Testing Effects of Social Rejection on Aggressive and Prosocial Behavior: A Meta-Analysis

Abstract

Social rejection elicits profound feelings of distress. From an evolutionary perspective, the best way to alleviate this distress is to behave prosocially, minimizing the likelihood of further exclusion. Yet, examples ranging from the playground to the pub suggest rejection commonly elicits aggression. Opposing theoretical perspectives and discordant empirical results have left a basic question unanswered: does rejection more commonly elicit prosocial or aggressive behavior? We conducted three meta-analyses (one with studies measuring aggressive behavior; one with studies measuring prosocial behavior; and one with studies measuring both aggressive and prosocial behavior; N=3,864) to quantify: (1) the extent to which social rejection elicits prosocial or aggressive behavior; and (2) potential moderating effects on these relations. Random-effects models revealed medium effects such that social rejection potentiated aggressive behavior (k = 19; d = 0.41, p < .0001) and attenuated prosocial behavior (k = 7; d = 0.59, p < .0001), an effect that remained consistent even when participants were given the option to behave prosocially or aggressively (k = 15; d = 0.71, p < .0001). These results cast doubt on the theory that rejection triggers prosocial behavior, and instead suggest it is a robust elicitor of aggression.

Introduction

Given the fundamental human need to belong (Baumeister & Leary, 1995), social rejection often elicits profound distress (Eisenberger et al., 2003; Otten et al., 2017; Schaan et al., 2020). From an evolutionary perspective, it is adaptive to repair or form new relationships after rejection to protect against further exclusion (Baumeister & Leary, 1995; Weerdmeester & Lange, 2019; Williams & Nida, 2011). Yet, ample evidence from the playground to the pub demonstrates that rejection commonly elicits aggression (DeWall & Bushman, 2011; Leary et al., 2003; Chester & DeWall, 2016). Using meta-analytic techniques, we sought to answer a critical debate in the literature: Does social rejection more commonly elicit prosocial behavior or aggressive behavior? Alternatively, is there support for each outcome, but only under certain conditions? Isolating these conditions could help explain contradictory findings, and inform the design of future research and interventions aimed at reducing rejection-elicited aggression.

Evidence for rejection-elicited prosociality largely stems from work demonstrating increased affiliative behavior following rejection (e.g., Lee & Shrum, 2012; Schaan et al., 2020; Williams et al., 2000). Laboratory studies suggest that rejected, relative to included, individuals conform, cooperate, and work harder on group tasks, show increased desire to work with others (Lee & Shrum, 2012; Schaan et al., 2020; Williams & Sommer, 1997), and engage in more nonverbal behavioral mirroring (Lakin et al., 2008). However, it is less clear whether rejection elicits overt prosocial behavior, such as sharing money or providing positive feedback. Still, other studies demonstrate that rejection provokes aggression. For example, rejection potentiates relational aggression, such as publicly embarrassing others (Buckley et al., 2004), and providing negative evaluative judgements (Rajchert & Winiewski, 2016; Reijntjes et al., 2010, 2011; Weerdmeester & Lange, 2019). Similar effects emerge for physical aggression: rejection increases willingness to deliver aversive noise blasts (Beyer et al., 2014; Rajchert & Winiewski, 2016; Twenge et al., 2001) or hot sauce, causing pain (Ayduk et al., 2008; Reiter-Scheidl et al., 2018; Webster & Kirkpatrick, 2006).

These inconsistent results fuel the debate about whether prosocial or aggressive behavior is more likely after social rejection. One prior meta-analysis attempted to resolve this debate (Gerber & Wheeler, 2009), claiming that aggression emerged when participants were unable to regain a sense of control, and affiliative behavior (e.g., nonverbal mimicry) was elicited when participants could regain a sense of belonging. However, less than half of the studies included in this meta-analysis involved the actual expression of aggressive or prosocial behavior (only 16/40), and instead used imagined scenarios or mood measurement. Moreover, because behavior was subjectively coded as either restoring a sense of control or belonging, they were not able to tease apart the effects of rejection on aggressive or prosocial behavior (Baumeister et al., 2009). Thus, the effects of rejection on aggressive and prosocial behavior, and the moderators that may impact this relation, remain untested.

The present work takes several important steps towards resolving this debate. We test the effects of rejection on objective outcomes by operationalizing prosociality and aggression using a measure of behavior, rather than simply inferring the psychological state of the individual. This is important because emotions and intent to behave do not always correlate with actual behavior (Ajzen et al., 2004; Linn, 1965; Wong & Sheth, 1985). We define prosocial behavior as an action that results in beneficial outcomes for others, including withholding punishment when the only alternative is delivering punishment (Brittian & Humphries, 2015), and excluding affiliative behavior that does not directly benefit others (e.g., nonverbal mimicry; Van Baaren et al., 2004). It also excludes behavior that simultaneously confers explicit personal benefit (e.g., helping so your team wins a prize; Twenge et al., 2007), as this behavior cannot be separated from motivations for personal gain (Eisenberg, 1986). We define aggressive behavior as “any behavior directed toward another individual that is carried out with proximate intent to cause harm” (Anderson & Bushman, 2002). This definition excludes behavior enacted without intent to harm (e.g., accidentally tripping someone), imagined aggression (e.g., thinking of tripping someone), and emotion-based ratings (e.g. feeling angry). These exclusionary criteria reflect the fact that emotions are not a reliably proxy for behavior, not all aggression stems from anger, and people do not reliably forecast their own aggressive behavior (Anderson & Bushman, 2002; Averill, 2012).

Age may moderate the expression of rejection-elicited prosocial and aggressive behavior

Although social rejection occurs across the life span (Kann et al., 2018; Liu et al., 2013; Eisenberg et al., 2002), age may impact rejection-elicited behavior. For example, sensitivity to social rejection peaks during adolescence (London et al., 2007; Hafen et al., 2015), thus adolescents’ behavioral responses following rejection may differ from adults. Aggressive behavior decreases from early to mid-childhood, increases again in adolescence and young adulthood, before decreasing again in later adulthood (Liu et al., 2013; Fraser, 1996). Prosocial behavior on the other hand is first observed in childhood and continues to increase in frequency into adolescence and early adulthood (Eisenberg et al., 2002). Indeed, some studies suggest prosocial and aggressive behavior have an inverse relation across development (Obsuth et al., 2015). Yet, the relation between age and rejection-elicited prosocial and aggressive behavior is poorly mapped. We will therefore test the moderating effect of age on rejection-elicited prosocial and aggressive behavior.

Contextual factors may moderate the expression of rejection-elicited prosocial and aggressive behavior

Behavior elicited by social rejection is likely impacted by a myriad of contextual factors. Because contextual factors covary with task design features, we will test the extent to which several aspects of the rejection event or features of behavioral outcomes moderate aggressive and prosocial behavior.

Methods of rejection.

Given the large number of social rejection induction paradigms, some may be more or less likely to induce prosocial or aggressive behavior. The moderating effects of several rejection-related features were explored.

Rejection/Exclusion or Ostracism.

Social rejection has been defined as explicit exclusion from a social group or relationship (Leary, 2010; Williams & Nida, 2017), or receiving negative social feedback from an individual (Rappaport & Barch, 2020). Social exclusion is often defined by a wider range of events that revolve around group membership. These events include threats to one’s social belonging (Riva & Eck, 2016), or negative social feedback (e.g., rejection) from more than one individual about aspects of group membership (Rappaport & Barch, 2020). Ostracism is defined as being ignored, “left out” (Williams & Nida, 2017), or implicitly excluded from a group (Williams, 2001; Zadro et al., 2004). Despite these differences, we believe there is utility to quantifying the extent to which aggressive and prosocial behavior are elicited by social events that broadly threaten the fundamental need to belong (DeWall & Bushman, 2011). For the purpose of this manuscript, we refer to these social events, which include social rejection, exclusion, and ostracism, as “rejection” or “social rejection”. We acknowledge that this may be an imperfect solution given that subtle but important differences in these social experiences have been associated with distinct psychosocial outcomes. For example, explicit rejection and exclusion produces a sense of social loss and withdrawal from social contact, rumination about the social interaction, and increased feelings of agitation (Camacho et al., 2003; Maner et al., 2007; Molden et al., 2009). Ostracism produces a sense of failure to achieve social gain, reengagement in social contact, and thoughts about actions one should have taken (Higgins, 1997; Maner et al., 2007; Molden et al., 2009). Despite evidence for differential effects of explicit rejection and ostracism on affect, cognition, and behavior, prior meta-analyses have not considered these potentially critical task effects on prosocial and aggressive behavior. To test this relation, we performed a moderation analysis contrasting behavior elicited by social rejection/exclusion and ostracism. We grouped social rejection and exclusion together because of their conceptual similarity and because there were not a sufficient number of studies that discriminated between these two event types. Specifically, we coded studies as modeling rejection/exclusion when participants were explicitly provided the information that a group or individual did not wish to interact with them, or gave them explicit negative feedback. For example, in the “Get Acquainted Paradigm” (Twenge et al., 2001), after interacting with a small group of individuals, participants nominated two people with whom they wanted to perform a subsequent task. The participant then receives purported feedback that is rejecting/exclusionary or accepting/inclusionary - no one or everyone in the group wanted to work with them, respectively. We coded studies as modeling ostracism when participants were not explicitly told they were being excluded, but when they were ignored or implicitly left out of a group. For example, in a standard Cyberball paradigm, two ‘players’ stop passing the participant a ball after two initial tosses; however, participants are never explicitly told that they are being excluded from the game.

Reasons for rejection.

Another common paradigm difference is whether the participant knows why they are being rejected. For example, in studies that use the Cyberball or Get Acquainted Paradigm, participants do not typically know what has motivated their peers’ behavior. In other studies, participants are given very specific reasons about why they are being rejected (e.g., Kirkpatrick et al., 2002; Reijntjes et al., 2010). For example, participants receive feedback that “they do not seem fun to hang out with” after reading the participant’s profile (Reijntjes et al., 2010). It is possible that a participant may be able to reappraise, and therefore cope with the feedback, when they do not know why they are being rejected. This could, in turn, influence behavior. Thus, we also tested whether knowing or not knowing the reason for being rejected moderated the degree of prosocial or aggressive behavior. This moderator is independent of the “Rejection/Exclusion or Ostracism” moderator described above. Specifically, while studies in which the reason for rejection is known always modeled social rejection/exclusion (rather than ostracism), studies in which the reason for rejection was unknown included studies which modeled rejection/exclusion as well as ostracism. Thus, we are investigating the effects of knowing the reason of rejection rather regardless of whether the study modeled social rejection/exclusion or ostracism.

Behavioral paradigm moderators.

Similarly, given the large number of behavioral paradigms used to measure aggression or prosociallity, some may be more or less likely to induce such behavior. Thus, the moderating effects of behavioral paradigm features were explored.

Target’s identity.

Some studies have shown that rejected individuals are less likely to punish innocent strangers and more likely to punish perpetrators of rejection (Twenge et al., 2001). Yet other studies (DeWall, 2007; Rajchert & Winiewski, 2016; Twenge et al., 2001) and real-world examples, such as school shootings, demonstrate that aggression towards innocent bystanders is common. Similarly, Gerber and Wheeler’s meta-analysis (2009) found that individuals were more likely to exhibit prosocial behavior to an innocent stranger, but identity of the target had no effect on aggressive behavior. Because aggressive and prosocial behavior were subjectively coded in that meta-analysis based on the need being fulfilled (need for control or need to belong), and the majority of studies did not actually quantify behavior, it is still unclear whether target identity moderates objective prosocial or aggressive behavior. We test this relation in the current meta-analyses.

Primary or secondary domain.

Aggression in the primary domain is defined as an outcome that causes physiological harm or impacts a biologically unlearned need (e.g., delivering hot sauce and therefore causing pain; Ayduk et al., 2008). In contrast, aggression in the secondary domain causes psychological or higher order harm (e.g., stealing prize money from a partner; Reijntjes et al., 2011). Individuals judge acts that threaten a primary domain (e.g., physical violence) as more harmful and serious than acts that threaten a secondary domain (e.g., stealing money; Herzog, 2008; Herzog & Einat, 2016; Warr, 1989). Because hurting someone by way of a primary compared to a secondary domain may seem more egregious to participants, it is critical to test if domain moderates the effects of social rejection on participant behavior.

Cost of the behavior.

A common criticism of experimental paradigms used to model aggression is that, unlike real-world experiences, there are few consequences for a participant’s actions. However, studies that include a consequence for aggression have found that participants are willing to punish others even if there is a personal cost to doing so (e.g., dictator game; Will et al., 2015). Similarly, prosocial behavior in the real-world often comes at a cost to an individual (Kraft-Todd & Rand, 2019; Fehr & Fischbacher, 2003; Zhang & Epley, 2009), and perceptions of prosocial behavior are influenced by the cost of that behavior (Kawamura et al., 2021; Flynn & Adams, 2009). While some studies show that participants will help others despite a personal cost (Fehr & Fischbacher, 2003; Penner et al., 2005; Gneezy et al., 2012), others find helping is reduced by personal cost (Zhang & Epley, 2009; Sommerville et al., 2018; Andreoni & Miller, 2002; Bode et al., 2015). Therefore, we will also test whether facing consequences for one’s behavior moderates aggressive and prosocial behavior.

Summary and Goals of the Meta-Analyses

For our first aim, we conducted three separate meta-analyses that test the effects of social rejection on: 1) aggressive behavior; 2) prosocial behavior; and 3) behavior when given the options to be aggressive or prosocial. Second, we tested if demographic (e.g., age) and contextual factors (e.g., methods of rejection and behavioral paradigm factors) moderated the effect of rejection on aggression or prosocial behaviors in each of the three meta-analyses. Lastly, we tested for evidential value of findings and publication bias within the published literature.

Methods

Identification and Selection of Studies for Inclusion in Meta-Analyses

Relevant studies assessing the effects of social rejection on subsequent aggressive or prosocial behavior were identified with a search of PsycINFO, Pubmed, and Scopus databases. A grey literature search was performed for unpublished dissertations with a search of ProQuest. The search strategy involved combining search terms “Social Rejection” or “Social-Rejection” or “Social Inclusion” or “Social Exclusion” or “Ostracism” or “Peer Rejection” or “Peer Acceptance” or “Social Pain” and (Aggress* or “Aggressive” or “Aggression” or “Prosocial” or “Pro-social” or “Helping Behavior” or “Hurting Behavior”). In accordance with methods suggested by Card (2012), we utilized more specific search terms (e.g., “Helping Behavior”) due to the very high number of hits more general search terms (e.g., “Help”) yielded (e.g., over 54,000 hits total from PsychInfo (n=35,444), PubMed (n=1,896), and Scopus (n=16,658)). English-language articles from peer-reviewed journals with human subjects were included in these meta-analyses. Even when they may have been highly relevant (e.g., Warburton & Williams, 2005), we did not include book chapters or data discussed only in the context of reviews as they did not meet our criteria for peer-reviewed experimental studies. Additionally, studies included in a previous meta-analysis on social rejection and subsequent behavior (Gerber & Wheeler, 2009) were screened for inclusion criteria (see Supplementary Materials for full description of studies included/excluded from that meta-analysis). Finally, subscribers from four listservs (sponsored by the Society for Personality and Social Psychology, the Society of Experimental Social Psychology, the Society for Psychological Study of Social Issues, and the Social and Affective Neuroscience Society) were contacted with a request to submit published or unpublished manuscripts related to the topic. The search was conducted on September 1, 2021.

Four independent coders (MQ, TC, JLF, and HG) reviewed the titles and abstracts identified in the search to select eligible manuscripts. Because our goal was to understand the effect of social rejection on aggressive and prosocial behaviors, to be eligible for inclusion, studies must have had: (a) behaviorally manipulated social rejection in a laboratory setting (i.e., not evoked through remembering past rejection experiences); (b) included a manipulation of present rejection rather than imagining future rejection; (c) included a rejection condition as well as a comparison condition (acceptance/control); (d) used an objectively measured behavior as a dependent variable (i.e., not intentions or affect); and (e) had no intervention/manipulation after social rejection, as it may affect aggressive/prosocial behavioral responding (e.g., French, 2015; Warburton et al., 2006). For example, we excluded one prominent study in which control over exposure to unpleasant noises following social rejection was manipulated prior to subsequent aggressive behavior because need for control may have influenced aggression (Warburton et al., 2006). Furthermore, because we were interested in how social rejection affected behavior compared to a control group, studies must have (f) implemented a between-subjects design (i.e., a participant cannot be exposed to an acceptance or rejection manipulation and then another manipulation). We made this decision because it is possible that a participant would behave differently in a control condition if first exposed to social rejection. Moreover, our definition of prosocial behavior was behavior that conferred benefit to others, rather than benefit to oneself. Thus, studies that measured affiliative behavior such as unconscious body language/mimicry (e.g., Lakin et al., 2008) or the choice to work with others rather than alone (Maner et al., 2007) were excluded. Furthermore, if a study compared the effects of two types of rejection (Gaertner et al., 2008; Buckley et al, 2004) on aggressive/prosocial behavior the study was excluded. We made this decision given that such studies would not be theoretically comparable to the rest of the papers in these meta-analyses since participants experienced multiple types of rejection. Next, the full texts were examined to determine whether they met the aforementioned inclusion criteria.

Data Abstraction and Statistical Analyses

For studies that reported two measures of aggression, we used the average of the two measures/variables in our analyses. For example, in studies that used the Reaction Time Game, participants may choose the volume and duration of an aversive noise to deliver to the other player. If both the intensity and duration were reported, we used an average of the two. For studies that did not explicitly report on the number of participants per condition (k = 14), researchers were contacted with a request to provide necessary information. Using methods consistent with Gerber and Wheeler (2009), for studies whose authors failed to respond to requests for information (k = 12), sample sizes were estimated based on splitting the total reported N (see Tables 1, 2, and 3). Eleven of these studies reported using random assignment to groups, and were thus assumed to have an equal number of participants per condition. For the remaining one study (Chow et al., 2008), we also replicated methods used by Gerber and Wheeler (2009), and assumed an equal number of participants per condition. When possible, we selected data presented as raw means and standard deviations. If a study presented dichotomous outcomes (k = 1; Galbavá et al., 2021), data were treated as events ratios (Cochrane, 2021). If a study presented insufficient data to calculate effect sizes (k = 20), researchers were contacted with a request to provide necessary information. Eleven of the 20 authors responded with the requested information. The nine remaining papers were excluded from analyses (Aydin et al., 2017; DeWall et al., 2010; Gaertner et al., 2008; Jopp, 2015; Kothgassner et al., 2017; MacKenzie & Baumeister, 2019; Mallott et al., 2009; Smith, 2015; Watson-Jones, 2013).

Table 1.

Characteristics and Effect Size Information of Study Samples Included in the Aggression Meta-Analysis (k = 19)

Study	Total na	Standardized Mean Differenceb	Age	Type of Rejectionc	Target of Behavior	Domain	Publication Status	Cost to Participant	Knowledge of Reason for Rejectiond
Ayduk et al., 2008	129	0.34	Adult	R/E	Rejecter	Primary	Published	No	U
Beyer et al., 2014	34	0.44	Adult	Ostracism	Rejecter	Primary	Published	Yes	U
Bugni, 2016	94^	0.83	Adult	Ostracism	Stranger	Primary	Unpublished	Yes	U
DeBono & Muraven, 2014, Study 1	45	0.57	Adult Adult	Ostracism	Rejecter	Primary	Published	No	U
DeWall 2007, Study 1	67^	0.50	Adult	R/E	Stranger	Primary	Unpublished	Yes	U
DeWall 2007, Study 3	70	−0.44	Adult	R/E	Stranger	Primary	Unpublished	No	U
Gallardo-Pujol et al., 2013	57	0.64	Adult	Ostracism	Rejecter	Secondary	Published	Yes	U
Geniole et al., 2011	63	−0.19	Adult	Ostracism	Rejecter	Secondary	Published	Yes	U
Heppner et al., 2008 Study 2	40^	0.92	Adult	R/E	Rejecter	Primary	Published	Yes	U
Kirkpatrick et al., 2002, Study 1	88^	0.83	Adult	R/E	Rejecter	Primary	Published	No	K
Kirkpatrick et al., 2002, Study 2	74^	0.53	Adult	R/E	Rejecter	Primary	Published	No	K
Logue, 2006	81	0.17	Adult	R/E	Rejecter	Primary	Unpublished	Yes	U
Nicholls, 2014	58	−0.20	Adult	R/E	Stranger	Primary	Unpublished	No	U
Pickett et al., 2016	135	0.15	Adult	R/E	Rejecter	Primary	Published	Yes	K
Rajchert & Winiewski, Study 1	89	0.44	Adult	Ostracism	Stranger	Secondary	Published	No	U
Rajchert & Winiewski, Study 2	120	0.42	Adult	Ostracism	Rejecter	Primary	Published	No	U
Reiter-Scheidl et al., 2018	85	0.35	Adult	R/E	Rejecter	Primary	Published	No	U
Twenge et al., 2001, Study 5	34^	1.16	Adult	R/E	Stranger	Primary	Published	Yes	U
Webster & Kirpatrick, 2006	81	0.82	Adult	R/E	Rejecter	Primary	Published	No	K

Note:

^{a ^ =}studies in which the sample size for each group was estimated as an equal division of the total sample;

^bdirection of study effect is rejection group > acceptance/control group;

^cR/E = rejection/exclusion tasks;

^dU =reason for rejection unknown; K = reason for rejection known

Table 2.

Characteristics and Effect Size Information of Study Samples Included in the Prosocial Meta-Analysis (k = 7)

Study	Total na	Standardized Mean Differenceb	Agec	Type of Rejectiond	Target of Behavior	Domain	Publication Status	Cost to Participant	Knowledge of Reason for Rejectione
Coyne et al., 2011	40	0.57	Youth	Ostracism	Stranger	Secondary	Published	No	U
Lee & Park, 2019, Study 2	360^	0.35	Adult	Ostracism	Stranger	Secondary	Published	Yes	U
Maner et al., 2007, Study 5	24^	1.60	Adult	R/E	Rejecter	Secondary	Published	Yes	U
Twenge et al., 2007, Study 2	20^	1.86	Adult	R/E	Stranger	Primary	Published	Yes	U
Weerdmeester & Lange, 2019	34	0.18	Adult	R/E	Rejecter	Secondary	Published	Yes	U
Cuadrado et al., 2021	48	0.75	Adult	R/E	Rejecter	Secondary	Published	Yes	K
Choy et al., 2021	232	0.04	Adult	Ostracism	Stranger	Secondary	Published	Yes	U

Note:

^{a ^ =}studies in which the sample size for each group was estimated as an equal division of the total sample;

^bdirection of study effect is acceptance/control group > rejection group;

^cYouth = Individuals < 18 years of age;

^dR/E = rejection/exclusion tasks;

^eU =reason for rejection unknown; K = reason for rejection known

Table 3.

Characteristics and Effect Size Information of Study Samples Included in the Aggression/Prosocial Meta-Analysis (k = 15)

Study	Total na	Standardized Mean Differenceb	Agec	Type of Rejection	Target of Behavior	Domain	Publication Status	Cost to Participant	Knowledge of Reason for Rejectiond
Chen, 2012	55	0.87	Adult	R/E	Rejecter	Secondary	Unpublished	No	K
Chow et al., 2008	74^	1.33	Adult	Ostracism	Rejecter	Primary	Published	No	U
DeBono et al., 2020, Sample 1	133^	0.52	Adult	R/E	Rejecter	Secondary	Published	No	U
DeBono et al., 2020, Sample 2	145^	0.38	Adult	R/E	Stranger	Secondary	Published	Yes	K
DeWall 2007, Study 2	46	0.88	Adult	R/E	Stranger	Secondary	Unpublished	No	U
Leiro & Zwolinski, 2014	206	0.94	Adult	Ostracism	Rejecter	Secondary	Published	No	U
Liebke et al., 2018	56	−0.02	Adult	R/E	Stranger	Secondary	Published	No	U
Liu et al., 2018, Study 1	62	0.66	Adult	Ostracism	Rejecter	Secondary	Published	No	U
Rajchert et al., 2018	64	0.94	Adult	R/E	Rejecter	Secondary	Published	Yes	K
Reijntjes et al., 2010	121	1.55	Youth	R/E	Rejecter	Secondary	Published	No	K
Reijntjes et al., 2011	185	1.43	Youth	R/E	Rejecter	Secondary	Published	No	K
Scherer, 2012, Study 1	80	0.50	Adult	Ostracism	Rejecter	Secondary	Unpublished	No	U
Smith, 2011	99	0.23	Adult	Ostracism	Rejecter	Primary	Unpublished	Yes	U
Lutz & Schneider, 2020	140	0.14	Youth	R/E	Rejecter	Secondary	Published	No	U
Galbava et al., 2021	196	0.33	Adult	Ostracism	Rejecter	Secondary	Published	Yes	U

Note:

^{a ^ =}studies in which the sample size for each group was estimated as an equal division of the total sample;

^bdirection of study effect is rejection group > acceptance/control group;

^cYouth = Individuals < 18 years of age;

^dR/E = rejection/exclusion tasks;

^eU =reason for rejection unknown; K = reason for rejection known

Meta-analytic Procedure

We conducted three separate meta-analyses. The first included studies in which participants had the opportunity to engage in aggressive behavior. The second included studies in which participants had the opportunity to engage in prosocial behavior. The third included studies in which participants had the opportunity to engage in either aggressive or prosocial behavior. Analytic methods were identical for each meta-analysis and are therefore described jointly below. A supplemental meta-analysis which combined all studies, was also performed (see Supplement).

Using the Comprehensive Meta-Analysis (CMA) statistical software (Version 3.3.070), we computed standardized mean difference effect scores for each study (Borenstein et al., 2009). The effect size represents standardized mean group differences between the social rejection group and the comparison group with respect to subsequent aggressive and/or prosocial behavior. The effect size (d) was computed by taking the difference between the means of the comparison group and the rejected group and dividing by the pooled standard deviation of the two groups and multiplying by a correction factor based on the total samples size (Lipsey & Wilson, 2001). Each study’s effect size was weighted before aggregating by multiplying its values by the inverse of its variance. This allowed studies with larger samples to contribute more to the effect size estimate compared to studies with smaller samples (Martire et al., 2004). To ease interpretation of results, across all three meta-analyses (as well as the supplemental analysis), positive d’s indicate that social rejection elicited more aggressive or less prosocial behavior, whereas negative d’s indicate that social rejection elicited less aggressive or more prosocial behavior. Thus, for the first meta-analysis that tests the effect of rejection in studies of aggressive behavior, positive d’s reflect more, and negative d’s reflect less aggressive behavior. For the second meta-analysis that tests the effect of rejection in studies of prosocial behavior, positive d’s reflect less, and negative d’s reflect more prosocial behavior. For the third meta-analysis that tests the effect of rejection in studies including both aggressive and prosocial behavior, positive d’s reflect more aggressive/less prosocial behavior, and negative d’s reflect less aggressive/more prosocial behavior. I² values are reported to index the total percentage of variability in a set of effect sizes arising from between-study differences (e.g., I² = 50 indicates that 50% of variability is due to sampling error and 50% to between-study differences; Higgins & Thompson, 2002).

We estimated overall effects and heterogeneity in the effect sizes using random-effects meta-analytic modeling. We then tested whether heterogeneity was accounted for by six separate moderators (age, rejection technique, reason for rejection, target’s identity, domain of behavior, cost of behavior) using the analogue-to-analysis of variance procedure in CMA (Viechtbauer, 2005). Each of the six mixed-effects analyses explored whether relations between social rejection and aggressive/prosocial behavior varied depending on each proposed moderating factor. See Table 1–3 for the coding of moderator categories and Table S1 for the number of studies in each category across meta-analyses. Analyses were not conducted when there were no studies in one category. Results for analyses that included only one study in a category are reported in Supplementary Materials for completeness, but are not discussed further due to their very low statistical power.

Lastly, we aimed to test if published studies provided evidence for an underlying effect of evidential value and for publication bias. To assess for evidential value in published studies, we conducted a p-curve analysis. A p-curve analysis estimates the probability of obtaining each p-value if the null were true, and determines if findings have evidential value or were underpowered. T- or F-values and degrees of freedom from each study were entered into the publicly available p-curve calculator (Version 4.06; http://www.p-curve.com/). These analyses indicated evidential value (Statistical Power estimate = 87%; 90% CI 75% - 94%) (for more details see Figure 3). A random-effects version of Egger’s test (Egger et al., 1997) of funnel plot asymmetry (see Figure 4) also found no bias in distribution of effect sizes for small and large size studies (t(39)=1.51, p=0.14), suggesting that our meta analyses were not meaningfully affected by publication bias.

Figure 3.

P-curve analysis of evidential value.

Binomial Tests:

Studies contain evidential value (right skew): p < 0.01

Studies’ evidential value inadequacy (Flatter than 33% power): p = 0.87

Continuous Tests:

Studies contain evidential value (right skew): Z = −8.36, p = 0.00

Studies’ evidential value inadequacy (Flatter than 33% power): Z = 4.69, p > 0.99

Note: The p-curve analysis included all studies (k = 41) across the three separate meta-analyses.

Figure 4.

Funnel plot of Standard Error by Standard Difference in Means.

Note. The funnel plot includes all studies (k=41) across the three separate meta-analyses for ease of interpretation.

Results

The initial meta-analytic search yielded 2,121 titles and abstracts. Two researchers responded to our listserv postings, and neither report met eligibility criteria. Abstract screening revealed that 152 articles were eligible for full text review. The two-way mixed-effects intraclass correlation coefficient revealed good agreement among raters (ICC = 0.89, 95% CI = 0.87 – 0.91). Disagreement was resolved through discussion with the senior author (JMJ), who was uninvolved in the full text review process. After full text reviews and the exclusion of ten studies whose authors didn’t respond to our request for more data, 41 studies from 36 papers met inclusion criteria (see Figure 1 for more details and Prisma diagram). Total sample sizes ranged from 20 to 360, M = 94.24, SD = 66.46. Overall results from each meta-analysis are reported in Table 4.

Figure 1.

PRISMA flowchart showing selection of studies for meta-analysis on social rejection elicited aggressive and prosocial behavior.

Table 4.

Summary of Results for Meta-Analyses and Moderators

Meta-Analysis of Social Rejection Elicited Behavior	Overall effect (d)	Q-values of Moderation Effects on Behavior (Q-Values)
		Age	Type of Rejection	Knowledge of Reason for Rejection	Target of Behavior	Domain	Cost to Participant
Aggressive Behavior (k=19)	0.41*	N/A	0.10	0.90	0.08	0.29	0.30
Prosocial Behavior (k=7)	0.59*	0.01	3.97*	0.25	0.66	6.36*	0.01
Aggressive and Prosocial Behavior (k=15)	0.71*	0.79	0.11	3.24	2.21	3.17	0.01

^*significant effects (p < 0.05)

Note: cells in grey were underpowered to test moderating effects (see Table S1 for more details)

Meta-Analysis 1: Effect of Social Rejection on Aggressive Behavior

This meta-analysis included 19 studies (from 16 papers). The effect of rejection relative to comparison conditions (e.g., acceptance/control; direction of effect: rejection > acceptance/control) on aggressive behavior ranged from d = −0.44 to 1.16. A random-effects model (Borenstein et al., 2009) revealed an aggregate weighted effect size of d = 0.41, 95% Confidence Interval (CI) = 0.24 – 0.59, p < .0001. Thus, there is a medium effect (Cohen, 1988, 1992) such that social rejection potentiates subsequent aggression.

Moderators of the relation between social rejection and aggressive behavior.

There was heterogeneity in effect sizes for the relation between social rejection and aggressive behavior, Q(18) = 45.30, p < .0001, I² =60.27. No studies in this meta-analysis included youth, therefore we were unable to test for moderating effects of age (see Table S1 for more details). Thus, analyses were conducted to test for five of the six proposed moderators.

Methods of rejection.

Type of rejection (rejection/exclusion or ostracism; Q(1) = 0.10, p = 0.75) and knowledge of reason for rejection (known or unknown; Q(1) = 0.90, p = 0.34) failed to moderate aggressive behavior.

Behavioral paradigm moderators.

There were no effects of the target’s identity (rejecter or innocent bystander), Q(1) = 0.08, p = 0.77, costs of the behavior, Q(1) = 0.29, p = 0.59, nor the type of outcome domain of the aggression (primary or secondary), Q(1) = 0.30, p=0.58 on subsequent behavior.

Meta-Analysis 2: Effect of Social Rejection on Prosocial Behavior

This meta-analysis included 7 studies (from 7 papers). The effect of rejection relative to comparison conditions (e.g., acceptance/control) (direction of effect: acceptance/control > rejection) on prosocial behavior ranged from d = 0.04 to 1.86. A random-effects model (Borenstein et al., 2009) revealed an aggregate weighted effect size of d = 0.59, 95% Confidence Interval (CI) = 0.23 – 0.95, p < .0001. Thus, there is a medium effect (Cohen, 1988, 1992) such that social rejection attenuates prosocial behavior.

Moderators of relations between social rejection and subsequent prosocial behaviors.

There was heterogeneity in effect sizes for the relation between social rejection and subsequent prosocial behavior, Q(6) = 23.01, p = .0001, I² =73.92. Several moderators (age, reason for rejection, domain, and cost of behavior) included only one study in a category (see Table S1 for more details). Results for these analyses are reported in Supplementary Materials for completeness, but are not discussed further due to their very low statistical power. Thus, analyses were conducted to test for two of the six proposed moderators.

Methods of rejection.

There was an effect of type of rejection (rejection/exclusion or ostracism) on prosocial behavior, Q(1) = 3.97, p = 0.05. Specifically, although participants exhibited attenuated prosocial behavior when they were explicitly socially rejected/excluded, d = 1.02, 95% CI= 0.32 – 1.73, p = .004, there was no effect of ostracism on prosocial behavior, d = 0.26, 95% CI= 0.01 – 0.52, p = .06.

Behavioral paradigm moderators.

There were no effects of target identity (innocent bystander or rejecter) on prosocial behavior, Q(1) = 0.66, p = 0.42.

Meta-Analysis 3: Effects of Social Rejection on Aggressive and Prosocial Behavior

For this meta-analysis, we also included a subset of studies that allowed participants to behave either aggressively or prosocially. For example, if a participant has the ability to donate money or take away money from another player (Liebke et al., 2018), or give a positive or negative job evaluation (DeWall et al., 2007). Although we acknowledge that aggressive and prosocial behavior are not opposite sides of the same coin, such studies provide an opportunity to test whether rejection elicits prosocial or aggressive behavior when they are given both options. This meta-analysis included 15 studies (from 14 papers). The effect of rejection relative to comparison conditions (e.g., acceptance/control) (direction of effect: rejection > acceptance/control) on aggressive behavior ranged from d = −0.02 to 1.55. A random-effects model (Borenstein et al., 2009) revealed an aggregate weighted effect size of d = 0.71, 95% Confidence Interval (CI) = 0.46 – 0.96, p < .0001. Thus, there is a medium effect (Cohen, 1988, 1992) such that social rejection potentiated subsequent aggression.

Moderators of relations between social rejection and subsequent behaviors.

There was heterogeneity in effect sizes for the relation between social rejection and subsequent behavior, Q(14) = 77.95, p < .0001, I² =82.04. Thus, analyses were conducted to test for the six proposed moderators.

Age of sample moderation effects.

Age of the sample (youth or adult) did not moderate the effect of rejection on subsequent behavior, Q(1) = 0.79, p = 0.37.

Methods of rejection.

There were no effects of type of rejection (rejection/exclusion or ostracism; Q(1) = 0.11, p = 0.75), nor knowledge of reason for rejection (known or unknown; Q(1) = 3.24, p = 0.07) on subsequent behavior.

Behavioral paradigm moderators.

There were no effects of the target’s identity (rejecter or innocent bystander; Q(1) = 2.21, p = 0.14), costs of the behavior (Q(1) = 3.17, p = 0.08), nor the type of outcome domain of the aggression (primary or secondary; Q(1) = 0.01, p=0.91) on subsequent behavior.

Discussion

Results across three meta-analyses support the idea that rejection strongly potentiates aggression and attenuates prosocial behavior. This is not due to participants lacking behavioral choices. In fact, in meta-analysis three, we demonstrate that individuals exhibit aggression even when they have the opportunity to behave prosocially. Moreover, these effects are quite robust, as they are largely insensitive to moderating factors ranging from age to numerous contextual features that vary across experimental paradigms.

Rejection resulted in more aggressive, or less prosocial, behavior with effect sizes that were medium in size. This contradicts the theory that individuals behave prosocially to reduce the likelihood of future rejection (DeBono et al., 2017; DeWall & Bushman, 2011; Leary et al., 2003). This discrepancy may be linked to the fact that rejection-elicited prosocial behavior has often been operationalized as affiliative behavior (e.g., nonverbal mimicry; Lee & Shrum, 2012), rather than behavior that directly confers explicit personal benefit. For example, Gerber and Wheeler (2009) found that individuals behaved in a prosocial manner following rejection if they had the opportunity to satisfy both their need to belong and for control simultaneously. However, their meta-analysis included studies that measured affiliation, rather than purely overt prosocial behavior. We found marginally non-significant evidence for rejection-elicited prosocial behaviors such as sharing money, resources, or providing positive feedback. In fact, not a single study we included demonstrated potentiation of rejection-elicited prosocial behavior and only four studies (DeWall, 2007; Geniole et al., 2011; Liebke et al., 2018; Nicholls, 2014) showed evidence for attenuated aggressive behavior following rejection. Thus, the mechanisms driving more overt prosocial behavior may not be the same as those leading to affiliative behavior. Given the small number of studies included in this meta-analysis (k=7), future work may illuminate under which conditions, if any, rejection elicits overt prosocial behavior. Importantly, in our third meta-analysis which included studies in which participants were given the opportunity to behave aggressively or prosocially, we found that rejection led to increased aggressive behavior. This finding makes a critical contribution to the literature, showing that rejection-elicited aggression is not simply a result of not having the opportunity to behave prosocially.

We also found limited evidence that contextual factors moderated the effects of rejection on aggressive behavior. For instance, knowledge for reason for rejection (known/unknown), the target identity (rejecter/innocent stranger), and the cost of the behavior failed to moderate the effect of rejection on aggressive or prosocial behavior. These results were consistent with Gerber and Wheeler (2009) who found that target identity influenced prosocial, but not aggressive, responding. Thus, rejected individuals behave aggressively regardless of who they were rejected by, whether they know why they are rejected, and whether they perceive a cost for their behavior. This supports our overall finding that rejection is a robust elicitor of aggressive behavior.

However, in meta-analysis two, we found that method of rejection (social rejection/exclusion or ostracism) moderated the relation between rejection and prosocial behavior. Specifically, when individuals were explicitly socially rejected/excluded, prosocial behavior was attenuated. However, ostracism failed to impact prosocial behavior. Gerber and Wheeler (2009) found that ostracism had a larger negative effect on participants’ self-esteem than explicit social rejection. Moreover, they found that ostracism, but not explicit rejection, led to a moderate increase in arousal. However, they did not test if this effect on self-esteem and arousal led to a change in aggressive or prosocial responding. Our findings suggest that the impact of ostracism on self-esteem and arousal does not subsequently influence behavioral responding. Additionally, our results support prior work that posits that ostracism and explicit rejection/exclusion are separable constructs that can result in different psychological outcomes (Wesselmann & Williams, 2017). For example, Lutz and Schneider (2020) found that rejected individuals were more threatened in their needs for belonging and self-esteem than ostracized individuals. Our results also suggest that prosocial behavior may be more sensitive to contextual factors than aggressive behavior. However, fewer studies were included in the meta-analysis that tested effects of social rejection on prosocial behavior (k = 7) compared to those included in the meta-analysis that tested effects of social rejection of aggressive (k = 19) or both aggressive and pro-social behavior (k = 15), where no moderating effects of rejection/exclusion versus ostracism were found. Thus, results should be interpreted with caution.

Limitations and Future Directions

Despite its novelty and many strengths, these meta-analyses are not without limitations. First, only seven papers tested the effect of rejection on prosocial behavior in the absence of aggressive response options. Given prevailing support for the theory that rejection-elicited prosocial behavior is evolutionarily adaptive, it is surprising that this number is so low. Further research is needed to investigate under what conditions, if any, overt prosocial behavior is elicited by rejection. Identifying moderators that promote prosocial, rather than aggressive behavior, following rejection, could inform novel interventions aimed at preventing aggression. Similarly, we utilized four databases to search for studies (ProQuest, Scopus, Pubmed, and PsycINFO) and requested unpublished data from listservs associated with four relevant research societies. Although we are confident that our meta-analyses comprehensively include the relevant literature, it is possible that a wider search may have identified additional unpublished data.

Moreover, while we tested for the moderating effects of age on rejection-elicited behavior in meta-analyses three, this analysis was admittedly underpowered. The vast majority of studies (37/41) included adult only samples. Thus, while age failed to moderate the relation between social rejection and behavior, results should be interpreted with caution (see Supplement for better powered exploratory analyses across all studies). This is an important limitation because peer-based aggressive behavior typically increases during adolescence (Brown & Larson, 2009; Moffitt, 1993), and violence is a leading cause of death for adolescents and young adults (CDC, 2014). Moreover, moderators that influence the relation between rejection and aggression may differ in adolescents and adults. For example, given that there are vast neurodevelopmental changes that occur during adolescence, especially in brain regions implicated in social cognition (Choudhury et al., 2006; Blakemore, 2012, 2008), different contexts may be more salient and evocative for adolescents than for adults. Therefore, interventions for aggressive behavior in adolescents and adults may need to target different cognitive factors. Furthermore, because adolescents who exhibit aggressive behavior have increased risk for aggression in adulthood (Bierman & Wargo, 1995), understanding neural mechanisms that promote adolescent peer-based aggression would inform novel interventions aimed at treating adolescent aggression, before it becomes an instantiated problem. Thus, it is crucial to identify the mechanisms underlying rejection-elicited aggressive behavior in adolescence, and the contextual factors that influence that behavior.

Additionally, we were unable to test for several individual difference factors that may influence behavioral responding following rejection. For example, no study reported gender-specific outcomes. This is a crucial future direction as prior experimental and sociological studies have found reliable gender differences in self-reported and actual peer-based aggressive behavior, especially in adolescence (Bettencourt et al., 1997; Harrendorf et al., 2010; Björkqvist, 2018; Österman et al., 1998; Card et al., 2008). Additionally, we were unable to test for the moderating effects of trait-related differences known to influence rejection-elicited aggression, such as self-esteem (Kirkpatrick et al., 2002), narcissism (Twenge & Campbell, 2003), and rejection sensitivity (Ayduk et al., 1999, 2008; Sommerfeld & Shechory-Bitton, 2020). Therefore, future studies should test if personality traits and susceptibility to rejection moderate the effects of rejection on aggressive/prosocial behavior. Lastly, factors related to the psychological experience of the task, such as perceived control over rejection and behavior, and emotional reactions to rejection, may also influence behavior following social rejection (Gerber & Wheeler, 2009). However, published manuscripts do not typically report results from every variable collected or analysis conducted. Because of this, meta-analyses that aim to test factors that differentially influence psychological experiences elicited by a task, can only include studies that directly report those results, or have to make subjective assumptions about the experiences paradigms might evoke. Moving towards an open science framework, where all variables collected within a study (e.g., gender; emotional reactions to the task) are accessible, is needed to better understand such important potential moderators.

Similarly, although we were able to test whether broad paradigm-related factors (e.g. rejection/exclusion or ostracism, domain of the behavior) moderated the relation between rejection and aggressive or prosocial behavior, we were unable to test how more specific paradigms impacted behavior. This was due to the high degree of variability in methods and paradigms used (six different prosocial behavior tasks, 15 different aggression tasks). Thus, an important future direction for the field is to identify gold standard paradigms.

Lastly, we had also hoped to describe potential neural mechanisms driving rejection-elicited aggression. The dearth of research on this topic prevented a systematic review of the literature (see narrative review in Supplement). Aggressive behavior elicited by social rejection likely results from complex neural response patterns, and much more work is needed to map these relations. Doing so may ultimately allow us to identify novel treatment targets to prevent aggressive behavior.

Figure 2.

Standard Difference in Means by Study for a) Meta-Analysis 1: Effect of Social Rejection on Aggressive Behavior; b) Meta-Analysis 2: Effect of Social Rejection on Prosocial Behavior; c) Meta-Analysis 3: Effects of Social Rejection on Aggressive and Prosocial Behavior.

Note: Effect sizes < 0.00 reflect attenuated aggressive/potentiated prosocial behavior; effect sizes > 0.00 demonstrate potentiated aggressive/attenuated prosocial behavior.

Statement of Relevance:

To our knowledge, these meta-analyses are the first to directly test whether social rejection elicits aggressive or prosocial behavior. By including a comprehensive collection of both published and unpublished research studies, and examining a wide variety of previously untested moderators, we show that social rejection robustly elicits aggressive behavior and inhibits prosocial behavior. Additionally, we demonstrate that aggressive behavior following social rejection is not simply a function of limited choices in response options. In fact, aggressive behavior was evoked even when the option to engage in prosocial behavior was provided. Furthermore, we conducted a comprehensive narrative review of the neural mechanisms underlying social rejection-elicited aggressive and prosocial behavior to supplement primary analyses. Overall, we believe that our work makes a critical theoretical contribution to the field.