SR and CM are employed by Animation Research Limited, a commercial animation company. This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials.
Conceived and designed the experiments: JH JCJ JJ. Performed the experiments: JH JCJ. Analyzed the data: JH JCJ JJ DB. Contributed reagents/materials/analysis tools: DB CM SZ. Wrote the paper: JH JCJ JJ DB HW.
Social psychology is fundamentally the study of individuals in groups, yet there remain basic unanswered questions about group formation, structure, and change. We argue that the problem is methodological. Until recently, there was no way to track who was interacting with whom with anything approximating valid resolution and scale. In the current study we describe a new method that applies recent advances in image-based tracking to study incipient group formation and evolution with experimental precision and control. In this method, which we term “in vivo behavioral tracking,” we track individuals’ movements with a high definition video camera mounted atop a large field laboratory. We report results of an initial study that quantifies the composition, structure, and size of the incipient groups. We also apply in-vivo spatial tracking to study participants’ tendency to cooperate as a function of their embeddedness in those crowds. We find that participants form groups of seven on average, are more likely to approach others of similar attractiveness and (to a lesser extent) gender, and that participants’ gender and attractiveness are both associated with their proximity to the spatial center of groups (such that women and attractive individuals are more likely than men and unattractive individuals to end up in the center of their groups). Furthermore, participants’ proximity to others early in the study predicted the effort they exerted in a subsequent cooperative task, suggesting that submergence in a crowd may predict social loafing. We conclude that in vivo behavioral tracking is a uniquely powerful new tool for answering longstanding, fundamental questions about group dynamics.
Social psychology has been described as the study of how individuals behave in groups [
The mechanism of group formation—physical proximity—is at the heart of any investigation into human social behavior, because any convergence of individuals is typically nonrandom and psychologically significant. The classic literature on “personal space” delineates a proximal boundary into which only close others are permitted [
A number of previous field studies have used physical proximity to predict self-identified group membership and shared beliefs and values. For example, Freeman and Webster [
Even simulated models of group formation have used physical proximity as a measure of social grouping. Recently, Gray and colleagues extended previous models of cooperation [
Given the significance of proximity for predicting and understanding how people will affiliate, it is surprising that more is not known about the parameters of incipient groups. At least part of this gap is methodological: there has been no way to keep track of who is approaching whom with anything approximating valid resolution, scale, and control. Because groups in the field do form spontaneously and unpredictably, and involve people whose characteristics and motivations are unknown, observing them, much less controlling them, is a challenge. Previous social psychology studies have often relied either on indirect or approximate indicators of proximity (e.g. rooms in a residence hall), or ecologically deficient laboratory paradigms that operationalize groups in terms of decision-making tendencies and attitudes (see [
However, advances in technology have made it possible to resolve this tradeoff by tracking the movements, proximities, and identities of large groups of people with very high temporal and spatial resolution. In the current study we apply this technology in a new method of “in-vivo spatial tracking” to track sets of 40–50 strangers interacting in a 600m2 space. We use this technique to answer several basic questions about the first stages of social group formation: (1) Do people physically group based on shared physical traits? (2) Do these traits also predict the physical position of individuals in their groups? And (3) What does variability in social distance say about future prosocial behavior?
When groups first emerge from a relatively indistinct whole, what individual characteristics determine what they look like? Although little research has examined the critical moment of physical group formation, a wealth of literature suggests that groups do not assemble randomly, but on the basis of social “fault lines.” According to Social Identity Theory, for example, individuals identify with and favor others who share common features, even arbitrary or superficial ones. For example, “minimal group” research shows a robust preference for others (e.g., in a laboratory resource allocation task) who supposedly share one’s preference for abstract art or one’s nametag color [
A fundamental dimension of incipient groups is their physical dynamic: where individuals position themselves relative to one another. As sets of individuals first calve off from a larger whole, physical position is likely to have a dramatic effect on the nature and effectiveness of the social group that ultimately evolves, since social information is extracted automatically from visual information alone. Taylor and Fiske [
Although no empirical research in social psychology has studied the physical structure of incipient groups, Pennebaker’s theoretical work on “social physics” provides a useful metaphor [
While our first two questions concern small groups that form out of larger wholes, our third question concerns the larger groups themselves and the inferences that can be drawn from their structure. There is no consensus on the “natural” density of crowds, even though there are known—and potentially competing—effects of others’ presence on behavior. According to Zajonc’s Drive Theory, the presence of other people can either make individuals feel aroused and self-aware, or relaxed and anonymous, depending on the size of the group and the transparency of individual performance [
Individuals’ physical “submergence” in a group can be operationalized as the mean distance between them and all other group members, where individuals with the lowest mean distance are the most embedded [
In sum, the current study provides the first direct look at the size, composition, dynamics, and implications of incipient social grouping, the moment at which merely coincidental gatherings of human beings first crystallize into psychologically significant subsets. Using a new method of spatial tracking, we test three hypotheses that have been foreshadowed by previous literature, but never directly tested. First, we predict that physical groups form on the basis of superficial characteristics (in this case, gender and attractiveness). Second, we predict that attractive individuals and women are especially likely to be positioned in the center of their groups. And third, we hypothesize that crowd embeddedness can significantly predict people’s tendency to socially loaf.
One hundred and seventy-two students (41 men, 130 women, 1 “other,” not included in the gender analyses; mean age = 21.43, SD = 4.42; 114 white/ethnic majority, 58 non-white/ethnic minority) volunteered in exchange for NZ$40 to cover their travel expenses. Participants were run in one of four sessions, which took place back-to-back over a single day in the Forsyth-Barr Stadium in Dunedin, New Zealand. The study as a whole was described to participants as “an exploration of the viability of using [the Forsyth-Barr Stadium] for social science research.” Participants were aware that the study might be filmed, but not that their movements would be tracked. No participant indicated any concern about, or awareness of, the mounted camera.
This study was approved by the ethics committee at the University of Otago. All participants provided written informed consent to participate in the study.
Our method draws in part from the growing literature in animal behavior that employs real-time tracking to gather data on animals and insects as they forage [
These issues aside, we also note that both sociometric studies and image-based tracking studies have previously been conducted in observational settings, such as the workplace [
After providing informed consent (including consent to appear anonymously on video), participants were assigned a participant number, which they wore on an orange baseball cap (orange improved the contrast of their heads against the ground, thereby minimizing tracking errors; participants were told that the cap’s purpose was to make their participant number visible). All participants were first photographed by a research assistant before being led, one-by-one in order of their participant number, into view of the video camera—a 20m × 30m space marked off by 1m high crowd control barriers (see
Aerial view of participants during the group formation task, from which proximity and movement data were extracted.
Finally, participants were informed that, prior to the study, the experimenters had scattered 500 1-inch washers randomly around the experimental area, and that the participants would have to work together to pick them up and deposit them in a large basin at one corner of the experimental area. Participants were told that the experiment would end only when all the washers had been deposited in the basin. Participants began this task in the middle of the experimental space, and upon a signal by the experimenter, proceeded to pick up washers and deposit them, one at a time, in the basin. Once all washers were collected, participants were paid and dismissed; they were given a full debriefing by email one week later, including an option (which none took up) to have their data removed from analyses.
Following data collection, we determined the x-y pixel coordinates of each participant’s head, within each video frame, for the entirety of the experiment. The tracking was done automatically using software custom-designed by Animation Research Limited (ARL), a New Zealand-based company specializing in software tracking. The software works by automatically extracting sets of image patches for each participant and finding these patches in the subsequent frames of the video sequence using computer vision techniques, such as template matching [
The location data were subsequently linked to individuals’ demographic data, which they had provided online prior to the study. Physical attractiveness was added to the data file by collecting off-line ratings of the participants’ photographs (taken on the day of the study) from three hypothesis-blind research assistants (two females, one male). Coders’ responses showed acceptable internal consistency, Krippendorff’s α = .75, and were therefore averaged into a single attractiveness score. All analyses reported in this paper controlled where appropriate for effects of two experimental manipulations (reported elsewhere) that were conducted across sessions, but which were not relevant to the questions examined here.
Participants formed a total of 227 groups across a total of 36 trials. Group sizes were normally distributed, with a mean of 6.8 members (SD = 3.0), and a median and mode of 6; groups ranged from 2 to 20 members in size (see
Frequency histogram displaying the distribution of group sizes across the 227 groups formed during the study.
To examine group composition, we used participants’ attractiveness and gender to predict these same traits among the other members of the groups they joined. For example, to the extent that individuals cluster by attractiveness, participants’ own attractiveness should predict the attractiveness of their groups. Thus, gender, attractiveness, and control variables (group size, participant age, and participant minority status) were entered into two repeated measures multilevel models predicting group gender and attractiveness composition respectively, with the 9 experimental trials nested within 172 participants. Group gender and group attractiveness were modeled as level 1 variables since they varied across trials, while participant gender and attractiveness were modeled as level 2 variables. Intercepts were modeled as varying across participants to account for the nested data structure. Parameters were estimated using a restricted maximum likelihood algorithm. In these models, our hypotheses regarding gender and attractiveness were partially supported: there was a non-significant gender-matching effect,
In a second pair of multilevel models, we examined dynamic group composition effects by adding trial number as a level 1 predictor and modeling the cross-level interaction of trial number with participant gender and attractiveness, respectively. That is, in one model, we entered gender, trial number, and their interaction term as predictors of group gender composition (while controlling for participant attractiveness), and in a second model we entered attractiveness, trial number, and their interaction term as predictors of group attractiveness composition (while controlling for participant gender). Both models included group size, participant age, and participant minority status as control variables. In these models, the aforementioned main effects of gender and attractiveness did not change, but were qualified by a significant gender × trial interaction,
To examine the centrality of participants’ position in their groups, we calculated the mean distance between each individual and the geometric center of each group he or she joined. These distances, which were normally distributed with a mean, median, and mode of 1.1m (SD = .23m), were then analyzed in a multilevel model similarly to the composition effects. Attractiveness, age, minority status, and gender were each entered as main effect predictors. As in the composition model, group size was entered as a control variable (group size correlated
To test the relationship between social distance and cooperation, we first calculated the distance that each participant stood from all other participants in their experimental session, averaged across the entire five minute mingling phase of the study. This “mingling distance” (
This study represents a unique approach to studying an unknown phase of the development of social groups, the moment when unrelated individuals make the first overtures of affiliation to others. Although many social scientists, in diverse fields, have conceptualized affiliation as physical distance, none have had the methodological tools to quantify distance under macro-level, yet controlled conditions. Using a ceiling-mounted camera, state-of-the-art color matching software, and a series of original statistical scripts, we were able to calculate unobtrusively participants’ locations in space at high physical and temporal resolution. From these basic data we can compute not only basic descriptive statistics about group formation, duration, and constitution, but also study how these indices are influenced by other behavioral, individual difference, and motivational variables.
The purpose of the study was to explore just a few of these metrics, both to fill gaps in our understanding of group formation, but also to demonstrate the potential of our methodology. First, we found that groups formed around attractiveness and, to a lesser extent, gender: attractive people, and women, were more likely to join groups of other attractive people and women, respectively, although these homophily effects decreased as participants formed subsequent groups. Second, we found that attractiveness and gender were significantly related to participants’ likelihood of standing in the geometric center of these groups, such that women and attractive participants were more likely to be closer their group’s center. It is not clear in the correlational analyses we conducted whether attractive people and women became the seeds of groups of similar others—attractors in Pennebaker’s terminology—or whether they sought out the center of groups after they started to form [
Finally, we were able to predict participants’ cooperation from their embeddedness among the full participant sample as they mingled at the beginning of the study. Participants who were closer on average to other participants at the beginning of the study, were also the ones who were less cooperative at the end of it, that is, who searched the experimental space more slowly in a task that required everyone’s help. The results are consistent with the association between anonymity and social loafing—the first such demonstration in the literature—although further research is required to identify the true mechanism behind this relationship. Specifically, we cannot at the present time be certain of the motivations behind participants’ crowding and foraging styles—or, indeed, whether the two behaviors are causally related. At the same time, the fact that participants’ embeddedness
Our study also generated descriptive data about groups that raise interesting new questions. For instance, why did groups converge around a median of 6? One possibility is that the number relates to the capacity of working memory (i.e., Miller’s “magic number” of 7 +/- 2 [
Of course, the “natural group size” is likely dependent on a number of contextual factors, such as the degree of ambient noise, type of conversation, cultural differences, or group composition. For example, Dunbar and colleagues’ analysis of “freely forming conversational groups” determined a “magic number” of four, which they attributed to constraints on language processing (groups greater than four are too loud and busy for conversations to continue) [
More generally, we propose that the current methodology—constructing a stadium-size laboratory and applying state-of-the-art tracking technology to participants’ social behavior within it—could be used to test major social psychological theories with a unique balance of external validity and control. Indeed, there are hundreds of theories that are well suited for in-vivo spatial tracking; the three questions we examine here are only a sample of the hypotheses that could be tested in a “stadium laboratory.” Readers may, of course, take issue with the ecological validity of our paradigm (we seldom interact with friends and family within empty stadiums). However, by definition,
(XLSX)
(DOCX)
The work on this paper was supported by a Large Grant from the UK’s Economic and Social Research Council (REF RES-060-25-0085) entitled “Ritual, Community, and Conflict” (to JJ and HW) and an award from the John Templeton Foundation entitled “Religion’s Impact on Human Life” (to JJ, HW, and JH). We are grateful to David Barton, Susie Campbell, Laura McCambridge, Cathy Ng, Helen Owen, and Arnika van der Berg for assistance on the day of the experiment. We are also grateful to Leila Morgenroth and Sophie Luther for managing many of the logistical details of the experiment. Finally, we thank Tamlin Connor, Evan Carr, and Hannah Samuelson for assistance with analyses.