Fear in groups: Increasing group size reduces perceptions of danger

Abstract

When we face danger or stress, the presence of others can provide a powerful signal of safety and support. However, despite a large literature on group living benefits in animals, few studies have been conducted on how group size alters subjective emotional responses and threat perception in humans. We conducted five experiments (N=3,652) to investigate whether the presence of others decreases fear in response to threat under a variety of conditions. In studies 1, 2 and 3, we experimentally manipulated group size in hypothetical and real-world situations, and found that fear responses decreased as group size increased. In studies 4 and 5 we again used a combination of hypothetical and real-world decisions to test whether increased anxiety in response to a potential threat would lead participants to choose larger groups for themselves. Participants consistently chose larger groups when threat and anxiety were high. Overall, our findings show that group size provides a salient signal of protection and safety in humans.

Evolutionary biologists have proposed that our primate ancestors first began forming social groups around 52 million years ago (Shultz et al., 2011). Across species, the jump from solitary to more complex social structures have been shown to come with many evolutionary advantages that improve individual fitness, including protection, resource sharing, and access to mates (Alexander, 1974; Krause & Ruxton, 2002; Pollard & Blumstein, 2008; Roberts, 1996). In animals, the size of a group is a powerful predictor of survival. When a threat is present, larger groups dilute risk, decreasing the chance of any one individual being targeted by a predator (Foster & Treherne, 1981; Hamilton, 1971; Lehtonen & Jaatinen, 2016). Groups also allow for joint vigilance, which increases safety and preserves energy for each individual in the group (Elgar, 1989). It has been proposed that, as social animals, humans have evolved to see the presence of others as a strong signal of safety, and that our reliance on others for emotion regulation is an adaptive trait which helps us preserve energy and psychological resources (Beckes & Coan, 2011). The presence of others can have systematic effects on mood, attention and behavior (Park & Hinsz, 2006), while social isolation can increase feelings of vigilance, anxiety and vulnerability (Hawkley & Cacioppo, 2010). Despite this, we know little about how the presence and size of a group affects emotional reactions to danger in humans.

Research on humans and other social animals has shown that the presence of others can have profound effects on reactions to aversive experiences. In what is termed social buffering, individuals facing a threat show reduced behavioral, emotional and biological stress responses during and after the event when there is another individual present, compared to when they are alone (Cohen & Wills, 1985; Hennessy et al., 2009; Kikusui et al., 2006; Kiyokawa & Hennessy, 2018). In humans, the presence of a partner during stress has been shown to reduce negative emotions and perceptions of pain, as well as decreases in physiological and neural stress responses (J. a. Coan et al., 2006; Conner et al., 2012; Eisenberger et al., 2011; Mulej Bratec et al., 2020; Uchino et al., 1996). These effects are observed even when the partner can do nothing to prevent or mitigate the threat, mere presence of a supportive other is enough (Mulej Bratec et al., 2020; Qi et al., 2020). Current research on social buffering in humans suggests that these effects are produced when the presence or perceived support of other people alters one’s perceptions of potentially stressful events. The availability of aid or resources can change an individual’s appraisal of a situation, including their own ability to cope, which in turn affects behavioral and emotional responses to stress (Cohen, 2004). However, most experimental studies on social buffering in humans have only examined the effects in dyads, where support is provided by only one other person (e.g. Coan et al., 2006; Conner et al., 2012; Mulej Bratec et al., 2020).

In animal studies, there is evidence that the presence of additional conspecifics can lead to increased buffering effects (Kiyokawa et al., 2018; Stanton et al., 1985), but this has not been fully investigated in humans. However, there is evidence that the presence of a group of others can impact responses to threat. When given the choice, humans exhibit a strong tendency to affiliate with others. Studies have shown that people are more likely to choose to join a group than to be alone when anticipating painful shocks (Gerard & Rabbie, 1961; Schachter, 1959), or when their mortality is made salient (Wisman & Koole, 2003). Being part of a group can lead to decreases in vigilance (Gomes & Semin, 2020), threat perception (Fessler & Holbrook, 2013) and neuroendocrine stress responses (Häusser et al., 2012), as well as increases in risk taking (Chou & Nordgren, 2017). Observational studies that have directly examined group size show decreases in individual vigilance with increasing group size (Dunbar et al., 2002; Wirtz & Wawra, 2010), as well as changes in affiliative gripping of others in the group (Dezecache et al., 2017). A study on risk perception also showed that participants viewed taking part in a study involving an electric shock as less risky when there were more people involved (Yamaguchi, 1998).

However, this previous research on group responses to threat leaves several open questions. First, these studies do not experimentally control group size in order to examine incremental effects of additional group members or test differences between small and large groups. While it may seem intuitive that increasing group size would increase safety, this may not always be the case. Larger groups could result in increased perception of threat, since they could increase the likelihood of group panic which would hinder escape (Helbing et al., 2000), or could result in a bystander effect where larger groups deter individual intervention (Latane & Nida, 1981). These possibilities make it particularly important to experimentally investigate the effects of group size in these situations. Second, previous studies have not examined group effects on emotional responses to threat. In non-threatening situations, the presence of a group has been shown to decrease negative emotions while maintaining positive ones (Park & Hinsz, 2015). Since negative emotions like fear and anxiety are an important part of human stress responses, and can be impacted by social buffering (Mulej Bratec et al., 2020), it is important to fully understand how they can be influenced by group membership in threatening situations.

This leads to two complementary hypotheses regarding the relationship between group size and emotional reactions to danger. First, based on findings that the presence of others can decrease individual risk and buffer responses to stress by providing a signal of safety, increases in group size should lead to decreases in fear and anxiety in response to threat. Second, since the anticipation of danger can lead to desire for affiliation, individuals showing high fear and anxiety when facing a threat should show an increased preference for joining larger groups. To test these hypotheses, we conducted five studies to investigate how the presence and size of a group influences feelings of fear and anxiety during a threatening event, focusing specifically on the threat of violence or physical danger. We predicted that participants would display reduced negative emotional responses in response to a threat when part of larger, compared to a smaller, groups (Studies 1, 2, and 3); and that they would seek out the presence of others and prefer larger groups when their own level of subjective fear in response to a potential threat was high (Studies 4 and 5).

Study 1

In Study 1, we tested the hypothesis that the mere presence of a group of other people during a threatening situation would decrease fear in response to the threat, compared to being alone. Online participants viewed a series of images depicting an approaching threatening individual. Participants were assigned to one of two conditions, for half the participants the threatening figure was alone, while for the other half there were thirteen other non-threatening individuals (bystanders) present. Participants were asked to imagine how they would feel and react in this situation. This scenario examines the simplest and most straightforward mechanism for how the presence of a group may influence threat perception and subsequent emotional responses. Being in a group dilutes an individual’s risk of being attacked, and group members may be able to provide direct (fighting off the attacker) or indirect (going for help) protection.

Methods

Participants

683 participants were recruited via Amazon’s Mechanical Turk service (MTurk). We set an intended sample size of 300 participants in each of the two conditions. While MTurk has been shown to produce reliable data for psychological experiments, samples tend to contain participants who fail to adequately complete studies and need to be excluded (Thomas & Clifford, 2017), and therefore we recruited more than our intended sample. Of the original participants, 44 were rejected for incomplete data or for completing the survey too quickly (more than 2 SD below mean survey duration). This left 639 (318 females, mean age 35.7+/−12.0) participants in the analysis, 339 in the Alone condition and 300 in the Group condition. All participants were required to be from the United States, and were paid $0.75 for participating. Before beginning the survey, participants provided informed consent in accordance with the Columbia University IRB, which approved all study procedures.

Stimuli

The stimuli were two sets of five black and white images created in Adobe Photoshop showing a threatening figure approaching the viewer. The figure is in an open outdoor space (a parking lot). Each of the five images showed the same environment, with the threat progressively approaching the viewer in each one, as distance has previously been shown to be closely tied to threat perception (Cesario & Navarrete, 2014). In one set of images, the figure is shown alone, while in the other set there are thirteen bystanders scattered across the scene. Six bystanders remain stationary in all five pictures, while the other seven appear to move closer or farther away. The number of bystanders moving closer vs. farther away was balanced across the pictures so that there was always some bystanders close to the viewer and some farther away.

Procedure

Participants first saw an image of the threatening figure alone on a white background, and were asked to rate how scary they thought he was on a scale from 1 (not at all scary) to 7 (extremely scary). Participants then saw all five pictures in their set in a random order. They were asked to imagine that they were actually in the scenario shown (meaning that they were standing where the camera was, with the figure approaching them). They rated how afraid they would feel and how easy it would be to escape on 7-point scales. Next, they were given 5 potential actions, and asked to rank them based on what they would be most likely to do. The five actions were: walk the other way; yell, scream or call for help; prepare to attack; run away; and ignore.

Analysis

We constructed a multilevel model to assess the effects of group presence on fear using the lmer function from the lme4 package (Bates et al., 2013) in R. The package LmerTest (Kuznetsova et al., 2017) was used to calculate p-values. In the model, the categorical variable of condition (alone or group), continuous variable of threat distance (farthest to closest), rater’s sex and rater’s initial rating of the threatening figure were entered into the model as fixed effects. In addition, since we believed there would be individual differences in anxiety and sensitivity to threat between participants, we also controlled for each person’s trait anxiety score on the State-Trait Anxiety Scale (Spielberger et al., 1979). The model included unique intercepts for each subject, as well as well as unique slopes for each participant for the effect of distance.

To analyze the action choice data, we conducted a multinomial logistic regression using the mlogit (Croissant, 2003) package in R. We focused on the first choice action for each scenario. In this model, the probability of choosing any of the defensive actions was compared to the probability of choosing to ignore the threat. The predictors for this model were the main effects and interaction of condition and threat distance, as well as the main effects of fear rating (mean centered) and initial threat rating (mean centered). Results of this analysis can be found in the Supplemental Material.

Results

The results of a multilevel regression showed that while fear ratings increased as the threat came closer (B=.553, se=.035, p<.001), subjects in the Group condition gave lower fear ratings than subjects in the Alone condition for every distance (B=−.459, se=.139, p=.004, Figure 1A). The interaction between condition and threat distance was small and not statistically significant, meaning that the group had the same effect at all distances (B=.060, se=.049, p=.250). The model controlled for individual levels of trait anxiety and ratings of the threatening figure’s scariness, made before seeing the main images. Both of these factors positively predicted fear ratings (trait anxiety: B=.014, se =.004, p<.001; figure scariness: B=.417, se = .028, p<.001.) There was no main effect of participant’s gender, although there was an interaction where female participants gave higher fear ratings than males in the Alone condition, but not in the Group condition. However, the Group condition had lower average scores than the Alone condition for both genders (main effect: B=.024, se =.153, p=.87; interaction: B=.421, se=.209, p=.044). Full model results are shown in Table S1. The presence of the group also influenced what action participants predicted they would take in response to the threat, making them more likely to ignore the threat and less likely to flee; these results can be found in the Supplementary Materials.

Figure 1:

(A) In Study 1, participants in the Alone condition made higher fear ratings than participants in the Group Condition for all threat distances. Lines show model estimates with error bars showing 95% confidence intervals. (B) Participants assigned to larger groups had lower ratings of fear than participants in smaller groups, but only when the group was present (blue). When the same participants viewed images where they were alone with the threat, they reported higher fear than when the group was present, and these values did not vary by group size (red). Lines show model estimates with 95% confidence intervals, points show mean ratings for each condition, with 95% confidence intervals.

Study 2

Study 1 demonstrated that the presence of a group that had the ability to dilute individual risk and provide protection and help led to decreases in fear and changes in defensive behavior. In Study 2, we expanded on this finding by testing the hypothesis that this effect of group presence on threat perception and fear would increase incrementally with the size of the group. A new group of online participants reacted to a similar hypothetical scenario as Study 1, but participants were randomly assigned to sets of images with between 1 and 15 bystanders. Again, participants rated their predicted level of fear for each image.

Method

Participants

2324 participants (1506 Female, mean age 36.68+/−12.54) were paid $0.75 to complete the survey via MTurk. No participants completed the study in less than two standard deviations below the mean time, so data from all participants was analyzed. Before beginning, participants completed an online consent form approved by the Columbia University IRB. When starting the survey, participants were randomly assigned to a group size condition, with an intended goal of approximately 100 per group. The size of the conditions ranged from 88 to 114 individuals (57–68% female).

Stimuli

For this study, we created sets of fourteen images for each of the fifteen conditions. Unlike in Study 1, group presence was a within-subjects variable, with every subject seeing images both the alone and group conditions. In each set, twelve images showed the assigned number of bystanders present (1–15) and two showed no bystanders. Across the 12 bystander images, the position of the threat (near or far) and the positions of the bystanders (near, middle-distance, or close) was counterbalanced twice with different bystanders.

The background image was the same as in Study 1. However, a number of participants reported that they did not find the original figure scary, so we replaced the figure with a man holding a knife, making the threat less ambiguous. A new set bystander images was compiled from Google Images. Feedback from Study 1 also indicated that some bystanders were seen as being unable to offer help (e.g. an old woman), so we chose to only use adult male bystanders. Each figure had been previously evaluated by a group of independent raters (N=134) who rated the figures on several measures related to their perceived ability to protect or keep others safe. The final figures were chosen from the middle of this distribution, excluding figures that were especially high or low on perceived protectiveness. Images were then created using Adobe Photoshop using combinations of 20 bystanders.

Procedure

As in Study 1, participants first saw a picture of the threatening figure on a white background, and rated how scary they thought he was. Next, they saw the fourteen images in a random order. For each image, they answered four questions on 7-point Likert scales. These questions were: “How afraid would you feel?”; “How safe would you feel?”; “How easy would it be to escape?” And “How likely would it be that someone would help you?” They also chose the action they would be most likely to take. The action choices were modified from Study 1: run, yell and ignore were the same, but walk away was removed and attack was changed to “prepare to fight” so that the fight option would be more defensive, rather than aggressive.

Analysis

We again constructed a multilevel model to analyze the effect of group presence and group size on fear, using the same R packages as Study 1. This model contained group size (continuous), condition (alone or group) and threat distance (far or near) as fixed effects, along with their interactions. Again, each subject’s initial rating of the threatening figure was also included, along with trait anxiety score and gender. The model gave each subject their own intercept, as well as their own slope for the effects of distance and condition.

Action choice data was analyzed using a multinomial logistic regression using the mlogit package in R (Croissant, 2003), with Ignore as the reference category. This model contained all the same predictors as the main model, with the addition of fear ratings for each image.

Results

Replicating Study 1, the multilevel regression model (Table S3) showed that participants reported significantly higher fear for images where they were alone compared to when the group was present (B=1.322, se=.159, p<.001). When the group was present, each additional group member produced an additional decrease in fear (B=−.040, se=.006, p<.001, Figure 1B), although this effect was small compared to the contrast between alone and group. The model also showed a significant interaction between group presence and group size (B=.038, se=.005, p<.001) where group size only affected fear ratings for images where the group is present, and has no effect on fear for images where there are no bystanders. This shows that participants in the higher group size conditions were not more likely overall to give lower fear ratings, but that the effect is specific to the presence of the group. As in study 1, fear ratings were higher when the threat was closer (B=1.186, se=.086, p<.001), when initial ratings of the threat were higher (B=.520, se=.014, p<.001), and when trait anxiety was higher (B=.004, se=.001, p=.008). On average, men gave lower fear ratings than women, especially when alone (main effect: B=−.308, se=.042, p<.001; interaction: B=−.114, se=.037, p=.002), but there was no difference between genders in the effect of increasing group size (B=−.011, se=.008, p=.191). Overall, the results of Study 2 indicate that the presence of even one other person has a large effect on threat response, and that this effect increases with additional others.

Study 3

Studies 1 and 2 present evidence that the presence of others can lead to reductions in fear in response to a threat in scenarios where being part of group provides explicit protection. However, these studies used hypothetical situations in which neither the threat nor the bystanders are real. Imagined reactions may be very different from reactions that would take place in response to a real threat (Camerer & Mobbs, 2017). For the next study, we aimed to measure the effect of group size on fear during a real-world experience. Furthermore, while the two previous studies examined risk dilution, evidence from previous research on social buffering indicate that other individuals can still signal safety and increase an individual’s feeling of competence in a threatening situation even if they cannot do anything to directly stop the threat (e.g. Coan et al., 2006; Conner et al., 2012). In this study, rather than a direct physical threat, we induced fear using a horror movie, to test whether the presence of others would decrease fear even when others could do nothing to dilute risk.

To test this, we had participants watch a horror movie either alone or with a group of other people. While horror movies do not present any real danger, they are specifically made to provoke fear reactions in audiences, and have previously been used for emotion inductions in lab settings (Ackerl et al., 2002; Lee & Andrade, 2015; Straube et al., 2010). We had participants watch a movie in our lab, and report their feelings of fear throughout, predicting that fear would be lower for participants watching in larger groups compared to those in smaller groups or alone.

Method

Participants

For this study, 274 participants (152 Female) were recruited from the Columbia psychology department participant pool and received course credit for participation. Participants signed up online and were not aware that the study involved a movie or that they were possibly signing up as part of a group. 29 of these participants (18 female) were excluded from analysis for the following reasons. 16 either declined to participate in the study upon hearing that it involved a scary movie (N=12), or stopped participation before the movie was over (N=4). If a participant chose not to participate before the start of the study, the rest of the group continued (and was counted as a smaller group). If the participant left in the middle, data from the entire group was excluded (8 excluded across 2 groups). Stopping participation either before or during the movie did not appear to be related to group size, as it occurred at roughly the same frequency across all conditions. Five additional participants were excluded because of technical issues, which caused the movie to stop. In all, 245 participants were included in the analysis. The mean age was 21.2 years (SD=4.09) with 110 males and 135 females. Participants completed the study either alone (N=27), or in groups of two (N=32), three (N=33), four (N=28), five (N=35), six (N=30), seven (N=28) or eight (N=32). We set a sample size target of as close to 30 participants in each condition as possible, although three conditions (alone, four and seven) did not reach this threshold before the end of the last semester of data collection. All participants provided written informed consent before the start of the study, and all procedures were approved by the Columbia University IRB.

Stimuli

The movie used for this study was an adapted version of The Conjuring (Wan, 2013). Although the full length of the film is 112 minutes, we edited it down to just over 60 minutes while keeping the main plot intact and understandable, and also maintaining a level of scariness comparable to the original. The final cut of the film was created by the research team using iMovie. Scenes that were not central to the main plot (including sub plots, opening and closing credits, exposition, and repetitive scenes) were removed. Several scenes were rearranged or combined in order to cut length while still keeping the story intact and preserving important details about the characters and plot. In some cases, fade-to-black transitions were added to smooth abrupt jumps where scenes had been removed. The shortened film was then divided into five sections; with six two-minute breaks inserted at the beginning, end and between each section. The sections were roughly equal in length, each between 10 and 11 minutes in length. The breaks consisted of a white screen with black writing, instructing participants to fill out a set of questions.

Experimental Procedures

The movie was shown in a lab conference room. Participants sat on one side of a long table, and the movie was projected onto the opposite wall. Chairs were set up along the table according to the group size, and seating was randomly assigned. Before the start of the movie, participants completed the Trait subscale of the State-Trait Anxiety Inventory, and the Interpersonal Support Evaluation List (ISEL), a questionnaire designed to assess how much participants tend to rely on others for help and support.

In the Alone condition, a single chair was located in the center of the table, and 1–4 pictures of neutral faces were placed above the questionnaires. The pictures used for the Alone condition were four neutral faces taken from the Karolinska Directed Emotional Faces (KDEF) set.

During each of the six rating periods, participants filled out question sets. The questions during the movie asked participants to rate their own level of fear as well as the fear of the other people in the room. They first rated themselves, and then answered the same questions for each of the others. Ratings were made on Likert scales from 1 (not at all) to 7 (extremely). The final set of questions after the end of the movie also included additional questions about horror movie preferences and viewing habits. Participants were instructed not to discuss their ratings with one another. The experimenter checked on participants during each break, but was not present in the room during the movie.

Analysis

Data were analyzed using a multilevel regression models in the R package lme4 (Bates et al., 2013). In the first model, we used the fixed effects of group size, sex, trait anxiety and ISEL score to predict average fear ratings for each participant across the six time points. The model included separate intercepts for each group. In the second model we investigated the effect at each time point. Fixed effects were added for group size (continuous) and the time period the rating was made. Time was analyzed as a factor, with Time 1 as the reference category, because we expected the effect to differ based on the time point of the movie. Interacting group size and time allowed us to examine the effect of group size at each point in the movie. We also accounted for trait anxiety and ISEL scores, as well as participant gender, and the interaction between gender and group size. Random intercepts were added for each individual and each group. Since the results of this analysis reflected the effects at each time point in relation to Time 1, we also conducted post-hoc tests on the slope of the effect of group size using the Phia package in R (Martinez, 2013). We conducted one analysis predicting each participant’s own fear, and a second predicting their ratings of others (see Supplemental Material). The two models were the same except that for the “others” model we added a fixed effect for the participant’s own fear, and added random intercepts for each dyad.

Results

First, we tested whether group size predicted fear, with fear ratings averaged across all time points. Results showed that there was a significant negative effect of group size on average fear (B=−.098, se=.042, p=.021), where participants in larger groups reported lower fear. In this model, trait anxiety positively predicted fear (B=.028, se=.009, p=.003), but sex (B=−.267, se=.169, p=.114) and ISEL score (B=.024, se=.016, p=.125) did not (Table S5).

To further explore this effect, we used a second model to predict fear ratings based on group size and the rating period in which the rating was made, and their interaction, with each time period modeled as a separate factor (Table S6). This produced an estimate of the effect of group size for each of the six rating periods, with Time 1, before the movie’s start, serving as the baseline. The main effects for the rating period factor showed that fear was higher for all ratings made during the movie then it was before the movie’s start (T1 (Intercept): B=1.851, se=.110, p<.001; T2: B=.808, se=.092, p<.001; T3: B=1.429, se=.092, p<.001; T4: B=.579, se=.092, p<.001; T5: B=1.432, se=.092, p<.001; T6: B=1.112, se=.091, p<.001), but that fear levels fluctuated across the course of the movie (Figure 2A).

Figure 2:

Average fear rating fluctuated across the course of movie (A, black points with 95% Cls), with high variability across individual participants (A, blue lines). However, participants watching in large groups reported lower fear than those watching in smaller groups or alone at all time points except at baseline before the start of the movie (B). After correcting for multiple comparisons, slopes during times 3–6 were significantly different from zero, with the effect being most evident during rating periods 3 and 5, when the movie was scariest. In B, points represent means for each group size with 95% Cls. Line shows model estimate for the effect of group size on fear with 95% Cls.

Results for the effect of group size provided evidence that, at all times after Time 1, participants in larger groups experienced less fear than those who were in smaller groups or alone. At Time 1, before the start of the movie, there was no effect of group size (B=.063, se=.049, p=.193). At each subsequent time, there were significant interactions indicating that for these time points, increasing group size predicted a decrease in fear to a greater extent than it did at baseline, (T2: B=−.125, se=.041, p=.002; T3: B=−.181, se=.041, p<.001; T4: B=−.179, se=.041, p<.001; T5: B=−.217, se=.041, p<.001; T6: B=−.180, se=.041, p<.001). We conducted post-hoc tests on the slopes for the effect of group size at each time point (Table S7). This showed that the slopes for Times 2, 3, 4, 5 and 6 were all negative (as the group size increased, fear ratings decreased). After correction for multiple comparisons, the effects were significant for the four time points with the highest average fear (Time 3: B= −.118, X2=5.998, p=.043; Time 4: B=−.116, X2=5.742, p=.049, Time 5: B=−.154, X2=10.145, p=.008; Time 6: B=−.117, X2=5.866, p=.046). Figure 2B shows the effect of group size on fear at each time point.

As in the first model, trait anxiety showed a positive main effect on fear ratings (B=.029, se=.009, p=.002), but there was no effect of sex (B=−.307, se=.160, p=.071) or ISEL score (B=.029, se=.010, p=.118). Overall, this provides evidence that participants in larger groups were less afraid than in smaller groups, and that the magnitude of the effect was strongest when fear was high.

Study 4

In the first three studies, we show that subjective perceptions of threat and fear during scary situations can change based on the presence and size of a group of other people. However, in each of these studies, participants were assigned to a particular group, and had no control over either the size of the group or the identity of the other individuals. However, if being in a group reduces fear, we should also expect that experiencing fear or anxiety in response to a threat may drive people to seek out others and prefer to be in larger groups.

To test whether anxiety in anticipation of a threat would lead to a preference for larger groups, we conducted two studies in which the potential threat was a shark attack while swimming. Since participants would be imagining and anticipating the threat, rather than be faced with it directly, we wanted to make the threat as salient as possible. Therefore, as in Studies 1 and 2, risk dilution and explicit protection provide potential mechanisms for group preference. However, since the shark was presented as only a possibility, the presence of others could further provide a signal of safety and calm.

In Study 4, 400 participants on MTurk were randomly assigned to one of two scenarios that asked them to imaging swimming in a body of water. In the first condition, the scenario described swimming in the ocean, and mentioned the possibility of sharks. In the second, the scenario described swimming in a lake, and no additional danger was mentioned. They then answered questions about their expected levels of anxiety and preferred group size. We predicted that participants who reported high expected anxiety would also prefer to swim in larger groups, especially those in the ocean scenario.

Method

Participants

400 participants were recruited online using MTurk. 200 participants were randomly assigned to the ocean scenario, and the other 200 participants received the lake scenario. Participants were paid $0.50 for completing the task, which took approximately three minutes to complete. Participants in Study 4 did not complete trait anxiety or demographic measures. All 400 participants answered all the survey questions within two standard deviations of the mean duration, so no data was excluded. All procedures were approved by the CalTech IRB.

Experimental Task

During this experiment, participants were first presented with a very brief scenario and then asked to answer a series of questions. Both scenarios asked participants to imagine that they were about to go out into the water a group of other individuals. In the “lake” scenario, participants were asked to imagine that they were about to swim in a lake. In the “ocean” scenario, participants were asked to imagine that they were about to swim in the ocean. The wording for both scenarios was otherwise identical, except for the addition of the following sentence in the ocean scenario: “Since you will be in the open water, there is a possibility of sharks being in the vicinity. It is unlikely that you will be attacked by a shark, but there have been cases of sharks attacking swimmers.” The control scenario (Lake) did not include any mention of sharks or of other possible dangers present

After reading the scenario, participants were presented with a series of questions, which were the same for both groups. The survey questions posed to the participants after they read the scenario were the same for both groups. The key questions of interest regarded their predicted anxiety about going into the water alone (ranked from 1 (Not anxious at all) to 5 (Too anxious to go in the water at all), and the number of people they would prefer to go into the water with (ranked from 1 (Alone) to 7).

Analysis

Survey responses were codified into a numerical scale, and we used linear regression models in R to assess the effects of anxiety and condition on group size. Welch two sample t-tests were used to assess the difference in means between the two groups.

Results

We first tested for group differences between the Ocean and Lake conditions on anxiety. Participants who anticipated swimming in the ocean reported greater anxiety at the thought of going in the water alone (t(392)= 4.14, p<.001. We then constructed a regression model using anxiety, treatment condition and their interaction to predict the number of people each participant wanted in the water with them (Table S10). There was a main effect of anxiety where the more anxious an individual was to go into the water alone, the more people they preferred to be in the water with them (B=.653, se=.109, p<.001). There was no main effect of condition on group size (B=.280, se=.175, p=.111), although there was a significant interaction between condition and anxiety (B=.290, se=.145, p=.047, Figure 3A), indicating that the effect of anxiety on group size is steeper in the ocean condition. Interestingly, we also observed a positive relationship between anxiety and group size in the lake condition. Although anxiety was lower on average than in the ocean condition, some participants still reported high expected anxiety. It may be that, although the lake condition did not include an explicit mention of danger, participants who viewed swimming itself as anxiety inducing also were drawn to joining larger groups.

Figure 3:

In Study 4 (A), online raters with higher anxiety about swimming in a body of water expressed preferences for larger groups. This effect was especially pronounced for online raters in the Ocean condition, where the threat of shark attack was explicit (Blue). In Study 5, (B), anxiety in real beachgoers about swimming in the water alone also predicted a preference for entering the water with more people. Points show jittered individual responses, lines show model estimates with 95% confidence intervals.

The results of Study 4 confirmed our prediction that higher anxiety would result in preferences for larger group sizes. However, as with Study 1 and 2, this was an online study that relied on participants imagined responses to a hypothetical scenario. In Study 5, we sought to replicate this result in a group of real beach-goers.

Study 5

As a follow-up to Study 4, Study 5 (n=44) tested the ecological validity of the previous findings by surveying a diverse group of beach-goers across four different beaches in southern California. Participants were asked if they were interested in participating in an hour-long study on “stress reduction” which would entail getting on a boat docked at the beach, going one mile into the ocean, swimming 20 feet away from the boat and relaxing in the water for 5 minutes before having their stress levels measured. At the end of the explanation of the experiment, participants were told that there was a small risk of sharks being in the vicinity. Immediately after the scenario was explained, participants were asked to answer a series of questions about their preferences similar to those used in Study 4. The key research question was whether participants who reported having higher anxiety about going in the water alone after being primed with the possibility of shark attack would also report wanting to be with a larger number of people in the water.

Method

Participants

44 individuals (22 female, age range 18–54, mean age: 27.3) were recruited from across four different beaches in southern California (Long Beach and Venice Beach in the Greater Los Angeles area, and Mission Beach and Seaport Village in San Diego). The sample was diverse in terms of nationality, with four US states and nine different countries represented. In order to obtain this sample, approximately 120 people were approached and asked to participate. All procedures were approved by the Cal Tech IRB.

Procedure

Experimenters approached beach-goers and asked if they were interested in participating in an hour-long study on “stress reduction” which would entail getting on a boat docked at the beach, going one mile into the ocean, swimming 20 feet away from the boat and relaxing in the water for 5 minutes before having their stress levels measured. At the end of the explanation of the experiment, participants were told that there was a small risk of sharks being in the vicinity – this was intended to prime those who were afraid of sharks to respond to the subsequent questions with their fear response activated. Immediately after the scenario was explained, participants were asked to answer a series of questions about their preferences (very similar to those posed in Study 4, with the key difference being that participants had a more granular scale over which to report anxiety levels i.e. 1–7 as opposed to 1–5). After all questions were completed, participants were told that this had been part of a psychology experiment and were fully debriefed.

Analysis

A visual examination of the data showed that the responses to the group size variable were not normally distributed. Instead, it was bimodal with a large number of responses at the high and low ends of the scale, and fewer in the middle. Therefore, we used a non-parametric measure (spearman correlation) to test the relationship between anxiety and desired group size.

Results

Higher anxiety about being in the water away from the boat was linked to a desire to have more people in the water (rho=.375, p=.012, Figure 3B). There were no differences in either group size (t(42)=−1.643, p=.108) or anxiety (t(42)=−1.073, p=.289) based on gender. While this effect is only correlational, it is consistent with the effects found in Study 4 in a hypothetical version of this scenario, and again confirms the overall relationship between negative emotion and desire to be in a larger group.

Discussion

Our results suggest that the presence of a group has a powerful influence on human threat assessment and emotional response. In the first three studies, we experimentally manipulated the presence and size of groups during threatening situations, and showed that larger groups lead to decreases in fear. Study 1 showed that people are less afraid of a threat when others are present than when they are alone. Study 2 expanded on this finding, showing that even one other person can have a large influence on fear in response to threat, and that each additional person incrementally increases this effect. Study 3 demonstrated that the effect of being in a group could be found in the real world during a high-arousal, fear-inducing situation. Subjects watching a horror movie in a larger group expressed less fear than subjects watching in small groups or alone. In the last two studies, we found evidence that individuals with increased levels of anxiety in response to a potential threat are also more likely to prefer to be in larger groups. Study 4 demonstrated in an online sample that higher anxiety was associated with larger preferred group sizes, especially when danger was explicit, and Study 5 replicated this effect in a real-world situation.

Subjects in these studies were not exposed to any real danger, so it is possible that reactions found in our studies are different from reactions individuals would have when actually in life-threatening situations. However, we used a combination of methods to approach this issue. Studies 1, 2, and 4 used hypothetical scenarios depicting realistic threats. The use of hypothetical scenarios to study human threat responses is well established, and results have shown that the predicted responses to such stimuli mirror behavior seen in animals to real threats (Caroline Blanchard et al., 2001; Harrison et al., 2015). In addition, we demonstrated with Study 3 that being in a group reduced fear in response to the horror movie. Even though this presented no physical threat, the movie produced a high degree of negative emotional arousal and anxiety, and represents the type of real-world experience that people may face on a regular basis. Finally, in Study 5, participants believed that they would be engaging in a potentially dangerous activity (swimming in waters where sharks had been spotted), and their levels of anxiety predicted a desire to be with more people. Taken together, these various approaches provide converging evidence that threat responses are attenuated by the presence of a group.

However, this research relied heavily on the self-reports of participants. Such measurements can lead individuals towards certain responses, or fail to accurately measure individual experiences if participants do not wish to appear afraid in front of other participants or researchers. While this is limitation of the current work, future work can expand on these findings with additional, more objective, measures. One of the key findings from work on social buffering is that the presence of others when facing a threat can attenuate physiological stress responses (Hostinar et al., 2014), and this effect has been demonstrated in group settings (Häusser et al., 2012). Studying whether group size produces incremental changes in physiological responses, as well as psychological ones, would greatly increase our understanding of these effects.

In all of our studies, we showed a relationship between negative emotions and group size. However, just as the animal literature has shown that there are many different benefits to forming groups, it is important to consider that different mechanisms may have contributed to these effects across our studies. In Studies 1, 2, 4 and 5, where the danger is a specific physical threat, the presence of others may dilute individual risk, while also potentially providing help in the event of an attack. However, we also show that the power of groups to attenuate negative emotions goes beyond situations with a specific physical threat where the group dilutes risk. In Study 3, all subjects are exposed to the movie equally, and neither risk dilution nor physical protection is possible. Here the groups’ effects may be a result of emotional reappraisal, where participants observe the reactions of others and adjust their own emotions and behavior as a result (Cohen, 2004; Cohen & Wills, 1985). While these effects may seem separate, they all are paths to the same endpoint: a reduction in stress in response to a threat. Social buffering theory is based on the idea that the presence or support of others changes how we perceive and respond to threat, but the exact mechanisms differ based on situational factors. Others may provide overt aid, or may model more adaptive responses, or may distract from the threat. As long as the support available matches what is required for the situation, buffering and stress reduction can occur (Cohen, 2004). Future studies should examine whether the effect of group size found here varies depending on the specific ways in which others can provide aid.

It is important to consider situations where group size would not decrease fear. While the studies described here focused on specific physical threats and scenarios where the presence of others could reduce feelings of fear and anxiety, there are many situations where being with others could provide no protection, and possibly increase these feelings. In a time when social distancing has become the norm, being with others can increase danger, and neither physical protection nor social safety signaling will help when the threat comes from a disease. In addition there are many cases where other people are the source of danger or stress. For example, a common laboratory stressor is the Trier Social Stress Test (TSST) (Kirschbaum et al., 1993), in which participants give speeches or solve math problems while being evaluated by others. This task reliably elicits psychological and physiological stress responses, and increasing the number of evaluators only increases these effects (Goodman et al., 2017).

However, even in the face of socio-evaluative threat, being part of a group can still have beneficial effects. Studies using a variant (von Dawans et al., 2011) of the TSST in which participants are exposed to stress alongside other participants (who are not evaluators), have shown that being part of a group can positively influence behavior and physiological reactions during and after this kind of stress. For example, participants exposed to a manipulation to boost their feelings of group affiliation before the task showed lower cortisol levels than control participants in groups without the manipulation (Häusser et al., 2012). Also, participants who completed the group version of the TSST were shown to have increased levels of trust and prosocial behavior towards their group-mates after the task (von Dawans et al., 2012). While most studies on social buffering have found the largest effects when the subject knows the individual(s) they are with (Christenfeld et al., 1997; Coan et al., 2006; Edens et al., 1992), these findings provide a possible explanation for why we found these results in groups of strangers, particularly in Study 3. There is evidence that shared emotional experiences encourage social connections and bonding (Cheong et al., 2020), and that being part of a group strengthens feelings of affiliation and cohesion under conditions of stress, even when individuals start out strangers (Morris et al., 1976). It may be that we would see even stronger group effects if individuals knew each other (e.g. if participants in Study 3 had watched the movie with friends), but it is also possible that shared stressful experiences create feelings of affiliation that facilitate social buffering. More research is needed to further explore how, when and why both existing relationships and increased affiliation among strangers may influence group behavior.

It may feel as though these findings are unsurprising. Situations where others can provide help and protection against a threat are generally less dangerous than ones where we face a treat alone. However, these results expand on previous work on group dynamics and social buffering and demonstrate that being part of a group, even a group of strangers, can have a systematic influence on emotional responses to threats. They provide an important stepping stone for future work exploring the nuances and limits of these influences. In times of stress, it is common to seek out help, comfort, and protection from other people. Research has shown that our emotional states are closely linked with our social environments. For example, anxiety makes individuals more likely to seek out advice from others (Gino et al., 2012), sadness increases one’s desire for social connection (Gray et al., 2011), and acute stress can prompt increased pro-social behavior (von Dawans et al., 2012). Even when there is nothing to be done to alleviate the danger, we have a strong preference for being with others when facing situations involving uncertainty and threat (Schachter, 1959). In modern society, many of the most dangerous and stressful jobs are performed as part of a group. Firefighters, police officers and rescue workers all encounter dangerous and high-stress situations and must overcome subjective fear and stress in order to do their jobs and keep themselves and others safe. Further study on group dynamics under conditions of stress and danger is essential for understanding how people perceive and react to such situations.

Data Availability

The data that support the findings of these studies are available from the corresponding author upon reasonable request.