, and 2) empirically probe potential causes for differences in study outcomes. To this end, we focused on four standardised economic games -the Dictator Game, the Trust Game, the Ultimatum Game, and the Prisoner's Dilemma-as measures tapping into prosocial decision making, given that these have been most commonly used in prior research and can therefore enable us to generalize findings across studies (Baumard et al., 2013). We aimed to test the hypothesis that acute stress impacts social decision making, by either consistently increasing or decreasing prosocial behaviour, against the null hypothesis that acute stress does not impact prosocial behaviour in a consistent manner. Moreover, we explored whether commonly proposed moderators could explain the potential heterogeneity across studies.

Study selection
To select studies investigating the effects of acute stress on proscial behaviours using economic games we conducted a systematic literature review and searched three databases for the relevant literature: PubMed, Web of Science, and SCOPUS. For each database, we constructed filters which included the same words: ("Acute Stress" OR "Psychosocial Stress" OR "TSST" OR "cold pressor" OR "CPT" OR "MAST" OR "MIST") AND ("Dictator Game" OR "Ultimatum Game" OR "Donation Game" OR "Punishment Game" OR "Prisoners Dilemma" OR "prosocial" OR "social decision making"). For all databases, time and language filters were applied to only include research articles published until 30/06/2022. The output from each database was exported in a separate table, the tables were then combined into a single list and duplicates (articles present in more than one database) were removed from the final list of articles of interest. Each article in this database was then independently reviewed by two reviewers (authors one and two) to select studies that met the inclusion criteria: (1) Article written in English language; (2) Article containing original research (i.e., not reviews, opinions, etc); (3) Data had to be based on adult participants (aged 18 and above); (4) Study manipulating acute stress, using either of the following tasks: physical stressors, (e.g., cold pressor task (CPT)), psychosocial stressor (e.g., Trier Social Stress Test (TSST), Montreal imaging stress task (MIST)), social evaluative cold pressor task (SECPT), Maastricht Acute Stress Test (MAST)), or variants thereof, with a non-stress control or placebo condition; (5) Study reporting at least one economic game (i.e., Dictator Game, Trust Game, Ultimatum Game, Prisoner's Dilemma, or a variant thereof) to measure prosocial behaviour including two players. In addition, studies needed to include measures that could be extracted to calculate an effect size. A summary of the selection and exclusion criteria at each phase of screening is illustrated in the PRISMA flow diagram (see Fig. 2). In total, we identified 23 studies that were eligible for inclusion in this meta-analysis (see Table 1 and Table S2 in the Supplemental Materials). Importantly, several studies included more than one economic game or more than one experimental group (e.g., different populations or stressors). In those cases we created separate entries for each experimental group and each economic game (see next section). In addtion, whenever possible we extracted separate effects for the different gender/sex of the participants, different stress inductions, and different time-delays (from stressor to decision making).

Data synthesis and analyses
We first extracted means and standard deviations from the relevant publications. In cases where these data were not obtainable from the text or associated data-files, we first contacted the authors for the information, and if we received no reply then we used the figures provided in the manuscript (typically displaying mean and standard error of the mean) to manually extract data points from the graphical representation of data (using WebPlotDigitizer). 2 Effect sizes were calculated as a standardized mean difference (SMD; Hedges' g; Hedges and Olkin, 2014) between the stress groups and the control groups.
We decided to run two separate analyses: 1) to test the effects of acute stress on prosociality-as measured by the Dictator Game, Trust Game (Player A & Player B), and Ultimatum Game (Player A). While these tasks broadly measure different aspects of prosociality, we decided to include them in the same analysis-here, positive effect sizes indicate an increase in prosociality in the stress condition, compared to control. Importantly, task-specific results for each type of game are reported as well (from task-specific meta-analyses); 2) to test the effects of acute stress on costly punishment, 3 as measured by acting in the role of the responder in the Ultimatum Game (Player B). Here, positive effect-sizes indicate increased punishment in the stress condition, compared to control. Conversely, negative effect sizes indicate less punishment in stressed individuals, compared to control.
We used random-effects meta-analysis modeling to estimate the effects of acute stress on 1) prosocial behaviour, and 2) costly punishment. In order to account for repeated entries (or effects) per study we used a multilevel meta modelling approach, which allows the specification of random-effects and accounts for dependencies (i.e., potentially correlated effects) introduced by the individual studies (Harrer et al., 2019).
Here, we included each study (study-id), as a random error term in order to account for repeated observations within each cluster. We kept this error-structure in place for all analyses. Furthermore, for the two main analyses (i.e., prosocial decision making and costly punishment) we adjusted the degrees of freedom according to recommendations by Viechtbauer (Viechtbauer, 2010;Viechtbauer et al., 2015).
Following the two main analyses (i.e., prosocial behaviours and costly punishment) we ran a series of moderator analyses. All analyses were run separately for prosocial outcomes and costly punishment. First, as an individual difference measure, we divided the studies by the gender/sex of the participants to see if men and women behaved differently across tasks in the stress compared to the control conditions. Second, we were interested in identifying study design choices. To this end, we ran two moderation analyses: a) we divided the studies by type of stressor, focusing on three types of stressors, physical (CPT), psychosocial (TSST, TSST-G, MIST), and combined (SECPT, MAST); b) we divided the studies based on time-interval between the end of the stress task and the economic game. To this end, we divided the studies into two groups: studies with a delay of 10 min or less (i.e., immediately after the 2 Note, we decided to look at costly punishment as a separate effect from the other (more) prosocial effects reported here as it can be interpreted in terms of punishment instead and as such it might not be a "pure" measure of prosociality. Hence, we have decided to report this effect separately. However, some might argue that accepting any sort of offer is a prosocial action by itself. To circumvent this-admittedly complex-discussion we have included a metaanalysis with costly punishment as a prosocial effect in the Supplemental Materials as well. 3 We decided on a dichotomous classification of time-delay rather than using a continuous measure. Most studies with short delays did not provide exact timings, but rather "immediately" or "shortly" after the stress task. By dividing the tasks into early and late delays we avoid making judgment calls on the exact time-delay of the study protocol. Notably, the 10 min time-intervall is a rather arbitrary choice. This choice allows for all studies that were close to the stressor to be grouped together (with very similar timings) and to be compared to studies with longer delays (a much more heterogeneous group). This is largely reflecting the experimental choices made by the researchers, as evident from the distribution of the data. However, we note that this only allows for a rather crude comparison of different stress phases. end of the stress task) 4 ; and studies that had longer delays. Following this we ran two moderation analyses that aimed to tease apart differences in the dependent variable, mainly the automaticity of behaviour; to this aim we a) tested decision complexity by dividing effects based on the kinds of decisions participants had to make, namely, if they had the option for a (pseudo) continuous answer (i.e., the amount of money to be shared), or simply had to indicate if they would share a certain amount (binary choice: yes vs. no); (b) we divided effects based on the number of interactions, i.e., if the economic game was a one-shot game (i.e., one interaction only), or a series of independent one-shot games (i.e., n > 1). For the Dictator Game only, we tested whether or not the recipient of the act of sharing (i.e., another player or a charity) mattered to the outcome. Finally, for costly punishment, the available literature did not provide sufficient data, which is why only one moderation analysis (with gender/sex) could be conducted.

Transparency and openness
We adhered to the PRISMA 2020 guidelines, an evidence-based minimum set of items for reporting in systematic reviews and metaanalyses (Page et al., 2021). All meta-analytic data, analysis code, and research materials are available at https://osf.io/kjcua. Data were analyzed using R (Version: 4.1.0; R Core Team, 2020) using the 'metafor' package for mixed-effects meta-analysis (Version: 3.0-2; Viechtbauer, 2010). For the moderator analyses we utilized the moderator function of the 'metafor' package.

Study characteristics
The n = 23 studies resulted in a total of k = 77 extracted effect sizes with a combined sample size of N = 2197 (33.5 % women), ranging from n = 26 to n = 203 participants per study. There were a total of 61 effect sizes measuring prosocial behaviours (Dictator Game= 32, Trust Game Player A= 11, Trust Game Player B= 8, Ultimatum Game Player A= 4, Prisoner's Dilemma= 6), of which 26 had a positive effect size (42.6 %) and 35 had a negative effect size (57.4 %). In addition, we had 16 effects measuring costly punishment (Ultimatum Game Player B= 16), eight had a positive effect size (50 %) and eight had a negative effect size (50 %). Of all studies included, ten studies were conducted in Germany, five were in the United States of America, and seven were in other countries (Austria (2), Japan (2), Netherlands (1), Trinidad & Tobago (1), China (1)). All articles were published between 2007 and 2022. For a detailed list of all included studies see Table 1. We first report the test-statistics for each meta-analysis (or moderation), followed by a Q-test (a significant Q-test indicates significant heterogeneity in research findings). For each individual effect, we first report the SMD, confidence interval (CI) of the SMD, test statistics, and p value, if applicable, followed by the measures of homo-or heterogeneity, (2) Article containing original research (i.e., not reviews, opinions, etc); (3) Data had to be based on adult participants (aged 18 and above); (4) Study manipulating acute stress, using either of the following tasks: physical stressors, (e.g., cold pressor task (CPT)), psychosocial stressor (e.g., Trier Social Stress Test (TSST), Montreal imaging stress task (MIST)), social evaluative cold pressor task (SECPT), Maastricht Acute Stress Test (MAST)), or variants thereof, with a nonstress control or placebo condition; (5) Study reporting at least one economic game (i.e., Dictator Game, Trust Game, Ultimatum Game, Prisoner's Dilemma, or a variant thereof) to measure prosocial behaviour including two players. In addition, studies needed to include measures that could be extracted to calculate an effect size. 4 Two of the studies (Potts et al., 2019;von Dawans et al., 2018) included several experimental groups, but only one control group. We decided to include all experimental conditions separately, and compare the effects to the same control group. Importantly, this selection choice might have increased bias and artificially inflated precision of our estimates. In order to account for the repeated comparisons for the control group, we decided to adjust the participant numbers for each control group comparison to match the number of comparisons (i.e., for two comparisons the n of each control group was halved) (Higgins et al., 2019). For, Schweda et al. (2019) we included two comparisons, in-and out-group sharing for control vs stress groups, respectively: sample-size was not adjusted. Furthermore, Margittai et al. (2015), did not report a standard deviation for each level of social distancing, but we were able to calculate a standard deviation across all levels of distance (our dependent variable). For Schulreich et al. (2022) we compared the data from the post-intervention trials (control vs. stress) corrected for baseline levels (post-interventionpre-intervention). This adjustment was necessary in order to correct for reported baseline differences between the two groups. (continued on next page) J.P. Nitschke et al. specifically, Q-statistics, I 2 , and the prediction interval of the SMD (PI; Riley et al., 2011). We tested for publication bias using Egger's regression test (Egger et al., 1997) by modifying the models to include the square-root of the sampling variances of the effect size as a moderator (Habeck and Schultz, 2015).

Effects of acute stress on prosocial behaviours
The meta-analysis for prosocial behaviour resulted in a nonsignificant SMD of  Table 2 for an overview and Fig. 3 for a forest plot.
The meta-analysis testing the effects of acute stress on costly punishment resulted in a non-significant SMD of − 0.11 (95 % CI %[− 0.28, 0.07]), t(7) = − 1.45, p = .19), again indicating no effect of acute stress on the rate of punishment behaviours, compared to the control group. The test of effect heterogeneity was non-significant (Q(15) = 24.54, p = .06; I 2 = 36.83 %; 95 % PI for SMD: − 0.69, 0.48), indicating relative homogeneity in effects. The Egger's test for publication bias revealed no significant bias in the published literature (95 % CI[− 1.57, 0.33], p = .16) (see Supplemental Materials for a funnel plot). See Table 2 for an overview of all effect sizes and Fig. 4 for a forest plot.

Moderator analyses for prosocial behaviours
Next we ran a series of moderation analyses (see Table 3 for an overview). Note, that we first tested if the conditions (i.e., gender/sex, stressor-type, delay between stress task and economic game, binary decisions, number of decisions) differed compared to a reference group (e.g., did female participants differ from male participants), before we tested if that effect of interest was significantly different in the stress compared to the control groups (i.e., did the 95 % CI of the SMD include zero or not).
Overall, gender/sex did not modulate the effect of acute stress on prosocial behaviour (F(2, 58)= 0.10, p = .90; Q(58)= 118.48, p < .01). Neither was there a significant effect for men (coded as reference group)    These results show that for stressed individuals binary decisions compared to non-binary decisions were significantly more prosocial. Next, we compared the effect for binary decisions between the control group and the stress group and found that this effect included zero in the 95 % CI for the SMD. We also compared the effect for non-binary decisions between the groups and again this effect included zero in the 95 % CI for the SMD. Thus, despite a significant moderation effect, there was no significant difference between control condition and stress condition for either of the decision options.
In addition, we ran a moderation analysis to test the difference between one-round paradigms (only one decision, i.e. one-shot games) versus paradigms that included several (but independent) rounds (multiple decisions). We found no significant difference between ( Lastly, for the Dictator Game only, we subdivided sharing behaviour by target-type, i.e., who the recipient of the sharing was (anonymous coplayer (k = 17)(coded as reference group); charity (k = 10); hypothetical, i.e., social discounting (k = 5). Here, we found a significant moderation in sharing behaviour for target-type, F(2, 29)= 5. These results indicate that decisions about sharing behaviours differed for anonymous players, compared to charitable giving, with significantly lower levels of sharing in the stress group when donations were made to a charity. Decisions based on social discounting did not significantly differ from anonymous other players. Notably, while the effect for anonymous players was not significantly different for the stress versus the control group, in the stress group decisions made for charitable giving were significantly lower in the stress compared to the control groups (95 % CI for the SMD did not include 0).

Moderator analysis for costly punishment
For costly punishment (i.e., Ultimatum Game Player B) we ran one moderation analyses for gender/sex. There was no sufficient variation in task designs to run the other moderation analyses. For gender/sex we did not find a significant moderation F(1, 14)= 0.01 p = .91; Q(14)=

Fig. 4. :
Forest plot for the effects of acute stress on costly punishment. Higher numbers indicate increased costly punishment in the stress versus the control condition.  Table 3 for the results of the moderator analysis. For task specific effects of the moderation analysis see Table S3 in the Supplemental Materials.

Discussion
Does acute stress lead to increased, or decreased, prosociality? To answer this question, we conducted a meta-analysis that included 23 studies, 77 effects, and 2197 participants. We did not find a clear effect in either direction for prosocial decisions (or costly punishment) in the Dictator Game, Trust Game, Ultimatum Game, or Prisoner's Dilemma, thus supporting the null hypothesis that acute stress does not impact prosocial behaviour in a consistent manner. These results indicate that acute stress neither increases nor decreases prosocial behaviours consistently, as measured by these economic games. Despite restricting our analysis to studies which used these relatively standardised economic games, there was considerable heterogeneitity, even within a particular economic game, in terms of how the dependent variable was measured (e.g., a binary decision vs. more continuous measures of sharing), the type of stressors that were employed (e.g., physical vs. social), the study designs (such as variations in delays between stress induction and economic games), and the different gender/sex compositions of the samples. To investigate these divergent effects we tested these previously identified moderators. We found that decisioncomplexity moderated the impact of acute stress on sharing: binary decisions (i.e., share vs. not-share) resulted in increased prosociality for participants in the stress group, compared to studies that used more complex decisions. Conversely, we did not find an effect for non-binary decisions requiring participants to choose the amount of money on a continuous scale (i.e., "how much do you want to share?"). In addition, we found that in the Dictator Game, the target of the monetary allocation mattered. Stressed participants allocated more money to anonymous other players, compared to charities; and for charities, stressed individuals also allocated significantly less compared to control. Contrary to some previous studies, differences in experimental design (e.g., Margittai et al., 2015;Vinkers et al., 2013), type of stressor (e.g., von Dawans et al., 2018), or gender/sex of participants (e.g., Nickels et al., 2017;Prasad et al., 2017;Youssef et al., 2018) did not modulate the effects of acute stress on prosocial behaviour.

Gender/sex
We did not find evidence for differences in the effects of acute stress on men's and women's prosocial behaviour. This is seemingly at odds with some theoretical accounts, such as the tend-and-befriend hypothesis (Taylor et al., 2000), that suggest gender differences in the response to acute stress, with women showing a higher propensity for affiliative behaviours, compared to men. However, it is unclear if economic games, as used in the current context, are necessarily the most appropriate means for testing affiliation. In this sense, most previous studies have commonly used "stranger" stimuli (that are often abstract and not tangible), whereas affiliative behaviours often concern prosocial action towards more familiar others, and thus, individuals that are likely to act as support providers (compared to anonymous strangers; Depue and Morrone-Strupinsky, 2005;Taylor, 2006). This increased tendency towards prosocial behaviours during acute stress, with preferences for close others compared to strangers, has some evidential support in male participants (e.g., Margittai et al., 2015;Schweda et al., 2020). Regardless of these theoretical considerations, we found no differences in male and female participants, with both groups showing a great variety of prosocial behaviours-increases, decreases, and no changes-in response to acute stress, compared to controls. It is also important to point out that within gender/sex variations are greater than between gender/sex variations raising the question of whether gender/sex is a meaningful predictor of individual differences, or if other sources of person/context dependent variations outweigh possible gender/sex effects.
The endocrinological stress response differs between men and women. In this regard, a consistent finding is that psychosocial stressors, such as the TSST, result in a more pronounced cortisol response in men, compared to women (Kudielka et al., 2009;Liu et al., 2017). This would suggest, that if behaviours and decisions are affected by stress-induced levels of glucocortioids we can likely expect gender/sex specific differences. For example, research on social cognitive processes has found that men and women show different responses to acute stress when it comes to cognitive inference making. For example, Smeets et al. (2009) found gender/sex differences in a social inference making task (i.e., "what would another person do") that was tied to the cortisol stress response. Here, men benefited from increased cortisol, whereas women showed a detrimental effect of cortisol on inference making abilities. Similarly, Nitschke et al. (2022) could show that while women's ability to accurately track the thoughts and feelings of others was unaffected by the experience of acute stress, men's performance benefitted. Notably, in men, this increase in accuracy was tied to the stress induced glucocorticoid response. In contrast to this, Tomova et al. (2014) found that women showed a stress-induced increase in self-other distinction (potentially indicating a flexible separation of self related vs. other related experiences), whereas men showed a stress-induced decrease. This contrasting evidence highlights the heterogeneity in research findings, indicating possible hidden moderators or task dependent effects. However, we do not find any meta-analytic evidence for gender/sex specific effects on social behaviours following stress. Notably, only 33.5 % of the individuals included in the meta analysis were female. Furthermore, of the 23 studies included in this meta-analysis, 14 studies included both male and female samples which would allow for a direct comparison for gender/sex (if the goal was to explore gender/sex differences). Purely male or female samples make it difficult to disentangle true differences between groups and other factors, such as experimental design choices, which might interfere with or outweigh any more subtle gender/sex differences. Additional studies including female participants are needed, importantly including mixed samples. In addition, it is necessary to consider other factors that have been attributed to differential stress responsiveness, such as the female menstrual cycle and oral contraceptive use (for reviews: Kudielka et al., 2009;Liu et al., 2017).

Variations in stressor types
Our moderator analyses did not show divergent effects for stressor types across studies on social decision making following stress. This is in line with the inconsistencies in the literature on the effects of acute stress markers on decisions in economic games (see Supplemental Materials Table S4 for an overview of the association of acute stress markers on social decision making in the included studies; however, von Dawans et al., 2018).
The acute stress response, depending on the stressor, unfolds in different stages (Joëls and Baram, 2009). We would therefore expect to find different effects of stress depending on the time-delay between stressor onset and the social decision making task, specifically if the effects of decision making are tied to HPA axis activation. That we do not find reliable evidence of differential effects due to delay in our moderator analysis might indicate that the glucocorticoid response does not play a crucial role in the effects of acute stress on social decision making (in either direction), or that other factors might moderate this association (cf. Azulay et al., 2022;Schulreich et al., 2022;Schweda et al., 2019;Zhang et al., 2019). This is further corroborated by the findings that the type of stressor did not influence prosociality (at least not in the experimental designs included in current meta-analysis). Thus, other factors must be driving these differences in effects, and differences in outcome between studies. This might indicate that "stress" per se-that is the experience of an aversive situation-is pertinent for changes in social behaviour, including perceptions about the stressful situation (i. e., more or less social) and the participants' own perceived agency. This further highlights the idea that the acute stress response extends beyond the activation of just one system (e.g., cortisol response), but rather represents an intricate interplay of the various systems involved (Ali et al., 2020), and should be considered more holistically. Thus, the different stress systems, and potentially their interactions, need to be considered when evaluating the effects of acute stress on social behaviours. Of note, due to lack of systematic reporting of stress markers (e.g. cortisol, heart rate) in the studies discussed here, it was not possible to quantify specific effects of individual systems, or their interplay, in the current meta-analysis.

Variations in the dependent variable
We did find a significant moderation effect for decision complexity, specifically binary (e.g., share vs not-share) versus non-binary choices (e.g., "how much") on prosociality (p < .05) for stressed individuals. As such, this might indicate that choice complexity can affect some aspects of social decision making. Less complex decisions (binary choices; share vs. not-share) were associated with a positive effect size, i.e., increased sharing and increased trust, whereas more complex decisions (non-binary choices) were associated with a negative effect size, i.e., reduced sharing and trust. 5 This is in line with the idea that intuitive decisions-those that are less deliberate-tend to be more prosocial. For example, people tend to act more generously under time pressure, compared to when they are given more time to think about their actions (Rand, 2016). 6 However, as we did not have access to reaction-time data it is possible that participants still deliberated longer in these binary choice paradigms compared to other types of non-binary decisions, thus making them less automatic and intuitive. For example, von Dawans et al., (2019von Dawans et al., ( , 2018von Dawans et al., ( , 2012 used binary choice paradigms to measure prosocial actions following acute stress in three separate studies. Here, participants were presented with either-or-choices (e.g., share [15:15 split] vs. not share [30:0 split]; trust vs. not trust) across several rounds and several decision making tasks. For a trust game (Player A of the Trust Game) participants were not asked to choose a specific amount for the investment, but rather, were given the option to either not trust--consequently keeping all of the initial endowment-or to trust, which would result in either an equal split if reciprocated by the other player, or a loss of the investment (if the other player decides to not reciprocate). Similarly, in the Dictator Game participants were not asked "how much" they would share with another player, but rather were given the option to either keep the entire endowment for themselves or split it equally with another participant. This is in contrast to, for example, Singer and colleagues (2021) or Takahashi et al. (2007) where participants were only asked once at the end of the study to share a self-chosen (continuous) amount of money with others. Another way to gauge task complexity is to look at differences between these tasks. Differences between these tasks might offer some insight about decision complexity, with the Dictator Game being more simple in its application compared to decisions made in Ultimatum Games. Conceptualized like this (and similar to binary vs non-binary decisions), we find evidence that differential complexity in these economic games could, arguably, be an indicator for stress-dependent effects, with more complex decisions in the Ultimatum Game resulting, on average, in a negative effect size (i.e., stress decreased prosociality; SMD= − 0.22). Future studies will have to systematically manipulate choice complexity (or other potential markers of automaticity) to test this potential moderator. We also examined the number of decisions participants made in each task, and found no differences for the effects of acute stress. Thus, on average participants showed the same tendencies for games with fewer trials (i. e., less than 5 decisions per game), compared to games with more trials (i.e., more than 4 decisions per game). In studies which used tasks with multiple trials, each trial was pitched as independent, meaning that participants did not play repeatedly with the same partner, but instead with a new partner in each round. Presumably this independence should guarantee that each interaction is approached as a tabula rasa. However, FeldmanHall et al. (2015) could show that this presumed independence does not necessarily hold up under acute stress. In their study, participants played 36 one-shot rounds of the Trust Game with a new co-player for each round. Following stress (cold pressor task) participants were more likely to incorporate previous trials into their decision making, but only in a social context (trust game) and not in a non-social control task (lottery game). This resulted in more irrational social decisions by the participants who treated each round as dependent on the previous interaction, when they should not influence each other. These findings show that stressed individuals may process social information differently. One way to interpret this is that while in no-stress situations it may be adaptive to approach every new social interaction as a blank state, under acute stress the stakes for doing so might be higher, consequently shifting decision making to be more conservative. Another interpretation is that the integration of novel information during acute stress is impaired (Nitschke et al., 2020a(Nitschke et al., , 2019Schwabe and Wolf, 2012) which could possibly result in misinterpretation of the available data. However, specific mechanisms that highlight how acute stress impairs information processing for social decision making in the context of economic games is still unclear. Future research is needed to test this association more explicitly.

Social interaction partner
The majority of studies included in our meta-analysis focused on prosociality towards anonymous others. However, prosocial behaviours are of particular importance for interactions that involve familiar others, and in particular close others (Balliet et al., 2014;Engel, 2011). It is therefore likely that (perceived) closeness could moderate the effects of acute stress on sharing behaviours, as interactions with kin or familiar others provide opportunities for reciprocity, a cornerstone of prosociality (Fehr and Schmidt, 2006;Kurzban et al., 2015;Queller, 1985). A few studies have investigated whether or not acute stress and social decision making matters for different degrees of closeness (Margittai et al., 2015;Passarelli and Buchanan, 2020;Schweda et al., 2020;Steinbeis et al., 2015). The studies included in our meta-analysis which directly manipulated social distance found that it modulated the effect of acute stress on prosociality. When making decisions about prosocial behaviours, individuals have a tendency to prioritise close others over strangers, an effect that has been shown to be exaggerated under conditions of acute stress (Margittai et al., 2015;Schweda et al., 2020;Steinbeis et al., 2015;however, Passarelli and Buchanan, 2020). For instance, Margittai et al. (2015) looked at the effects of acute stress, compared to control, on social discounting behaviour. The authors found that men tested shortly after the end of the stress task (approximately 20 min) showed an increase in generosity towards close but not to distant others, compared to non-stressed men or men tested with a longer delay after stressor onset (approximately 90 min). Importantly, in the context of this study, Margittai and colleagues did not find any evidence that acute stress had an effect on overall generosity, i.e., 5 Note, when estimating the specific effect sizes from these two types of decisions, the 95 % effect-size confidence interval includes zero for both effects, indicating a non-significant effect for both types of decisions, i.e., the stressdependent decisions were not significantly different from the control group. 6 Of note, the findings on automaticity-that spontaneity is associated with greater cooperation-have been questioned in recent years, with systematic replication attempts showing considerably smaller effects compared to the original study by Rand and colleagues (Tinghög et al., 2013;Bouwmeester et al., 2017).
average generosity independent of social distance, compared to control. This indicates that if acute stress impacts prosociality, it might do so differently for different levels/degrees of closeness. Here, we found that participants under acute stress shared less in the Dictator Game when giving to a charitable cause (negative effect size), compared to anonymous players (positive effect size). Notably, while there was no significant difference for the stress versus the control group for anonymous others, stressed individuals shared less for charitable causes compared to participants in the control condition. This might indicate that co-players, even if anonymous (and not physically present), might feel more tangible (as they are, presumably, other participants in the experiment), compared to sharing with a cause, which is arguably more abstract (i.e., donations to charitable organizations, such as UNICEF, or causes supporting climate action). Stressed participants might be less inclined to donate to causes that seem removed from their stressful situation. As these levels of closeness are rather abstract, it might be interesting for future studies to further explore this by more rigorously manipulating social distance; does this effect depend on the type of charity, the importance of the cause for an individual (cf. Sollberger et al., 2016), or the perceived closeness of other players? Furthermore, it will be important to consider the nature of the interaction more closely; for example, does it matter if the play-partner is completely fictional (i.e., they do not exist), do person-dependent characteristics of the play-partner play a role (e.g., man or woman), or if the participants had an opportunity to meet/interact with the other player? In this regard it has been argued that prior social contact could modulate prosociality in subsequent economic games (Balliet, 2010;Declerck et al., 2020).
Notably, at this point, data are still sparse, as relatively few studies have investigated this relationship, and not necessarily systematically-thus, future studies will have to investigate these effects further.

Additional factors
Our analysis focused on the influence of stress on behaviour in a series of economic games. Whilst economic games are easy to implement, subtle changes in task designs (e.g. binary vs. non-binary choices) can have profound effects on performance, making comparisons across studies difficult (Nitsch et al., 2021). Moreover, following acute stress, it is unclear what drives altered performances in these tasks. Does reduced sharing in the Dictator Game reflect increased prosociality or a reduced preference for one's own rewards under stress (Berghorst et al., 2013)? Furthermore, when we help others in everyday life, this often comes at a cost of effort and/or time, rather than simply sharing (or investing) a financial endowment (Lockwood et al., 2017). In addition, outside the laboratory, helping behaviours often involve the reduction of others' suffering, or alleviating unpleasant states; rather than maximizing monetary wealth-as measured by the economic games used here (Crockett, 2013;Decety and Cowell, 2018;Hartmann et al., 2022;Lengersdorff et al., 2020). Thus, to better capture changes in prosocial behaviour following stress induction, we propose that future studies should employ additional measures of prosociality, or include a diverse set of measures (cf. Haesevoets et al., 2021), including measures that capture helping behaviours in everyday life more closely (Balliet et al., 2009;Galizzi and Navarro-Martinez, 2019). To this end, Forbes et al. (2022) recently demonstrated that when deciding whether to exert physical effort to obtain a reward, participants under acute stress prioritise their own rewards over those of others to a greater extent than participants in a control group. This fits with previous work demonstrating the tendency to avoid cognitive effort under stress (Bogdanov et al., 2021), and suggests that when effort costs are involved we see reductions in prosociality under stress. However, future studies are needed to test the robustness of effects showing altered effort-based decisions under stress. Lastly, it might be important to investigate the nature of the incentives in these economic games more directly. While the vast majority of these games are incentive-compatible-i.e., they result in tangible payouts (at least for some trials)(see additional information in the Supplemental Materials TS2), it will be important to explore the role of incentives further, including motivations to act a certain way more systematically (i.e., by varying payouts, giving participants opportunities to interact etc).

Costly punishment
The previously discussed economic games were concerned with sharing money (or monetary units) with another participant. This is different to the second player in the Ultimatum Game (Player B) who can decide to accept an offer (from Player A) or reject it. Acceptance is mutually beneficial, as both players receive a payout; however, rejecting an offer has negative consequences for both players-hence, costly punishment. While it can be argued that punishing behaviour in the Ultimatum Game is motivated by preferences for fairness, and as such could be considered prosocial, we decided to look at this more complex behaviour separately from the other economic games (Guala, 2012). For an analysis including all economic games in one meta-analysis see Supplemental Materials. As with the other economic games measuring prosocial behaviours, we do not find evidence that acute stress systematically alters costly punishment (SMD= − 0.11). Similar to the other economic games, neither the gender/sex of the participants nor stressor-type were significant moderators of this effect. Importantly, there are not enough studies to allow for a more systematic investigation of the factors that potentially influence the association between acute stress and costly punishment. Given the importance of repeated interactions in Trust games (FeldmanHall et al., 2015), in which stressed participants are more influenced by previous interactions, one exciting avenue will be to test how repeated iterations of the Ultimatum Game influence costly punishment under stress. Here, computational approaches, which can precisely quantify the influence of previous trials on subsequent decisions, could be especially useful (Zhang et al., 2020).

Limitations
It is important to understand the findings of this meta-analysis alongside its limitations. Here, we focused on prosociality as measured by a selection of commonly used economic games. This choice omitted other factors and constructs that are often considered measures of prosocial behaviours (Penner et al., 2004;Schroeder and Graziano, 2015), including moral judgments, helping behaviours, or compassion. As such, our findings represent a limited view of what prosociality constitutes. Thus, our conclusion-that acute stress has no consistent effects on prosociality-might only apply in the context of economic games.
Furthermore, while we used data from 23 studies resulting in 77 effects, for some economic games there was a lack of sufficient data to reliably estimate task-specific effects, or to run moderation analyses given the low variations in task design. For a meta-analysis, and particularly one that uses moderation analyses, the sample (n = 2197) is rather small, and therefore it might be less surprising that most of these effects were non-significant. As such it might be better to interpret the effect-sizes obtained directly. All effects observed are rather small (i.e., ranging from SMD= − 0.24 for the Ultimatum Game Player A to SMD= 0.03 for the Dictator Game). Given this, additional research might be needed to look at game-specific effects. In addition, the current analyses only includes studies that were peer-reviewed and published, hence accessible by searching available databases. While we did not find any evidence for publication bias, it is important to consider additional unpublished data. Since it is less likely for non-significant findings to be published (Franco et al., 2014), it is possible that these unpublished data would further substantiate our non-significant findings of acute stress on prosocial decision making.
Lastly, the findings discussed here are applicable to a narrow base of the population that is relatively young (20-35-ish), mainly university students, and mostly from Western, educated, industrialized, rich, and democratic (WEIRD) societies. As there are potential cross-cultural differences in sharing behaviour in economic games (Raihani and Bshary, 2012), exploring potential cultural differences in terms of acute stress effects is an exciting avenue for future studies.

Conclusion
Given its importance for our social togetherness, especially during times of increased stress, identifying and understanding factors that can facilitate or decrease prosocial behaviours are of particular interest. Research on acute stress suggests that it can both increase and decrease prosociality-or not influence it at all. Our meta-analysis shows that acute stress is not a reliable predictor for changes in prosociality (as measured using economic games), as we found no consistent directional effects of acute stress, compared to control, neither on prosocial decision making nor on costly punishment. While we found significant heterogeneity in the research findings reported, subsequent moderator analyses could not account for divergent findings-there were no moderating effects of participants' gender/sex, paradigm designchoices, or the type of stressor, despite all having been implicated as moderators in previous studies. The conclusion is not that acute stress does not impact social decision making, but rather that the effects of acute stress are inconsistent, at least in the experimental tasks and settings used so far. Our systematic and quantitative analysis revealed a high level of heterogeneity in research findings. Notably, this is despite the highly standardized nature of the stress protocols being used, as well as the relative homogeneity of the economic games. Given the evidence that acute stress has the potential to shift social decision making to be more-or less-prosocial, the focus of future research should be to understand under which conditions, and in which individuals, such shifts occur. Box 1.

Box 1
Questions for future research.

Who do we help under acute stress, and why?
Given the focus of the 'tend and befriend' hypothesis on affiliation with kin or close others (Taylor et al., 2000), how does acute stress impact social behaviour towards closer others compared to strangers or charitable organisations? Previous studies suggest we tend to help closer others more under stress (e.g., Margittai et al., 2015). Our meta-analysis suggests that under acute stress we donate more readily to anonymous other participants than charitable organisations. But what is driving these effects? For example, is this a form of stress-regulation, whereby acting prosocially towards closer others or actual people is more likely to reduce stress, does helping others possess hedonistic qualities that makes a stressed individual feel better about themselves.? It has been suggested that under threat one strategy would be to increase prosocial behaviour as a means for resource-building by affiliating with others, or by recruiting others as a means of social support (Faber and Häusser, 2021). Furthermore, acute stress may reduce the flexibility of our cognition resulting in behaviour based on habits rather than careful deliberationfrom 'thinking' to 'doing' (Schwabe and Wolf, 2009;Wirz et al., 2018) This raises the possibility that if prosocial behaviour is a habit (as is the case for many health care professionals), then acute stress may increase prosocial behaviour in certain circumstances and for certain people.

Is helping too much effort under stress?
Acute stress can lead to an avoidance of effort (Bogdanov et al., 2021) and prosocial behaviour (especially outside the lab) often involves energising our actions to help others (Lockwood et al., 2017). Economic games used to measure prosociality have rarely considered the role of effort in influencing prosociality. Recent evidence suggests that effort influences prosocial decisions under stress with people becoming less willing to exert effort for other people's rewards (Forbes et al., 2022). However, does this hold for different types of prosocial behaviour under stress, such as when we reduce the pain of others (Hartmann et al., 2022)?

How does stress influence prosociality across the lifespan?
Prosocial behaviour varies across the life span. For example, older adults tend to be more generous, especially to in-group members (Cutler et al., 2021a(Cutler et al., , 2021bLockwood et al., 2021;Riva et al., 2022). Additionally, older adults show differing physiological responses to acute stress (Kudielka et al., 2009). There is some preliminary evidence that would suggest differential effects of stress based on age (Sparrow et al., 2019). However, additional research using a broader variety of tasks is required. Stress induction techniques specifically aimed at younger samples (e. g., Buske- Kirschbaum et al., 2021) will be useful in determining these developmental effects.

How does stress influence other forms of prosociality?
The most common way to study prosociality in the laboratory is by using economic games, as discussed in the current paper; however, there are many other ways in which prosociality can be measured. First, while we chose the most commonly used economic games-in order to better compare these effects-other tasks exist, such as social risk taking games (Bendahan et al., 2017) and games that include more than the two players (to which we have restricted the current analyses). At this point, however, there are a limited number of studies that have implemented three player games (Wang et al., 2022;Zhen et al., 2021), and an inclusion in the current analyses was not warranted. Second, prosocialty in real-life situations extends beyond economic exchanges to behaviours ranging from to acts of helping-often through physical means, to moral decision making (Singer et al., 2021;Starcke et al., 2011;Youssef et al., 2012), and to various forms of social support (e.g., Bodenmann et al., 2015;Ditzen et al., 2019). In order to understand what acute stress does to prosocial actions it is important to extend this framework and look at helping behaviours more broadly, and in more naturalistic settings, and with non-monetary costs, such as time and effort.

Do individuals differ in their propensity to make social decisions?
Individual differences in empathic abilities, personality traits can play a major part in social decision making (Speer et al., 2022;Thielmann et al., 2020). Recent work suggests that acute stress impacts social behaviour differentially depending on some of these individual differences. For example, Schulreich et al. (2022) found that while acute stress alone had no impact on charitable giving (Dictator Game), high levels of cortisol in combination with high mentalizing abilities (assessed with a behavioural task) resulted in a stress-induced reduction in donations, but not in low mentalizers (however, see Zhang et al., 2019). In contrast, Azulay et al. (2022) found a positive association between higher trait levels of empathy (assessed via questionnaire) and higher levels of stress-induced cortisol on more generous behaviour. These differences highlight the importance of considering traits and aptitudes-such as empathy, prosocial concern, or affiliation-as possible moderating factors for the effects of acute stress on social decision making. Notably, this should include factors that might influence the acute stress response as well, such as social anxiety, self-esteem or tendencies for rumination.