Everyday, we visually perceive people not only in isolation but also in groups. Yet, visual person perception research typically focuses on inferences made about isolated individuals. By integrating social vision and visual ensemble coding, we present novel evidence that (a) perceivers rapidly (500 ms) extract a group’s ratio of men to women and (b) both explicit judgments of threat and indirect evaluative priming of threat increase as the ratio of men to women in a group increases. Furthermore, participants’ estimates of the number of men, and not perceived men’s coalition, mediate the relationship between the ratio of men to women and threat judgments. These findings demonstrate the remarkable efficiency of perceiving a group’s sex ratio and downstream evaluative inferences made from these percepts. Overall, this work advances person perception research into the novel domain of people perception, revealing how the visually perceived sex ratio of groups impacts social judgments.

In New York City’s Bryant Park, a man stands with two clickers in hand, tallying the number of people who enter the park—on one, the number of men and the other, the number of women. Dan Biederman, president of the Bryant Park Corporation, considers these daily counts to be important indices of civic health, “Go to any public space in the world. If it’s skewing overwhelmingly male, get out as soon as possible” (Paumgarten, 2007, p. 4). This quote, along with the sociological evidence that inspired it (Whyte, 1980), implicates a group’s sex ratio as a salient cue to threat and safety. Given the importance of ensuring physical safety for survival (Neuberg, Kenrick, & Schaller, 2011), we propose that naive perceivers are likely to be equipped to rapidly and accurately judge the ratio of men to women (i.e., sex ratio) in a group and to draw evaluative inferences about groups based on sex composition. Here, we combine insights from social psychology and vision science to test novel questions regarding whether people can accurately report a group’s sex ratio and, in turn, whether evaluative judgments of threat are tethered to a group’s sex composition.

After merely glimpsing a face, human perceivers rapidly and efficiently decipher others’ social category memberships (e.g., race, sex), emotional states (e.g., happiness, fear), and interpersonal traits (e.g., trustworthiness, dominance; Ambady, Bernieri, & Richeson, 2000; Freeman, Johnson, Adams, & Ambady, 2012; Todorov, 2012). Most of the work supporting these conclusions probed perceivers’ judgments of faces presented in isolation. This approach has been valuable in revealing the underlying mechanisms and downstream consequences of face perception, yet it also differs from how observers encounter others in everyday life—in groups. Although groups play a central role in our social world (Lewin, 1951), interpersonal relations (Baumeister & Leary, 1995), and social identity (Tajfel, 1982), we know surprisingly little about how groups are visually perceived. As such, contemporary researchers have called for a shift in focus from “person perception” (i.e., the processes and outcomes associated with perceiving individuals) to “people perception” (i.e., the processes and outcomes associated with perceiving groups; Phillips, Weisbuch, & Ambady, 2014).

Alongside social psychologists’ call for increased attention to perceptions of social groups, vision scientists have developed methods for studying the perception of stimulus groups or ensembles. Recent evidence indicates that observers quickly and accurately perceive sets of objects through ensemble coding, whereby perceivers simultaneously encode multiple stimuli and represent them as a summary percept (Alvarez, 2011). Ensemble coding appears to be a general process of visual perception, with perceivers forming accurate representations of an ensemble’s mean brightness (Bauer, 2009), physical orientation (Parkes, Lund, Angelucci, Solomon, & Morgan, 2001), and even spatial location (Alvarez & Oliva, 2008). This perceptual acuity also extends to summary representations of social stimuli such as faces and bodies, including walk motion (Sweeny, Haroz, & Whitney, 2013), eye-gaze direction (Sweeny & Whitney, 2014), and social category membership (Haberman & Whitney, 2007; Thornton, Srismith, Oxner, & Hayward, 2014). For instance, in one ensemble coding study, participants viewed a set of faces for 500 ms, morphed between two anchors—happy and sad. They then decided whether a newly presented face, taken from the morph distribution, was more or less happy than the average emotion of the group. Results indicated that accurate discrimination (defined as >75% correct) occurred for faces that were only a few morph degrees away from the actual mean emotion of the ensemble (Haberman & Whitney, 2007). Thus, ensemble coding facilitates the rapid and accurate representation of ensembles; however, the downstream evaluative consequences have yet to be fully explored.

Section: Choose Top of page Abstract << Study 1 Method Results Study 2 Method Results Study 3 Method Results Discussion References Present Research In three studies, we examined how a group’s sex ratio impacts perceivers’ ratings of that group’s threat. Because men are stereotypically associated with threat and aggression (Archer, 2004; Eagly & Steffen, 1986), we hypothesized that groups would be judged overall as more threatening as the ratio of men to women increased. We tested this hypothesis by using both explicit (Study 1) and a more indirect (Study 2) measures. Finally, in Study 3, we examined two potential mediators for why groups with more men are judged as more threatening. On one hand, based solely on numeric probability, more men within an ensemble might directly impact judgments of threat. On the other hand, more men within an ensemble might increase the perceived coalition of the men, which may then explain higher threat evaluations. Indeed, according to the male warrior hypothesis, men are more attuned to intergroup conflicts and are more aggressive when faced with intergroup conflicts (McDonald, Navarrete, & Vugt, 2012; Vugt, 2009). Furthermore, large group size conveys dominance and higher status (Cao & Banaji, 2017; Pun, Birch, & Baron, 2016). This possibility implies that perceived group coalition might mediate judgments of threat for the group. Overall, this work is among the first empirical studies to bridge social vision and ensemble coding research (also see Phillips et al., 2014), providing novel insights about the accuracy and social evaluative consequences of visual group perception.

Study 1 Section: Choose Top of page Abstract Study 1 << Method Results Study 2 Method Results Study 3 Method Results Discussion References

Method Section: Choose Top of page Abstract Study 1 Method << Results Study 2 Method Results Study 3 Method Results Discussion References Participants Power analyses, utilizing parameters derived from our models, were run with MPlus (Version 7.4) simulations following guidelines by Muthén and Muthén (2002). Given the high power afforded by our within-subject designs, these simulations determined that at least 20 participants were needed to detect an effect of the size estimated in our study at 90% power. Based on our lab’s prior rate of data collection, we collected data for 2 weeks to reach the desired number of participants. In fact, we exceeded the target number, recruiting 82 undergraduate participants (50 women, M age = 19.19, SD age = 1.55). Stimuli Ensemble stimuli consisted of 3 × 4 arrays depicting faces of White men and women from the Chicago Face Database Version 1.0 (36 men and 38 women; Ma, Correll, & Wittenbrink, 2015). We used a custom Python script to generate arrays, such that faces were randomly selected without replacement within trial, but with replacement between trials, from the available bank of face stimuli. The location of each face was randomized within each ensemble. Ensembles varied in the ratio of men to women from 0:12 to 12:0 in increments of one. For each of the 13 possible sex ratios, we generated a set of 300 unique ensembles, resulting in a total of 3,900 ensembles from which to sample. Procedure Stimuli were presented, and responses were recorded using Inquisit 4.0.6.0 software (Inquisit, 2014). The protocol consisted of two blocks presented in counterbalanced order. Each trial began with a fixation cross (500 ms), followed by a randomly selected ensemble (500 ms) about which participants provided judgments. In one block, participants estimated the number of men in each ensemble (0 men–12 men). In the other block, participants rated each ensemble along four evaluative dimensions: invitingness (1 = not at all inviting, 7 = extremely inviting), warmth (1 = extremely cold, 7 = extremely warm), hostility (1 = not at all hostile, 7 = extremely hostile), and threat (1 = not at all threatening, 7 = extremely threatening). Each block included 260 arrays randomly selected from the total population of 3,900 ensembles, with 20 ensembles for each level of sex ratio. After completing their judgments, participants provided demographic information and were debriefed.

Results Section: Choose Top of page Abstract Study 1 Method Results << Study 2 Method Results Study 3 Method Results Discussion References We analyzed data both including and excluding participants who had incomplete data (n = 13), and the results were unchanged regarding direction and significance. We therefore included data from all participants, but analyses that exclude participants who have incomplete data are reported in Online Supplemental Material. We hypothesized that estimates of the number of men and evaluative judgments of ensembles would vary as the ratio of men to women increased. Specifically, we predicted that numeric estimates would reflect the ratio of men to women in each ensemble and that overall judgments of threat would increase as the ratio of men to women increased. We used the R packages “lme4” and “lmerTest” to compute hierarchical linear models which account for within-subject variation and nesting within participant (Bates, Mächler, Bolker, & Walker, 2015; Kuznetsova, Brockhoff, & Christensen, 2015). All models included random intercept, random slopes with residual maximum likelihood estimation. All coefficients are unstandardized. We also tested potential interactions with participant gender (effect coded as −1 = men, 1 = women); in all studies, participant gender did not qualify any effect unless otherwise noted. First, we examined whether participants accurately perceived the sex ratio of ensembles by regressing participant’s numeric estimates of men onto sex ratio (mean centered). Consistent with hypotheses, as the ratio of men to women increased, so did participants’ numeric estimates of men in an ensemble, B = .52, SE = .02, t(77.63) = 28.10, p < .001, 95% CIs [.48, .55].1 Next, we tested whether the actual number of men in each ensemble affected threat judgments of groups. To achieve this, we calculated a mean threat score by averaging the four evaluative items (with warmth and invitingness reverse coded), such that higher scores represented higher threat. The composite of these 4 items showed high within-subject reliability (.94) based on the method outlined by Cranford et al. (2006). Results for individual measures are reported in Online Supplemental Material and align with the findings presented here. Regressing the overall threat judgment score onto actual sex ratio (mean centered), we found that as the ratio of men to women increased, threat judgments also increased, B = .06, SE = .01, t (78.65) = 6.55, p < .001, 95% CIs [.04, .07]. In summary, consistent with our hypotheses, participant’s numeric estimates of men in an ensemble corresponded to the actual number of men depicted in each ensemble. Importantly, perceivers also judged the group to be more threatening as the ratio of men to women increased. In Study 2, we sought to extend these findings beyond self-report measures by using a more indirect measure of evaluation: the affect misattribution procedure (AMP; Payne, Burkley, & Stokes, 2008). The AMP has been used as an indirect measure of both affective (e.g., fear) and semantic (e.g., personality traits, animacy) concepts, and a previous meta-analysis indicates it is a valid and reliable means of measuring attitudes on par with other more implicit tasks (Deutsch & Gawronski, 2009; Gawronski & Ye, 2014; Oikawa, Aarts, & Oikawa, 2011; Payne & Lundberg, 2014). Aligning with results from Study 1, we hypothesized that the arrays with higher ratios of men to women will prime the concept of threat, leading to greater misattribution of threat for those targets.

Study 2 Section: Choose Top of page Abstract Study 1 Method Results Study 2 << Method Results Study 3 Method Results Discussion References

Method Section: Choose Top of page Abstract Study 1 Method Results Study 2 Method << Results Study 3 Method Results Discussion References Participants As in Study 1, power analyses using MPlus (Version 7.4) simulations were conducted with parameters derived from our models. Based on these simulations, we determined that a sample size of at least 20 was needed to detect an effect of the size estimated in our study at 90% power. Again, based on our lab’s prior rate of data collection, we collected data for 2 weeks to reach the desired number of participants. During this time, we exceed the desired number, recruiting 102 undergraduate participants (74 women, M age = 18.92, SD age = 1.23). Stimuli The AMP utilizes two types of stimuli: primes and targets. Prime stimuli consisted of ensembles from Study 1 depicting five different sex ratios of men to women: 0:12, 3:9, 6:6, 9:3, and 12:0 (300 unique arrays per sex ratio). Target stimuli for the AMP traditionally consist of a set of Chinese characters (Payne et al., 2008). Many participants in our subject pool are familiar with Chinese, so we instead generated a set of 360 target stimuli that depicted unique combinations of Thai letters. Procedure Utilizing the AMP script developed for Inquisit software (Borchert, 2016), participants viewed pairs of images flashed sequentially on a computer screen. Participants’ task was to judge whether the second image—the target stimulus—appeared to be more threatening or less threatening than the average stimulus by pressing “E” or “I” on the keyboard (counterbalanced across participants). Each trial consisted of the following: a prime stimulus (i.e., an ensemble) presented for 75 ms, a blank screen for 125 ms, a target stimulus (i.e., Thai letters) for 100 ms, and a visual mask that remained on screen until the participant made a judgment. Participants completed 10 practice trials followed by 360 critical trials (randomized), consisting of 60 trials for each sex ratio and 60 additional trials that contained no prime. Upon completion, participants provided demographic information and indicated whether or not they knew Thai.

Results Section: Choose Top of page Abstract Study 1 Method Results Study 2 Method Results << Study 3 Method Results Discussion References Eight participants were excluded for providing invariant responses (e.g., rated all stimuli as more threatening), and one participant was excluded for indicating they knew Thai, leaving a total sample of 93 participants. We analyzed our data using the R packages lme4 and lmerTest, ensuring our models fit a binomial distribution. We coded AMP responses dichotomously (0 = less threatening, 1 = more threatening) and the sex ratio of prime stimuli continuously (from 0 [0 men:12 women] to 4 [12 men:0 women], mean centered).2 We then regressed AMP responses, a dichotomous outcome (0 = less threatening, 1 = more threatening), onto the prime stimulus sex ratio. As predicted, for each unit increase in the ratio scale of men to women, the odds ratio (OR) for categorizing the target as more threatening, compared to less threatening, increased by 1.07, B = .07, SE = .02, z = 2.89, p = .004, 95% CIs [.02, .12], OR = 1.07. This effect was qualified by a higher order quadratic trend (see Figure 1),3 whereby the likelihood of a target stimulus being judged as more threatening was relatively equal when women outnumbered men (i.e., 0 men:12 women, 3 men:9 women) and when men and women were equal (i.e., 6 men:6 women), but increased exponentially as men became the majority in an ensemble, B = .02, SE = .01, z = 2.53, p = .011, 95% CIs [.005, .04], OR = 1.02. Thus, when exposed to majority male groups, participants judged a subsequent target to be more threatening. These results corroborate the explicit evaluations from Study 1. Download Open in new tab Download in PowerPoint

Study 3 Section: Choose Top of page Abstract Study 1 Method Results Study 2 Method Results Study 3 << Method Results Discussion References In Study 3, we examined two potential underlying processes accounting for why a group of men may be viewed as more threatening. As argued in the introduction, one straightforward explanation maybe that because men are both stereotypically and verdically more associated with violence and aggression (Archer, 2004; Eagly & Steffen, 1986), perceived threat may increase as the ratio of men to women increases. Alternately, because coalitional aggression and violence is often enacted by men (McDonald et al., 2012; Vugt, 2009), the degree to which the men in an array are perceived to be a coalition may account for higher threat evaluations. We examine these two hypotheses in Study 3 by testing whether perceived number of men or the perceived coalition of the men in the group mediates the relationship between the sex ratio of men to women and threat evaluations.

Method Section: Choose Top of page Abstract Study 1 Method Results Study 2 Method Results Study 3 Method << Results Discussion References Participants We replicated the recruitment strategy for Study 1, collecting data for a total of 2 weeks. In this time, we recruited 99 undergraduate participants (68 women, M age = 19.42, SD age = 1.28). Stimuli We utilized the stimuli from Study 1; however, we excluded ensembles with men to women ratios of 0:12, 1:11, 2:10 and 10:2, 11:1, 12:0 because groups of one or two individuals would be unlikely to constitute a coalition, thereby rendering our measure meaningless for these groups. Thus, our pool of potential ensembles was 2,100, with 300 unique ensembles per sex ratio. Procedure Overall the protocol replicated Study 1, with two notable differences. First, the blocks were presented in sequential order, with numeric estimations first, followed by evaluative judgments, and a new block assessing perceived coalition of the group of men. For this coalition judgment, participants were instructed to respond to measures we adapted from the entitativity literature (Ip, Chiu, & Wan, 2006), “To what extent do the men seem more like a group rather than just a bunch of individuals?” (1 = Much more like a collection of individuals than a group, 7 = Much more like a group than a collection of individuals).4 To allow for tests of mediation, participants judged the same set of randomly selected ensembles from the total pool of potential ensembles, selected in the first block, in each subsequent block, and presented in a random order within each block. After completing the task, participants provided demographic information and were debriefed.

Results Section: Choose Top of page Abstract Study 1 Method Results Study 2 Method Results Study 3 Method Results << Discussion References As in Study 1, we hypothesized that numeric estimates of the number of men and threat judgments for ensembles would increase as the sex ratio of men to women increased. We also hypothesized that as the sex ratio of men to women increased, judgments of men’s coalition would also increase. Finally, we tested whether numeric estimates of the number of men or perceived men’s coalition mediated threat judgments. We used the same R packages and analytic strategy as Study 1, using hierarchical linear models which account for within-subject variation and nesting within participant. To test our hypotheses regarding mediation, we used the “mediation” package in R (Tingley, Yamamoto, Hirose, Keele, & Imai, 2013) and examined two different mediation models, one with the perceived number of men as the mediator of threat judgments and the other with judgments of men’s coalition as the mediator. For each model run, we set the simulation parameter at 1,000. No participants had incomplete data. Results for numeric estimations and threat judgments followed the same pattern found in Study 1. In terms of numeric estimations, as the ratio of men to women increased, so did participants’ numeric estimates of men in the ensemble, B = .33, SE = .02, t(89.12) = 20.21, p < .001, 95% CIs [.30, .37]. Unlike Study 1, the participant gender by sex ratio interaction did not reach significance, B = .004, SE = .04, t(88.55) = 0.11, p = .911. For threat judgments (which had high within-subject reliability, .97), we found that as the ratio of men to women increased, so did threat judgments, B = .04, SE = .01, t(83.51) = 6.04, p < .001, 95% CIs [.02, .05]. Consistent with our hypotheses, we found that as the ratio of men to women increased, participants judged the men in the group to represent a coalition, rather than a collection of individuals, B = .09, SE = .01, t(87.46) = 7.09, p < .001, 95% CIs [.07, .12]. Lastly, we tested whether perceived number of men and judgments of men’s coalition predicted threat judgments. As numeric estimates of men in the ensemble increased, so did threat judgments, B = .03, SE = .004, t(66.55) = 5.66, p < .001, 95% CIs [.02, .04]; however, the perceived coalition of the men in the ensemble did not predict higher threat judgments, B = .01, SE = .01, t(79.36) = 1.24, p = .218, 95% CIs [−.001, .003]. Finally, we examined two different mediation models to determine whether the relationship between sex ratio and threat judgments was mediated by the perceived number of men or judgments of men’s coalition. We conducted mediation tests using multilevel mediation analyses (see Zhang et al., 2016 for review), examining the average causal mediation effect (ACME) and the average direct effect (ADE). First, we tested whether participants’ numeric estimations of the number of men were a mediator of the relationship between sex ratio and threat judgments. Results indicated that the ACME was significant, ACME = .01, p < .001, 95% CIs [.002, .01]. We also found evidence for a significant ADE = .03, p < .001, 95% CIs [.02, .04]. Overall, the significant ACME indicates that participants’ perceived number of men mediated the relationship between sex ratio (i.e., the actual number of men) and threat judgments. We next tested whether the relation between sex ratio and threat judgments was mediated by men’s coalition judgments. In this model, the ACME did not reach significance, ACME = .0003, p = .610, 95% CIs [−.0001, .001]; there was however a significant ADE = .04, p < .001, 95% CIs [.02, .05]. Thus, we did not find evidence that men’s coalition judgments mediated the relationship between the actual number of men and threat judgments.

Discussion Section: Choose Top of page Abstract Study 1 Method Results Study 2 Method Results Study 3 Method Results Discussion << References Using a social vision approach, we found that participants efficiently extracted the sex ratio from an ensemble of faces and made trait judgments of that ensemble that were consistent with prevailing gender stereotypes regarding aggression and threat. Specifically, participants judged groups as more threatening as the ratio of men to women increased, and these judgments were evident in both explicit (Study 1) and indirect evaluations (Study 2). Results also showed that the relationship between the ratio of men to women in an ensemble and higher threat evaluations was mediated by the perceived number of men in the group but not by perceived men’s coalition (Study 3). Collectively, the current findings contribute to theoretical knowledge of person perception by demonstrating that people rapidly and accurately extract social category information from groups of faces, which impacts higher level evaluative judgments. While previous research in social psychology provides deep insight into the processes of perceiving, categorizing, and evaluating individuals (Johnson, Lick, & Carpinella, 2015), the current findings apply these concepts to the visual perception of groups. The process and consequences of group perception remain surprisingly understudied in social vision, especially given the salience of groups for human social life (Cosmides & Tooby, 2005; Neuberg et al., 2011). Indeed, determining whether a group is hostile or hospitable based on its social category membership is likely vital to determining one’s behavioral intentions toward the group. Our findings demonstrate the balance of men to women in a group affects interpersonal judgments in a manner consistent with social stereotypes (i.e., men are perceived as threatening). We also advance research on group perception by identifying one potential mediator, the perceived number of men, which accounts for our effect. Interestingly, while higher sex ratios of men to women increased judgments of men’s coalition, coalition judgments did not significantly mediate the relationship between sex ratio in an ensemble and threat judgments. This result suggests that, at least at rapid presentation times, merely perceiving a higher ratio of men to women increases perceived threat. This actuarial determinant of perceived threat is intriguing, given that men’s coalitional aggression often underlies intergroup violence (Archer, 2004; McDonald et al., 2012), and our observations can inform future work to probe this issue systematically. For instance, coalition-based threats might require greater visual salience, such as greater indicators of group affiliation and membership (Kurzban, Tooby, & Cosmides, 2001) or a common spatial location within an ensemble. The fact that perceivers generalize stereotypes about individuals to judgments about visually perceived groups might appear to be a logical extension of person perception research. What makes our findings especially striking is the remarkable speed with which participants perceived the sex ratio of a group. Although the presentation times in our study were too fast for perceivers to individuate each face in an ensemble, participants nevertheless formed an overall summary percept of the group. Thus, in as little as 75 ms, participants formed impressions of the sex ratio in a group and subsequently tethered their evaluations to that sex ratio. These findings parallel recent evidence from vision science, where researchers found that perceivers form accurate summary percepts of groups of stimuli at remarkably fast presentation times (e.g., 50 ms; Haberman & Whitney, 2007). Thus, much in the same way as perceivers discern the social category memberships and traits of others based on thin slices of behavior (Ambady & Rosenthal, 1992), perceivers are equipped to evaluate groups of faces with remarkable efficiency. The novelty and importance of our work is enhanced by our synthesis of ensemble coding (Alvarez, 2011) with social vision (Johnson et al., 2015), and this approach represents a fruitful avenue for future research to test questions about the inferences and consequences of group or ensemble perception. Of particular interest is how visually perceiving a group leads observers to develop an understanding of group norms and their sense of belonging within those groups (Goodale, Alt, Lick, & Johnson, under review) and how similar processes might operate for groups consisting of different social categories (e.g., race or age) and with a different phenotypic variability within each social category. The current studies pave the way for future research to test these possibilities about the perception and evaluation of groups comprising diverse identities. In summary, we discovered that perceivers rapidly and accurately represent the sex ratio of social groups and use that information to draw inferences and affordances about the threat posed by a group, with the perceived number of men, and not the men’s coalition, mediating this effect. While still nascent, the study of people perception (Phillips et al., 2014) poses numerous questions about the proximal processes and downstream consequences of our perception of social groups.

Acknowledgments The authors wish to thank members of the UCLA Social Communication Lab for their feedback on previous drafts of this manuscript and Ryan Tieu for his assistance with data management.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the National Defense Science and Engineering Graduate Fellowship awarded to the first author and the National Science Foundation Graduate Research Fellowship awarded to Brianna Goodale and David J. Lick.

Supplemental Material

The supplemental material is available in the online version of the article.

Notes 1.

There was a significant sex ratio by participant gender interaction, B = .06, SE = .02, t(66) = 2.85, p = .006, 95% CIs [.02, .10]. We decomposed the interaction by examining simple effects of sex ratio broken down by participant gender. Results showed that women’s numeric estimates of men were more strongly tethered to sex ratio, B = .54, SE = .02, t(49) = 23.62, p < .001, 95% CIs [.50, .59], than were men’s numeric estimates of men, B = .42, SE = .03, t(17) = 13.61, p < .001, 95% CIs [.36, .48]. However, both simple effects were significant and in the same direction, indicating that participants’ numeric estimates of men increased as the ratio of men to women increased. 2.

We also analyzed the data with prime stimuli treated as a categorical variable rather than a continuous variable. These results mirrored those reported here and can be found in Online Supplemental Material. 3.

For this quadratic trend, we found a significant interaction with participant gender, B = .02, SE = .01, z = 2.46, p = .014, 95% CIs [.005, .04]. We decomposed the interaction by exploring the simple quadratic effect separately for men and women. For women, the quadratic trend was significant, B = .03, SE = .009, z = 3.25, p = .001, 95% CIs [.01, .05]; for men, the quadratic trend was not significant, B = −.01, SE = .02, z = −0.81, p = .418, 95% CIs [−.05, .02]. We should note, however, that we have few men (n = 18 after exclusions), thus strong interpretations should be cautiously drawn. 4.

Immediately following the men’s coalition question, we also asked the same question but replaced the word men with women. Since this question was not central to our hypotheses regarding men’s aggression, we report analyses in Online Supplemental Material. Overall results were in the predicted direction (e.g., as the sex ratio of men to women in the ensemble increased, perceived women’s coalition decreased). Multilevel mediation analyses revealed that women’s coalition judgments mediated threat judgments, unlike the finding for men (see Online Supplemental Material).