B482 – BC – OG Behavior Research Methods 2009, ?? (?), ???-??? doi:10.3758/BRM. Cooperation versus competition in a mass emergency evacuation: A new laboratory simulation and a new theoretical model John Drury University of Sussex, Brighton, England Chris Cocking London Metropolitan University, London, England Steve Reicher St. Andrews University, Fife, Scotland Andy Burton University of Nottingham, Nottingham, England Damian Schofield RMIT University, Melbourne, Victoria, Australia Andrew Hardwick University of Sussex, Brighton, England Danielle Graham St. Andrews University, Fife, Scotland and Paul Langston University of Nottingham, Nottingham, England Virtual reality technology is argued to be suitable to the simulation study of mass evacuation behavior, because of the practical and ethical constraints in researching this field. This article describes three studies in which a new virtual reality paradigm was used, in which participants had to escape from a burning underground rail station. Study 1 was carried out in an immersion laboratory and demonstrated that collective identification in the crowd was enhanced by the (shared) threat embodied in emergency itself. In Study 2, high-identification participants were more helpful and pushed less than did low-identification participants. In Study 3, identifica- tion and group size were experimentally manipulated, and similar results were obtained. These results support a hypothesis according to which (emergent) collective identity motivates solidarity with strangers. It is concluded that the virtual reality technology developed here represents a promising start, although more can be done to embed it in a traditional psychology laboratory setting. Virtual reality (VR) technology has been argued to be helped in the development of important theoretical ad- useful in a number of domains of psychological research. vances. Thus, for example, Mintz’s (1951) classic sim- This article reports three studies applying this technology ulation study used a bottle containing several corks on for the first time to the field of emergency evacuation be- strings held by participants, representing the potential for havior, in order to test a new model of such behavior based exit jamming that would take place if every individual on the social identity approach. tried to evacuate simultaneously. Mintz suggested that Early research on mass evacuation behavior relied ineffectual escape in an evacuating crowd is due to in- on anecdotal evidence and data from the military (e.g., dividual calculation of costs and benefits, rather than to Strauss, 1944). In the 1950s and 1960s, however, the in- a contagious outburst of mass irrationality, as assumed troduction of new laboratory experimental techniques by the early mass panic models (e.g., McDougall, 1920; J. Drury, j.drury@sussex.ac.uk 1 © 2009 The Psychonomic Society, Inc. 2     Drury et al. Ross, 1908; see Chertkoff & Kushigian, 1999, for a re- the past 30 years, where the social identity approach has view of mass panic theories). provided explanations across the range of group-based Subsequent laboratory simulations were more sophis- phenomena (Turner, 1999; Turner & Oakes, 1997). ticated than Mintz’s (1951) design and were marked by Mathematicians and engineers have also filled the gap increased use of digital technology, such as computerized left by experimental social psychologists. In both aca- light displays to represent levels of crowding. On the basis demia and applied settings, increased use has been made of this research, a number of variables have been identi- of computer simulations of crowd behavior to predict and fied as playing a role in the extent of competition versus explain the contours of mass evacuations (e.g., Helbing, cooperation in escape behavior in emergency evacua- Farkas, & Vicsek, 2000). However, it has been argued that, tions. These include group size and perceived threat (Kel- for any sense of realism, such computer models need not ley, Condry, Dahlke, & Hill, 1965), exit time (Chertkoff, only human data (including that from laboratory simu- Kushigian, & McCool, 1996), leadership (Klein, 1976), lations), but also accurate and up-to-date psychological and “rational” contingencies and emotional arousal theory (Langston, Masling, & Asmar, 2006; Still, 2000). (Gross, Kelley, Kruglanski, & Patch, 1972). However, in the last 3 decades, the field has largely A Social Identity Approach to been abandoned by experimental social psychologists. Mass Evacuation Behavior Before describing in detail the VR experimental paradigm Kugihara (2001), an engineer, developed a sophisti- employed in the present studies, we will outline how this cated computer-based experimental simulation of a mass abandonment came about and, hence, explain why a new evacuation situation. Each participant sat in a booth and experimental paradigm is needed. watched a display screen of lights indicating whether oth- ers were aggressive or made concessions toward each Methodological Problems and “Solutions” other as they tried to escape from a painful stimulus. The The key methodological issue confronting the develop- evidence from his studies showed that responses in such ing experimental studies of emergency evacuation is that contexts could be understood in terms of (aggressive or of simulating the impending threat of death. It has been ar- concessive) group norms (i.e., rules for conduct), which gued that, without some kind of strong motivation provided were more salient in larger than smaller groups. These by the experimenter to approximate this in some way, par- group norms were in turn explained in terms of salient ticipants in emergency simulations will not act in the ways social identities. observed in real life (Muir, 1996), so the processes behind As well as the methodological innovation involved, these behaviors cannot be studied. Classic laboratory stud- this research was important for its attempt to apply the ies, such as that by Kelley et al. (1965), induced motivation social identity approach (Tajfel & Turner, 1979) for the through a (bogus) threat in their experimental participants. first time to the field of evacuation behavior. The social Experimenters might claim that the social and applied im- identity approach has become the dominant perspective portance of such studies outweighs any possible suffering in group processes and intergroup relations. But the field on the part of participants. However, ethical guidelines, of evacuations and other mass emergency behavior has from both the American Psychological Association and remained relatively untouched by it. the British Psychological Society, have become tighter The present study followed Kugihara (2001) in apply- since the 1960s. Classic experimental simulations of mass ing the social identity approach to mass emergency evacu- evacuations would, therefore, now risk falling foul of the ation behavior. However, we sought to go beyond his work stipulation that such studies are acceptable only when there both methodologically and in terms of theory. Rather than is no stress-free methodological equivalent. being concerned with the contingencies of social identity In the absence of a laboratory method that was both contents (i.e., the different behavioral outcomes of differ- psychologically engaging and yet ethically sound, the ent identity-based norms), the studies described in this field of mass evacuation behavior has largely been left to article focused on the logic of social identity processes. In sociologists and others, who have used archival data rather particular, we sought to draw upon the principles of self- than experiments (e.g., Aguirre, Wenger, & Vigo, 1998). categorization theory (SCT; Turner, 1982, 1985; Turner, Arguably, therefore, it is outside of experimental social Hogg, Oakes, Reicher, & Wetherell, 1987) to explain the psychology that most of the contributions to the field have fact that cooperation, helping, and other expressions of been made in recent years. In particular, the normative solidarity are common among strangers in mass evacua- approach, which stresses the continuity between everyday tions (e.g., Blake, Galea, Westeng, & Dixon, 2004; Chert- rule-governed behavior and that in emergency situations koff & Kushigian, 1999; Clarke, 2002; Drury, Cocking, & (Johnson, 1987a, 1987b, 1988), and affiliation theory Reicher, in press; Proulx & Fahy, 2003). (Mawson, 2005; Sime, 1983), which emphasizes the im- What is of theoretical interest in such behavior is the portance of preexisting interpersonal ties in limiting panic fact that it carries a cost to the helper—in exit delay and responses, represent the current state of theory in this area even risk to personal safety. The social identity approach (Aguirre, 2005). For mainstream social psychology, there- and SCT in particular suggests a possible explanation. fore, the topic of mass evacuation behavior represents a According to SCT, cognitive representations of the self theoretical lacuna that needs to be addressed. This ab- take the form of self-categorizations, which may range in sence is particularly striking given pertinent theoretical inclusivity from the personal (definitions of what makes advances in the wider field of group processes made in us unique) to the shared, collective, and even univer-  Virtual Reality Study of Mass Evacuation     3 sal (definitions that classify us with others). Collective ­identity—that is, seeing oneself as interchangeable with other ingroup members on some relevant dimension— means seeing these other ingroup members as, in some sense, part of self. This, in turn, means caring about them and acting in their interests, even where these others are not known or even personally liked. SCT principles have been applied successfully to ex- plain emergency-related phenomena such as crowd con- flict (Reicher, 2001) and mutual support among group members (Haslam, O’Brien, Jetten, Vormedal, & Penna, 2005). Moreover, experimental studies of helping be- havior are in line with the hypothesis put forward here that shared ingroup membership enhances cooperation (Levine, Cassidy, Brazier, & Reicher, 2002; Levine, Figure 1. Prosser, Evans, & Reicher, 2005; Levine & Thompson, 2004). Furthermore, a recent comparative interview study of 11 mass emergency events showed that high- mated feature films have short sequences that last a few ­identification survivors saw, received, and gave more seconds that may take a standard PC hours or even days aid to others than did those low in identification (Drury to render (to create the individual frames of the anima- et al., in press). However, these experimental studies of tion). The evacuation environment, on the other hand, is helping were not simulations of emergency evacuations; rendered in real-time as a VR world. It is possible for the and the comparative interview study, by its nature, lacks user to navigate around the environment freely to view the experimental control. Mass panic theories would sug- evacuation from other angles and to interact with virtual gest that the presence of a crowd and the increased risk characters in the simulation. of death or injury incurred by helping others would each In order to achieve the real-time rendering of the world, reduce cooperative behavior and increase competition in a technique called low-polygon modeling was used.2 Al- an emergency evacuation. Hence, there is a need for an though, in reality, objects are made up of a lot of very fine experimental test of the hypothesis that collective identity detail in terms of their shape, it is often possible to reduce in a mass evacuation enhances cooperation and reduces this complexity by ignoring intricate details. Neglecting competition among crowd members. fine detail can potentially reduce the realism of the world, Experimental control (manipulation and control of but this compromise increases the ability for a computer relevant variables) and behavioral measures need to be to process the positions of the hundreds of thousands of combined with a design that avoids falling foul of current polygons that make up the 3-D virtual environment and guidelines on the prevention of distress to research par- render them to the screen. A low-polygon model, textured ticipants. We therefore sought to design a computer-based using digital images (in this case, of an underground rail simulation that engaged participants, yet without threaten- station) that can be cleverly mapped to the polygonal faces ing them with pain. To that end, the Underground Station of the object, can replace the detailed information lost by Evacuation Simulator (USES; Burton & Schofield, 2005) simplifying the mesh geometry. This gives the look of a was developed. Visualization Methodology: The Underground Station Evacuation Simulator The USES (Burton & Schofield, 2005) is an interactive computer simulation that visualizes an emergency evacu- ation at an underground railway station and presents the user with opportunities for cooperating and competing with others. The simulation is modeled using computer game technology (i.e., using similar graphical techniques, physics engines, and user interfaces). The graphical en- vironment within the evacuation simulation was created using 3-D computer-aided design packages similar to those used both in the computer games industry and to create animated models in films such as Shrek or The Incredibles.1 The models are created as a 3-D geometric mesh that is later wrapped in (or textured with) photo- graphic images that give the realism achieved in the final models (see, e.g., Figures 1–3). The simulated evacuation environment is, in fact, more like a 3-D computer game than an animated film. Ani- Figure 2. 4     Drury et al. relevant self-report variables—in particular, manipulation checks and feelings toward others. This VR paradigm avoids use of real or bogus threats of pain to prompt urgency of exit among participants by relying, instead, upon psychological immersion or engagement with the interactive visualization program. As such, the methodology is in part based on role-play techniques where participants are asked to cooperate in imagining a scenario (e.g., Ginsburg, 1979). The present methodology goes beyond this, however, through the use of graphics and computer game technologies that many people are familiar with and enjoy using. VR has been argued to be a useful methodological tool in a number of domains—­primarily, for its ability to enhance ecologi- cal validity while maintaining the control associated with the experimental method (Loomis, Blascovich, & Beall, Figure 3. 1999). More important, it also allows for empirical re- search in domains in which there might otherwise be ethi- cal restrictions. For example, Slater et al. (2006) used an detailed 3-D model, while retaining relatively low compu- immersive virtual environment to successfully replicate tational requirements. the classic study of obedience by Milgram (1963); previ- At the front end, the task facing the simulation user is ously, the perceived trauma caused to participants who to evacuate the station as soon as possible, while facing believed that they had harmed or killed another participant bottlenecks caused by the rest of the crowd. Users are able meant that the Milgram paradigm had been impossible to to push characters aside (any number of times) at the press reproduce. of a key. The amount of pushing is intended as a measure The present study is the first application of VR tech- of competition. Users also have to make decisions about niques, so far as we are aware, to the field of mass evacu- whether to stop and help (or to ignore) four separate ation or any other form of crowd behavior. Again, part people (two male, two female) that they encounter during of the rationale for the application of this methodology the evacuation, who are apparently injured. The number is the fact that, as has been discussed, real-life emergen- helped is intended as a measure of cooperation. Faced with cies involve stress and suffering. As such, it represents a one of the four victims in need, instructions on the screen unique attempt to develop a new method appropriate to invite the user to help or ignore him/her, again at the press the difficult but important theoretical question of explain- of a key. If they choose to ignore the victim, they con- ing cooperation versus competition in a mass evacuation. tinue on their way. If they indicate that they will help, some text appears on the screen thanking them for doing so. To Overview represent the personal costs incurred by helping in a real- Three studies were carried out using the USES to life mass evacuation, instances of stopping to help a fallen (1) explore the scope of this methodological technique character add to the exit time and, hence, the risk of death for investigating aspects of mass evacuation behavior and for the user. Each user’s responses, interactions, and tim- (2) test the usefulness of the social identity approach to ings are automatically recorded and saved in a data file. mass emergency evacuation behavior in relation to other The simulation was designed to be configurable, which theoretical models. allowed a range of experimental conditions to be experi- Study 1 was carried out in an immersion laboratory. enced by different users. A number of key dimensions and Conditions of common (vs. personal) fate were compared system variables can be varied through drop-down menus with a control (1) as antecedents of collective identifica- at the beginning of the program, such as the appearance tion and (2) in terms of cooperation (amount of helping (clothes) of the characters in the evacuation (red, blue, offered, time taken to offer help) and competition (i.e., mixed, or neutral) and the number of other evacuees. To amount of pushing). In Study  2, high and low group enhance the urgency of exit, a “danger-of-death” bar can identifiers were compared on amount of helping offered, be displayed at the top of the screen, the value of which pushing, and concern for others. Possible mediation re- increases over time but reduces as the participant nears lationships between group identification, concern, and the exit and safety. There is also a soundtrack of ambient (1) helping and (2) competition were also assessed, as was sounds from an underground railway. the relationship between the perceived danger of death The USES was designed to be embedded within a tra- and the amount of help offered. In Study 3, both collective ditional social psychology laboratory experiment. Thus, identity and group size were manipulated to determine the participants’ identity can be manipulated by a vignette their independent and interactive effects on both helping at the beginning of each trial (verbally, in a written text, and pushing. Mediation analyses were again carried out, or as a PowerPoint presentation) that casts them as either this time using group liking as a possible mediator. The psychological crowd members or individuals in an aggre- effect of danger of death on both helping and pushing was gate crowd.3 A posttest questionnaire can be used to assess assessed.  Virtual Reality Study of Mass Evacuation     5 Study 1 competition in both of the threat conditions than in the control. In a first study, the VR paradigm was used to explore the possible antecedents, as well as the behavioral conse- Method quences, of collective identification. Design. The design used was a one-way mixed design with three In terms of antecedents, according to SCT, one pos- levels. The independent variable was the mortality salience manipu- sible criterion leading otherwise disparate individuals to lation, with three conditions: control (no mortality salience manip- ulation); personal mortality salience, with participants instructed to define themselves in terms of a collective identification focus on their own death; and collective mortality salience (com- is a common fate: a perceived relationship to an external mon fate), with participants instructed to think of the possibility of force, group, or agent, within which members’ fortunes death that they shared with the others around them. are seen as one (Turner et al., 1987; cf. Campbell, 1958). Participants. Sixty undergraduate students from St. Andrews Put differently, an emergency or disaster can create a sense University took part in this study. Their age ranged from 19 to of “we-ness,” creating new social bonds in a crowd of sur- 25 years. Participation was voluntary. Dependent measures. Doosje, Ellemers, and Spears’s (1995) vivors (Clarke, 2002). It is therefore suggested that mass identification scale was employed, anchored by 1, strongly agree emergency evacuations, like other crowd events (Drury (i.e., high identification), to 7, strongly disagree (i.e., low identi- & Reicher, 2000; Reicher, 2001), might broaden people’s fication). Behavioral measures included number of times (out of level of identification, which, as has been shown above, is four) the victims were helped (vs. ignored), the hesitation time (in a possible basis of cooperation. seconds) taken to help the victim, and the number of times people However, although both field and journalistic evidence were pushed aside. Procedure. After being randomly allocated to one of the three support the idea that disasters can both bring people to- conditions, the participants in the two mortality salience conditions gether psychologically and encourage cooperation (e.g., were instructed to fill in the Mortality Attitudes Personality Survey Chertkoff & Kushigian, 1999; Drury et al., in press; Fritz (Greenberg et al., 1990). Next, they were told to “relax and clear & Williams, 1957; Ripley, 2005), these ideas have not your mind.” They were then instructed to take a few moments to been tested experimentally. imagine the scene in the 1987 Kings Cross underground rail fire of An alternative explanation for enhanced identification 1987, to enhance engagement with the visualization. and helping in an emergency can be derived from ter- Next, the individuals were presented with one of three scenarios, depending on their condition, which the experimenter read aloud to ror management theory (TMT; Greenberg, Solomon, & ensure that they fully understood. For the control, the scenario was Pyszczynski, 1997). TMT suggests that one of the quali- as follows: ties that makes us human is the ability to be conscious of 1. You have just got off the train and are on a platform in Kings our own death and the consequent need to avoid the terror Cross underground station. The station is busy. associated with this knowledge. Faced with the thought of 2. You and the others around you are hurrying to get out of the personal mortality, people create a sense of psychological station because the sales in Oxford Street have just began and endurance that, in turn, provides comfort (Jonas, Schimel, you all want to get some good bargains. Greenberg, & Pyszczynski, 2002; Osborn, Johnson, & 3. You must make your way out of the station as quickly as pos- Fisher, 2006). One way to create such psychological en- sible so you do not miss the best deals. durance is to “cling to the ingroup” (Castano, 2004)—in The personal mortality salience condition differed as follows: other words, to increase endorsement of a positively val- 3. All of a sudden you see smoke and someone shouts, “There’s ued social identity and its associated values (which may, a fire. You must get out quickly!” The fire is large and your life e.g., include altruism). is seriously in danger. You must get out of the station as quickly In terms of meta-theory, whereas SCT is a social, cog- as possible. nitive theory of the group as a collective self, TMT is a Finally, the collective mortality salience condition differed as theory of individual motivation. In relation to behavior in follows: life-threatening emergency evacuations, what is needed is 3. All of a sudden you see smoke and someone shouts, “There’s a test of whether common fate (SCT) or personal threat of a fire. You must get out quickly!” The fire is large and your life death (TMT)—or indeed both of them—enhances identi- is seriously in danger. You must get out of the station as quickly fication, which, in turn, should prompt more rapid helping as possible. While doing this, think of your own death and the and reduce competition. The present experimental para- deaths of the others around you in the station. digm affords precisely a way of doing this. By varying The keyboard facing the participant was connected to three pro- features of the vignette at the beginning of the evacuation, jectors throwing synchronized images of the visualization onto the we can manipulate presence and level of threat indepen- entire front wall of the lab (4 m wide 3 3 m high) and on part of the two side walls. These images filled the participant’s entire visual dently and test their separate effects. field. In addition, there was a surround sound system of speakers In Study 1, participants in the USES were divided into in the room.4 three conditions: control (“in a hurry to get to the shops”), Following the visualization, the participants filled in the self-report personal mortality salience (emphasis on personal death), questionnaire. and collective mortality salience (emphasis on common fate through collective death), with self-report measures Results taken of identification and behavioral measures of co- Identification. Group identification was found to vary operation (helping and time taken to help) and competi- across the conditions [F(2,57) 5 10.41, p , .001, h´ 5 tion (pushing). It was expected that there would be more .27]. Breaking this down, there was found to be signifi- self-reported identification, more cooperation, and less cantly greater identification in the personal mortality con- 6     Drury et al. dition (M 5 3.35, SD 5 0.94) than in the control condition Finally, there is also an issue about the practical feasi- (M 5 5.04, SD 5 1.25) [t(38) 5 4.81, p , .001] and bility of the VR paradigm as implemented in the present significantly greater identification in the collective mor- study. If we can run the study simply using a PC and yet tality condition (M 5 3.66, SD 5 1.48) than in the control demonstrate sufficient psychological engagement among condition [t(38) 5 3.17, p 5 .003]. However, there was participants, the USES can be incorporated into many no significant difference between the collective and the more research studies, since it would not have to rely on personal mortality conditions [t(38) 5 0.79, p 5 .43]. To- full immersion facilities, which are available in only a mi- gether these results showed that threat (whether presented nority of psychology department laboratories. as personal or collective) did indeed create common iden- tification in the emergency itself. Study 2 Behavioral measures. Given that personal and col- lective fate were found to induce common identification Study 2 was intended both to measure the level of en- to an equal degree, for the behavioral measures, planned gagement engendered by the USES methodology (this comparisons were carried out. It was expected that there time run without full immersion facilities) and to compare would be more help, less delay in offering help, and less competing theories of mass evacuation behavior. pushing in the (combined) mortality conditions than in Both mass panic theories and modified versions of the the control condition. In each case, t tests revealed mar- normative account suggest that increased perceptions of ginal effects in line with these expectations [help, t(58) 5 danger reduce helping behavior in an emergency evacu- 1.17, p 5 .13 (one-tailed); time taken to help, t(54) 5 ation (Johnson, Feinberg, & Johnston, 1994; Quarantelli, 21.47, p 5 .07 (one-tailed); and pushing, t(58) 5 21.24, 1957). Study 2 therefore examined the effects of both p 5 .11 (one-tailed); see Table 1 for means and standard danger and identification on helping behavior. We might deviations]. predict a simple main effect whereby helping would de- crease with the increasing rate of danger. However, the Discussion effect of danger on helping might be affected by identi- The key finding from this study is the replication in fication in one of two possible ways. The social identity the laboratory of the field observation that features of an approach would suggest that those in a crowd who define emergency evacuation itself can enhance collective iden- themselves psychologically as part of that crowd would tification (Clarke, 2002; Drury et al., in press). This is in offer more help, even at personal cost. We would therefore line with the prediction based on SCT that common fate expect the absolute amount of cooperation to be greater can lead individuals to redefine themselves in collective for high- than for low-identification participants. On top terms in emergencies, just as in other contexts. The com- of this, collective identification might moderate the ef- parable results obtained for personal mortality salience fect of increased danger on cooperation: As danger in- are consistent with the role of terror management in en- creases, low-identification participants should help less, dorsing social identification in such circumstances too, whereas this relationship might be attenuated for high- however. ­identification participants, who would continue to help to The pattern of behavioral evidence, although in the roughly the same degree irrespective of variations in the directions expected, was weak. However, the trends that perceived danger of death. were obtained in this first exploratory study provide at Any evidence that high identification is associated with least a prima facie case for considering seriously the hy- general cooperation not only counts against the panic pothesized role of collective identity in enhancing coop- model; it also goes beyond the affiliation approach (Maw- eration and reducing competition in a mass emergency son, 2005). The affiliation approach predicts cooperation evacuation. These results therefore offer us some encour- only with affiliates and, indeed, would suggest that, when agement for carrying out further studies using this VR the evacuating crowd consists of strangers in an unfamil- paradigm. iar setting, panic (i.e., competitive behaviors) will occur. On a theoretical level, the next step is to try to improve According to SCT, on the other hand, in the same way upon these results. In addition, adding manipulation that high identification increases cooperation and re- checks and further measures to a future replication would duces competition, it should also enhance the concern give us more confidence that the variables specified by participants express for others, as compared with low- SCT are indeed responsible for the pattern of findings ob- ­identification participants. Moreover, given that concern tained here. for others is the motivation for cooperation and is a func- tion of collective identification, we would expect there to Table 1 be a relationship of mediation between these variables. A Means (and Standard Deviations) for Helping, mediator is a variable that accounts for the relationship Time Taken to Help, and Pushing in Study 1 (in Seconds) between the predictor and the criterion variable (Baron Time Taken & Kenny, 1986). That is, strength of identification should N Helped to Help N Pushed indirectly affect helping through its direct effect on con- M SD M SD M SD cern for others. Strength of identification should predict Control 2.85 1.23 1.52 1.82 3.05 2.63 reduced pushing, as should concern for others, although Personal mortality salience 3.05 1.47 1.18 1.31 1.95 1.50 it is unclear whether the relationship between these three Collective mortality salience 3.40 0.68 0.71 0.99 2.60 2.56 variables should be one of mediation.  Virtual Reality Study of Mass Evacuation     7 Method Results Participants. Seventy-two students, 45 of them female, from Manipulation checks. Three engagement items (α 5 the University of Sussex took part in this study for course credit or .53) were conflated into a composite, the mean score of money. There were no differences on any of the relevant variables for which (7.34) was significantly above the scale midpoint, either of these subgroups. The age of the participants ranged from 18 to 49 years, the mean age being 22.19. indicating that the participants did indeed feel engaged Dependent measures. As in Study 1, behavioral measures in- with the visualization [t(71) 5 6.92, p , .001]. However, cluded number of times (out of four) the victims were helped (vs. judging from the mean scores of the composite measure ignored) and number of times people were pushed aside. Helping was of group identification (α 5 .89), there was no difference measured in relation to danger, indicated as a percentage on the screen in the level of collective identification across the two con- danger bar (see Figure 3), which was recorded in the data file. ditions [low-identification condition, M 5 6.12, SD 5 Self-report measures were taken of concern for others (five items; e.g., “I wanted to help other people in the evacuation”) and 2.58; high-identification condition, M 5 6.17, SD 5 2.59; identification with the crowd (two items; e.g., “I felt a sense of unity t(69) 5 20.24, p 5 .81]. with others evacuating the station”). Manipulation check measures Main effects. No gender differences were found for were also taken of self-reported psychological engagement with the any of the dependent variables, so the analysis excluded scenario depicted in the USES (three items; e.g., “I felt emotionally this variable. In order to test the effects of identification, engaged during the experiment”). For each item, 11-point Likert a median split was carried out on the self-report measure. scales were used (where 1 5 disagree strongly and 11 5 agree strongly). Those at the median (6.5, n 5 6) were eliminated in order Procedure. The participants were recruited to take part in “a Vir- to enhance the difference between those high (n 5 34) and tual Reality study of evacuation behavior.” On being seated at a PC, low (n 5 32) in group identification. they were given headphones to create a sense of immersion in the Danger-of-death values at Helping Opportunities 1 and 2 content of the program. were relatively low (M 5 25% and 51%, respectively), Unlike in Study 1, the only concern in the present study was that whereas at Helping Opportunities 3 and 4, the danger-of- of the behavioral and subjective effects of identification (rather death values were high (M 5 88% and 77%, respectively). than its possible antecedents). Therefore, we attempted to manipu- late high and low identification directly, through inducing the par- These were therefore simplified to make just two helping ticipants to imagine themselves in a psychological or an aggregate opportunities: low and high danger. The mean amount of crowd, respectively. help offered at each of these two points was crossed with The participants allocated to the high-identification condition high and low identification in a 2 3 2 mixed ANOVA to test wore badges and were asked to name three things about themselves for a possible moderation effect. There was a main effect as Sussex University students, whereas the low-identification condi- for danger of death [F(1,57) 5 30.76, p , .001, h´ 5 .39]. tion participants were each asked to name three things about them- selves as individuals. All the participants were then exposed to a Overall, people helped more when danger of death was low brief news video of the Kings Cross fire. In the high-identification (M 5 1.74, SD 5 0.53) than when it was high (M 5 1.08, condition, the participants read and heard the following via Power- SD 5 0.73). High identifiers (M 5 1.59, SD 5 0.18) were Point presentation immediately before the USES program began: also found to help more than low identifiers (M 5 1.32, 1. You and other Sussex University students have just been on a SD 5 0.21) [F(1,57) 5 8.33, p 5 .005, h´ 5 .17]. However, march in London about tuition fees. With a large crowd of other there was no interaction [F(1,57) 5 0.38, p 5 .54, h´ 5 .02]. Sussex students, you are now going home. You all make your Hence, although the high-identification participants offered way into an Underground station. Together, you head towards more help than did the low-identification participants on the platform from which you will catch your train . . . . each occasion, each set of participants were affected in the 2. You are just waiting for your train when there is some com- same way by the level of danger of death; there was no mod- motion along the platform. You suddenly hear someone shout “Fire! Get out, get out!” You look behind you and see smoke eration effect. billowing towards you at one end of the platform. The crowd The behavioral results for identification and help- of people around you looks scared. The air seems to be getting ing were echoed in the analysis of conflated self-report thicker and hotter and you start to choke on the smoke. You measures of concern for others (α 5 .66): Those high in realise that you may only have a few minutes to get back up to collective identification were more concerned for others ground level and survive. (M 5 8.60, SD 5 1.13) than were those low in collective 3. But there are other people trying to get out too . . . The station is still packed with other Sussex students from the march . . . . identification (M 5 7.10, SD 5 1.36) [F(64) 5 23.11, p , .001, h´ 5 .27]. In the low-identification condition, the wording was identical, ex- Before comparing high- and low-identification partici- cept for the opening and third sections: pants on pushing, a box-plot was used to identify and ex- 1. You have just been on a march in London about tuition fees. clude outliers (those pushing over 27 times; n 5 8). (Such You have left the rest of the crowd, and are now going home high scores may have reflected accidental keypushes, on your own. You make your way into an Underground sta- holding the key down continuously, or possible glitches tion. The station is busy with other people—some going home in this early version of the program, since the numbers from work, others are tourists, some have been shopping. You head towards the platform from which you will catch your seemed higher than should be possible in each short trial.5) train . . . . As was expected, those high in collective identification 3. But there are other people trying to get out too. The station (M 5 3.22, SD 5 3.58) pushed significantly fewer times is packed . . . . than those low in identification (M 5 8.22, SD 5 8.03) Following the visualization, the participants filled in the self-report [F(59) 5 11.46, p 5 .001, h´ 5 .17]. questionnaire. The visualization took between 14 and 17 min for Mediation analysis. To test the proposed indirect rela- each participant. tionship between group identification, helping, and con- 8     Drury et al. cern for others, bootstrapping procedures were employed, proach, with the effect of identification on helping being using the macro developed by Preacher and Hayes (2004). wholly mediated by concern for others. This method is argued to be more suitable than the Sobel Concern did not mediate the effect of identification on test for smaller samples (Preacher & Hayes, 2004). It also competition, however, although both identification with makes fewer assumptions than does the more traditional the crowd and concern for others separately reduced the Baron and Kenny (1986) mediation approach and, so, may amount of pushing displayed. Although identification and be more appropriate for data sets that do not necessarily concern for others are clearly related, perhaps the pathway meet all the parametric criteria (MacKinnon, Fairchild, & from identification to (reduced) pushing is not necessar- Fritz, 2007). First, 5,000 bootstrap samples were generated ily the same as that from identification to helping. One from the original data set. If the 95% confidence interval might offer help out of concern for others, but one desists for the estimates of mediation effect does not include zero, from pushing not out of concern, perhaps, but for other it suggests that the mediation effect is significant at the .05 identification-­based reasons—such as politeness, expec- level. The bootstrapping results indicated that the mean tations that ingroup others will themselves be considerate, mediation effects from collective identification through and lack of impatience. Since these possible factors were concern for others to amount of help given (b 5 .09 [CI: not measured in this study, however, these suggestions re- 0.03, 0.18]) was significant at p # .05. As the regression main speculations. equations showed, concern for others fully mediated the Taken together, these results clearly both further con- effect of collective identification on helping behavior. firm the potential of VR methodology for investigating Thus, collective identification predicted the amount of psychological aspects of mass evacuation (even in a set- help given [ β 5 .35, t(69) 5 3.12, p 5 .003]. Concern ting that is not fully immersive) and give some support for others also predicted the amount of help given [ β 5 to a social identity explanation of such behavior. The re- .64, t(68) 5 6.79, p , .001]. Collective identification pre- sults contradict most of the predictions of the mass panic dicted concern for others [ β 5 .46, t(67) 5 4.20, p # model and say something additional to the dominant af- .001]. Finally, when collective identification and concern filiation and normative approaches. Increases in the level for others were put into the equation together, concern for of danger did reduce helping, but these responses were not others remained significant, but collective identification disproportionate (exaggerated, irrational), as mass panic was no longer a predictor of helping [β 5 .07, t(66) 5 approaches would suggest. More important, as we have 0.67, p 5 .51]. seen, irrespective of level of danger, absolute levels of Using the same procedure, the bootstrapping results helping were greater for high- than for low-identification indicated that the mean mediation effects from collective participants. Neither the mass panic model nor the affili- identification through concern for others to amount of ation approach would predict such high levels of helping pushing (b 5 2.25 [CI: 20.69, 0.08]) was not signifi- and concern and reduced levels of pushing in a crowd of cant at p # .05. As would be expected on the basis of strangers, as occurred for the high-identification partici- the social identity approach, high collective identification pants. At best, the normative approach might suggest that negatively predicted pushing [ β 5 20.25, t(69) 5 22.12, everyday norms of courtesy would continue even in an p 5 .04], and concern for others also predicted pushing, emergency; but this approach fails to explain why this again in a negative direction [ β 5 20.34, t(60) 5 22.84, norm conformity would occur more for high- than for p 5 .006]. Yet when the two predictors were put into the low-identification participants. regression equation together, collective identification re- As useful as this study has been as a test run of the visu- mained significant, whereas concern for others was no alization program and an initial test of the social identity longer a predictor [ β 5 20.19, t(64) 5 21.36, p 5 .18]. account of mass evacuation behavior, however, there were a number of procedural issues that need to be addressed. Discussion In particular, the theoretical claims made here are some- The two sorts of manipulation checks taken in this study what weakened by the fact that part of the analysis relied yielded quite different results. Whereas the participants on a post hoc median split on levels of self-reported col- were not engaged as intended in the identity manipulation lective identification. contained in the vignette, they indicated that they did feel We can speculate on why the identification measure did engaged with the VR. not differ between the experimental conditions. One pos- However, there was sufficient variation in levels of iden- sibility is that the items were created simply for this study, tification among the participants to carry out a compari- rather than established measures of identification. Hence, son of low and high identifiers on the dependent variables, they may not have mapped cleanly onto the identification which were found to operate in the directions expected, construct. However, the consistency between the general giving us further reason to regard the USES as a feasible predictions and both the median split comparison analy- tool for this behavioral domain. sis and the mediation analysis suggests that the measures Thus, in line with predictions, those who were found to were not at fault. be high in identification with the rest of the crowd gave More plausible explanations for the apparent failure of more help, expressed more concern, and pushed less than the identification manipulation have to do with features of did those low in identification. The relation of mediation the manipulation itself. The use of badges and the name- between identification, concern, and helping was also three-things technique are tried and tested identity ma- exactly as would be predicted by the social identity ap- nipulations (Haslam, 2004). However, the manipulation  Virtual Reality Study of Mass Evacuation     9 procedure as a whole involved the subsequent provision impact theory, salience makes no difference to the effect of additional information that may have served to blur the of crowd size on reducing helping behavior. Each theory subjective distinction between conditions. would lead us to expect main effects and no interaction. On the one hand, the shared social category used in the However, a third possibility is one of moderation: Crowd vignette—demonstrators—may have little resonance for size might moderate the effect of collective identity, so contemporary U.K. students for whom the campaign over that the effect of collective identity on helping will be sig- tuition fees is far less prominent now than it was a num- nificantly greater in large than in small crowds. ber of years ago. It may not have been relevant enough to As before, helping was expected to be reduced and student participants to evoke identification with a crowd pushing greater when participants did not identify with the of other demonstrators. On the other hand, and perhaps a group. Although the predictions of social impact theory are more plausible explanation for the failure of the identifi- clear for helping, they are less clear for pushing, whether cation manipulation, was that the personal and collective with small or large crowds, since the person pushed is not experimental conditions were not sufficiently distinct. a potential source of influence in the same way as a needy The demonstration (collective identification) backstory victim. The predictive and possible mediating role of liking was held constant, and personal identification was ma- was also explored. Finally, we also sought a further test of nipulated by reference to the participant’s being among the predictions of the mass panic and affiliation models by an aggregate crowd only after being in the demonstration. examining the extent to which danger of death predicted It seems possible, therefore, that this additional personal- not only (lack of ) help to victims, but also spontaneous izing information was not prominent enough for those in pushing. A version of the mass panic model modified by the low-identification condition. the tenets of SCT suggests that increased danger would Through running a third and final version of the study, encourage competitive behavior such as pushing, but only our intent was to manipulate collective identification by for those in the low-identification condition. addressing these possible problems of social category rel- evance and insufficient difference between the vignettes. Method Successful experimental manipulation would serve to Design. This study used a 2 3 2 independent measures design, demonstrate more convincingly and cleanly the claimed with low versus high identification as the first level and number of others visibly present in the simulation (32 for high vs. 8 for low) causal relations between the variables of interest than in as the second level. the present study. Participants. Forty students from the University of Sussex were The third study was also intended to examine the pos- recruited to take part in the study. Their age ranged from 20 to sible mediating role of a further SCT-derived variable 25 years, with the mean age being 22. Seven were male, and the rest different from that employed in Study 2: liking or attrac- were female. All took part without inducements. tion to others, which should be a function of collective Dependent measures. Behavioral measures were the same as identity. Furthermore, the test of the relationship between those in Studies 1 and 2. Additionally, a measure was taken of the danger-of-death percentage at the time of each spontaneous push by perceived danger and behavior was developed by examin- the participant. Two self-report items were used to measure group ing whether not only helping, but also pushing is affected liking: “How positive did you feel towards the other people in the by the level of danger. Finally, by varying the size of the computer simulation?” and “How much did you like the other people crowd in the visualization, this study allowed for a test of in the computer simulation?” There were manipulation checks for SCT not only against panic and affiliation models, but psychological engagement with the VR (three items, based on those also against social impact theory (Latané & Wolf, 1981), in Study 2) and for perceptions of the size of the crowd. Scales were 1–11, where 1 5 agree strongly and 11 5 disagree strongly.6 Since which has previously been applied to predicting (lack of) the scale for collective identification was found to be reliable in help in “emergency” situations. Study 2, just one item from this scale was used: “How much did you feel part of a group during the simulation?” Study 3 Procedure. The participants began the study by reading a news report of the Kings Cross fire. They were then asked to close their Social impact theory (e.g., Latané & Wolf, 1981) sug- eyes and imagine the sights, noises, smells, and other sensations that would be associated with such a scenario. The high-identification gests that group size is a factor determining responses in participants then were given the following scenario: situations in which helping behavior might be involved. Specifically, total social impact is said to be a function 1. You have just been to an England football match at Wembley of strength, immediacy, and total number of influences. Stadium and are now on your way back to Brighton as you have university in the morning. You and the other England support- Hence, according to this theory, the larger the number of in- ers are making your way through the local rail station to the dividuals over which possible influence (e.g., from a fallen Underground, from where you can get the train back home. victim) is diffused, the smaller the influence. Thus, the big- 2. You are just about to board the underground train when you ger the crowd, the less likely a victim is to receive help hear someone shout “There’s a fire, get out, get out!” You look from others in the crowd (e.g., Latané & Nida, 1981). behind you and see large flames at one end of the platform In the present study, therefore, crowd size, as well as with people running away from the fire. Everybody around you looks scared, and you feel yourself starting to sweat and sense collective identity, was manipulated. SCT would suggest your heart pumping faster. The fire seems to be getting bigger that crowd size would make no difference to the amount rapidly and you start to choke on the smoke. You realise that of helping offered when salience of collective identity was you may only have a few minutes to get back up to ground level held constant, whereas from the point of view of social and away from the fire in order to survive. 10     Drury et al. The low-identification condition participants were presented with (M 5 8.61, SD 5 20.51) conditions [t(35) 5 0.56, p 5 .58], the following: although the means were in the expected direction. 1. You have spent a long day shopping in central London and Looking next at the possible effects of crowd size, there are now on the way back to Brighton as you have university in was no difference in the number of victims helped in the the morning. You are making your way through the local rail small and large crowd conditions [F(1,40) 5 0.00, p 5 station to the Underground, from where you can get the train 1.00, h´ 5 .00; M for each 5 .59]. Furthermore, there was back home. no interaction of crowd size and identification [F(1,40) 5 The rest of the scenario was otherwise identical to that in the first 1.27, p 5 .27, h´ 5 .03]. condition. The self-report measures of liking (α  5 .57) were To enhance collective identity, in the high-identification condi- conflated into a single measure and were found to echo tion, the characters in the visualization wore vests that were the same color, whereas in the low-identification condition the characters the behavioral results for helping: Those in the high- wore various vests that were different. ­identification condition (M 5 6.33, SD 5 1.21) liked oth- Half of the high-identification and half of the low-identification ers in the visualization significantly more than did those participants were exposed to a condition in which the crowd was in the low-identification condition (M 5 7.95, SD 5 1.51) small (8 other characters), as opposed to one in which the crowd was [t(38) 5 3.76, p 5 .001]. large (32 other characters). For the analysis of pushing, a box-plot was again used to The rest of the procedure was the same as that in previous trials. Following the visualization, the participants filled in the self-report identify statistical outliers (those over 40 pushes: n 5 3). questionnaire. Low-identification participants pushed more (M 5 18.39, SD 5 12.20) than high-identification participants (M 5 Results 9.26, SD 5 8.54), as was predicted [F(1,37) 5 8.27, p 5 Manipulation checks. The three engagement items .007, h´ 5 .20]. There was more pushing in the large crowd scaled well (α 5 .79) and were therefore combined in a (M 5 17.50, SD 5 11.58) than in the smaller one (M 5 single item. The mean for these was not different from the 10.05, SD 5 10.08) [F(1,37) 5 5.37, p 5 .03, h´ 5 .14]. midpoint [M 5 6.11, SD 5 2.16; t(39) 5 0.32, p 5 .75], There was a very marginal interaction between crowd size suggesting moderate psychological engagement with the and identification [F(1,37) 5 3.16, p 5 .09, h´ 5 .09]; VR. The item measuring the participants’ perceptions of the greatest amount of pushing took place with the larger the difference in size between the crowds was in the in- crowd and when identification was low (M 5 24.89), and tended direction but did not differ significantly between the least amount of pushing took place when identifica- high- (M 5 4.95, SD 5 3.23) and low-density (M 5 5.35, tion was high and the crowd small (M 5 8.40). SD 5 3.23) conditions [t(38) 5 0.36, p 5 .72]. Identifi- Mediation analysis. To test the possible relationships cation differed between the high- (M 5 5.85, SD 5 2.54) between collective identification, helping, and liking for and low-identification (M 5 8.45, SD 5 1.64) conditions others, the bootstrapping procedure was again employed. in the direction intended [t(38) 5 23.85, p 5 .001]. Five thousand bootstrap samples were generated from the Main effects and interactions. Danger-of-death val- original data set. The analysis indicated that liking for oth- ues at Helping Opportunities 1 and 2 were again relatively ers did not mediate the effect of collective identification low (i.e., below the midpoint; 10% and 46%, respectively), on helping (b , 2.001 [CI: 20.03, 0.25]). The regression as compared with Helping Opportunities 3 and 4 (85% and equations indicated that identification did indeed predict 56%, respectively). Hence, to look at any possible mod- the amount of help given [ β 5 20.38, t(39) 5 22.55, eration effects, the data were simplified by conflating Op- p 5 .02]. However, liking of others did not predict helping portunities 1 and 2 and then 3 and 4, respectively, to make [β 5 20.21, t(39) 5 21.33, p 5 .19]. Finally, however, just two helping opportunities: low and high danger. These in line with SCT, identification predicted liking of others two data points were then crossed with the high- and low- [β 5 .57, t(39) 5 4.26, p , .001]. identification conditions in a 2 3 2 mixed ANOVA. With outliers removed, the same tests were carried There was a marginal effect for danger of death [F(1,38) 5 out with pushing as the dependent variable. The boot- 3.23, p 5 .08, h´ 5 .08]. Means revealed that slightly more strap analysis indicated that liking for others did not help was offered when danger was low (M 5 1.30, SD 5 mediate the effect of identification on pushing (b 5 .32 0.69) than when it was high (M 5 1.05, SD 5 0.78). There [CI: 21.04, 1.53]). In the separate regression equations, was a main effect of group identification, with significantly it was found that there was no relationship between col- more help offered in high-identification (M 5 0.70, SD 5 lective identification and pushing [ β 5 .02, t(39) 5 0.12, 0.29) than in low-identification (M 5 0.48, SD 5 0.27) p 5 .91], nor did liking of others predict pushing [ β 5 .07, conditions [F(1,40) 5 6.42, p 5 .02, h´ 5 .15]. However, t(39) 5 0.41, p 5 .68]. there was again no interaction between danger of death and identification [F(1,38) 5 0.13, p 5 .72, h´ , .01]. Discussion For this study, we were able to gather average danger-of- In line with Study 2, it was found that those high in death values for instances of spontaneous pushing, allowing identification gave more help and pushed others less us to analyze the relationship between danger of death, iden- than did those low in identification. Yet, this time, the tification, and pushing. There was no significant difference finding was based on an experimental manipulation of in the mean danger-of-death level for pushing between the identification, rather than on a post hoc median split of high- (M 5 10.21, SD 5 11.71) and the low-identification self-reported identification. Again, identification pre-  Virtual Reality Study of Mass Evacuation     11 dicted helping. But this time, using a different mediat- urgency in participants without deception or the use of ing variable (liking instead of concern), no mediation threats or rewards for escape. Instead, techniques based on relation was found between identification and helping. role-play, combined with the latest computational graph- Although identification and liking were correlated, as ics technology, were employed to engage participants in SCT would suggest (i.e., we like those who we define as the emergency scenario, in a paradigm based on the fa- in our ingroup), there was no relationship between lik- miliar computer game format. Within this framework, all ing and helping. In fact, this too is consistent with SCT, other relevant variables could be controlled, manipulated, which suggests that we feel desire to help the crowd to and measured. Thus, we have been able to make a con- the extent that we identify with it, irrespective of whether tribution to an area of social psychology that, with a few we like individual members. Indeed, this finding also, at notable exceptions (Kugihara, 2001), has been largely ne- least indirectly, counts against the affiliation approach, glected by experimental social psychology and, hence, by where the emphasis is on interpersonal bonds in explain- mainstream social psychological theory. ing helping in an emergency. The method we have developed here has a further Crowd size did not affect amount of help given; hence, obvious advantage over most of the techniques used in these findings give no support to social impact theory. mainstream experimental social psychology, as well as in (The self-report measures of density might suggest that sociological survey research. Rather than relying on tick- our manipulation of this variable was not consciously per- boxes indicating what participants “intend” to do, the key ceived; however, the fact that, as will be discussed below, measures in the present study were behavioral: helping, the participants pushed more in larger groups suggests hesitation, and pushing. The self-report measures we took that there was indeed a difference in the awareness of were additional, not central to the analysis. density across conditions, as was intended.) Neither was The USES has not been produced for commercial there any evidence for a modified version of social impact availability. It was developed by the researchers for their theory, whereby SCT effects operate only in small crowds. own research purposes.7 However, we have done more These results are therefore in line with those of Levine and here than demonstrate the general utility and viability of colleagues (e.g., Levine et al., 2002), who, in a variety of VR technology for the domain of mass evacuation. We (noncrowd) scenarios, have found that collective identifi- have also specified the computational techniques re- cation, rather than diffusion of responsibility, explains the quired. The low-polygon modeling and digital-imaging extent of helping behavior. techniques described here can be employed to re-create The result for greater pushing in the large crowds, any other kind of scenario that researchers wish to use as although significant, may not be as psychologically the backdrop to an emergency evacuation, whether it be ­interesting—or indeed as psychological per  se—as it an office block, theater, or street. The low computational might first appear. People may barge and push more in a demands mean that the resultant program will be able to larger crowd simply because of physical constraints, rather run on most standard PCs and laptops. This is a practical than psychological tendencies toward personal selfish- consideration that experimentalists will find attractive, ness (Chertkoff & Kushigian, 1999; Cornwell, Harmon, since it makes the paradigm amenable to student proj- Mason, Merz, & Lampe, 2001). ects and other research outside of large-budget research As with Study 2, there was a main effect of danger of programs. death on helping behavior: the greater the danger, the The one clear weakness in the use of the new technique less helping—even though, once again, those in the high- lay not in the technology itself, but in our attempt consis- identification condition still gave more help in absolute tently to embed it in a traditional psychology laboratory terms. This finding is therefore consistent with both SCT setting. The framing of the VR with an identity manip- and Johnson et al.’s (1994) modified normative theory of ulation was clearly weak in Study 2. It has been noted emergency evacuation. Moreover, in line with a version previously that identity variables are harder than other of the mass panic model modified by the tenets of SCT, group-relevant perceptions (e.g., collective injustice and there was a marginal interaction suggesting that increased efficacy) to manipulate (Van Zomeren, Postmes, & Spears, danger encourages competition only for those low in 2008). Indeed, although there are a variety of techniques identification. for manipulating salience of real or laboratory identities, some are more robust than others (Haslam, 2004). Clearly, General discussion more consideration needs to be given to choosing the type of social categories that we think (student) participants The studies described here are the first to employ VR might relate to best (e.g., protester vs. football fan) and the technology to investigate experimentally one of the most best techniques with which to make such social categories important, yet difficult to study, topics in psychology: salient. Alternatively, we need to consider employing as mass evacuation behavior. In avoiding both the stress participants people for whom a given social identity is relied upon in many of the classic experimental studies most clearly relevant. For example, did the Scottish stu- and the intrusion and trauma that might be involved in dents in Study 1 find the London Underground setting interviewing survivors from actual emergencies, all three meaningful and engaging (or remote)? studies had no difficulty in getting ethical clearance from Having said this, taken together the effects and correla- the relevant bodies. They were able to induce a degree of tions of identification across the three studies allow us 12     Drury et al. to draw a number of provisional conclusions in terms of ent study, group membership in a high-intensity envi- psychological theory, as well as methodology. First, and ronment was manipulated. Haslam and Reicher’s (2006) most broadly, the results from these three studies suggest BBC prison simulation study took psychometric, physi- the usefulness of collective-level psychological theory ological, behavioral, and observational data to show the in understanding aspects of behavior in mass emergen- dynamic relationship between identification and stress. cies. Whereas in classic group psychology, panic in an With so many objective measures, as with the case of emergency is understood as the dissolution of collectiv- the behavioral measures taken here, it is hard to argue ity (Freud, 1921/1985), our analysis is consistent with the that simulation experiments in general are necessarily view that collective bonds may be strengthened and even the poor relations of studies of “real” group contexts. created through the experience of an emergency. Thus, the Thus, although it is ultimately necessary to supplement first study suggested that social identity could operate as the findings described here with field, archival, and in- a function of the (shared) threat when people experience terview data, rather than dwelling on the weaknesses of an emergency. the methodology, we would seek finally to contextual- Studies 2 and 3 showed that the amount of help of- ize our VR paradigm as a positive contribution to group fered by the participants was affected proportionately by processes research. levels of danger, but also that the participants responded Taken in this context, this study is a success, in that it differently to that level of danger depending on whether reconnects the discipline with some of its more creative their personal or collective identities were salient. Across and productive traditions. As we have seen, laboratory re- all three studies, collective identity was the key predic- search simulating mass emergency situations flourished in tor accounting for variations not only in cooperation of- the 1950s and 1960s. This was the height of the popular- fered, but also the (relative lack of) pushing. Study 2 also ity of the group dynamics approach to social psychology, showed that the effect of collective identity in helping was when group processes were studied behaviorally, in the mediated by concern for others. lab, rather than (as now) largely through questionnaires. Thus, we would argue that the social identity approach This period produced some of the most engaging, pro- to mass emergency evacuation we have tested and found vocative, imaginative, and readable studies found in social support for in the studies reported here takes us be- psychology textbooks today. Our aim has been to develop yond previous advances in the field in certain impor- an engaging experimental study that also meets modern tant ways. Although the normative approach (e.g., John- standards in research ethics. To the extent that we have son, 1987a, 1987b) clearly transcends the mass panic achieved this, contemporary theories of group processes, model by pointing to the predominance of sociality even such as SCT, can inform the field of mass emergency within life-threatening mass emergency crowds, it col- behavior, from which mainstream social psychology has lapses back into the mass panic model when it claims been disconnected for too long. that competitive behavior increases at times of greatest Author Note urgency (Johnson et al., 1994; cf. Quarantelli, 1957). The normative approach follows the affiliation model The research described in this report was made possible by Grant in emphasizing the role of preexisting interpersonal re- RES-000-23-0446 from the Economic and Social Research Council lationships in limiting panic (e.g., Cornwell, 2003; Fein- awarded to J.D. (University of Sussex), S.R. (St. Andrews University), berg & Johnson, 2001; Johnson et al., 1994). But the P.L. (University of Nottingham), and D.S. (RMIT University). Thanks to Luca Pietrantoni for statistical advice. Correspondence concerning this affiliation model too collapses back into the assumptions article should be addressed to J. Drury, School of Psychology, Univer- of the mass panic approach when it suggests that panic sity of Sussex, Falmer, Brighton BN1 9QH, England (e-mail: j.drury@ can occur when there are no familiar figures or features sussex.ac.uk). present (Mawson, 2005). What is missing from all these References models is an explanation for the evidence of positive af- fective, motivational, and behavioral relations to a crowd Aguirre, B. E. (2005). Commentary on “Understanding mass panic and of strangers in an emergency evacuation. By contrast, other collective responses to threat and disaster”: Emergency evacua- SCT suggests that the shift from personal collective tions, panic, and social psychology. Psychiatry, 68, 121-129. identity leads us to define others’ interests in line with Aguirre, B. E., Wenger, D., & Vigo, G. (1998). A test of the emergent norm theory of collective behavior. Sociological Forum, 13, 301-320. our own, instead of understanding them as counterposed. Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator vari- In such contexts, one might even help strangers at a risk able distinction in social psychological research: Conceptual, strate- to the personal self. gic, and statistical considerations. Journal of Personality & Social There are valid questions concerning how well the Psychology, 51, 1173-1182. Blake, S. J., Galea, E. R., Westeng, H., & Dixon, A. J. P. (2004). An processes analyzed here translate into genuine emergen- analysis of human behavior during the WTC disaster of 11 September cies, where the threat of collective death is real and not 2001 based on published survivor accounts. In Proceedings of Third imagined. Questions like this about the validity of simu- International Symposium on Human Behavior in Fire (pp. 181-192). lation studies of course extend beyond the specificities London: InterScience Communications. of this particular domain. In our defense, we can per- Burton, A., & Schofield, D. (2005). Underground station evacuation simulator (Version 3.1). Nottingham: University of Nottingham. haps take some encouragement from the high levels of Campbell, D. T. (1958). Common fate, similarity, and other indices participant engagement and face validity demonstrated of the status of aggregates of persons as social entities. Behavioral in another recent simulation in which, as in the pres- Science, 3, 14-25.  Virtual Reality Study of Mass Evacuation     13 Castano, E. (2004). In case of death, cling to the ingroup. European structure and panic: The impact of social bonds on individual action Journal of Social Psychology, 34, 375-384. in collective flight from the Beverly Hills Supper Club Fire. In R. R. Chertkoff, J. M., & Kushigian, R. H. (1999). Don’t panic: The psy- Dynes & K. J. Tierney (Eds.), Disasters, collective behavior, and social chology of emergency egress and ingress. Westport, CT: Praeger. organization (pp. 168-189). Newark: University of Delaware Press. Chertkoff, J. M., Kushigian, R. H., & McCool, M. A. (1996). Inter- Jonas, E., Schimel, J., Greenberg, J., & Pyszczynski, T. (2002). The dependent exiting: The effects of group size, time limit, and gender Scrooge effect: Evidence that mortality salience increases prosocial of the coordination of exiting. Journal of Environmental Psychology, attitudes and behavior. Personality & Social Psychology Bulletin, 28, 16, 109-121. 1342-1353. Clarke, L. (2002). Panic: Myth or reality? Contexts, 1, 21-26. Kelley, H. H., Condrey, J. C., Dahlke, A. E., & Hill, A. H. (1965). Cornwell, B. (2003). Bonded fatalities: Relational and ecological di- Collective behavior in a simulated panic situation. Journal of Experi- mensions of a fire evacuation. Sociological Quarterly, 44, 617-638. mental Social Psychology, 1, 20-54. Cornwell, B., Harmon, W., Mason, M., Merz, B., & Lampe, M. Klein, A. L. (1976). Changing in leadership appraisal as a function of (2001). Panic or situational constraints? The case of the M/V Estonia. the stress of a simulated panic situation. Journal of Personality & International Journal of Mass Emergencies & Disasters, 19, 5-25. Social Psychology, 34, 1143-1154. Doosje, B., Ellemers, N., & Spears, R. (1995). Perceived intragroup Kugihara, N. (2001). Effects of aggressive behavior and group size on variability as a function of group status and identification. Journal of collective escape in an emergency: A test between a social identity Experimental Social Psychology, 31, 410-436. model and de-individuation theory. British Journal of Social Psychol- Drury, J., Cocking, C., & Reicher, S. (in press). Everyone for ogy, 40, 575-598. themselves? A comparative study of crowd solidarity among emer- Langston, P. A., Masling, R., & Asmar, B. N. (2006). Crowd dynam- gency survivors. British Journal of Social Psychology. doi:10.1348/ ics discrete element multi-circle model. Safety Science, 44, 395-417. 014466608X357893 Latané, B., & Nida, S. (1981). Ten years of research on group size and Drury, J., & Reicher, S. (2000). Collective action and psychological helping. Psychological Bulletin, 89, 308-324. change: The emergence of new social identities. British Journal of Latané, B., & Wolf, S. (1981). The social impact of majorities and Social Psychology, 39, 579-604. minorities. Psychological Review, 88, 438-453. Feinberg, W. E., & Johnson, N. R. (2001). Primary group size and Levine, M., Cassidy, C., Brazier, G., & Reicher, S. (2002). Self- fatality risk in a fire disaster. In Human behavior in fire: Understand- categorization and bystander intervention: Two experimental studies. ing human behavior for better fire safety design (pp. 11-22). London: Journal of Applied Social Psychology, 32, 1452-1463. InterScience Communications. Levine, M., Prosser, A., Evans, D., & Reicher, S. (2005). Identity Freud, S. (1985). Group psychology and the analysis of the ego (J. Stra- and emergency intervention. How social group membership and in- chey, Trans.). In A. Dickson (Ed.), Civilization, society and religion clusiveness of group boundaries shape helping behavior. Personality (pp. 91-178). Harmondsworth, U.K.: Penguin. (Original work pub- & Social Psychology Bulletin, 31, 443-453. lished 1921) Levine, M., & Thompson, K. (2004). Identity, place, and bystander Fritz, C. E., & Williams, H. B. (1957). The human being in disasters: intervention: Social categories and helping after natural disasters. A research perspective. Annals of the American Academy of Political Journal of Social Psychology, 144, 229-246. & Social Science, 309, 42-51. Loomis, J. M., Blascovich, J. J., & Beall, A. C. (1999). Immersive Ginsburg, G. P. (1979). The effective use of role playing in social psy- virtual environment technology as a basic research tool in psychol- chological research. In G. P. Ginsburg (Ed.), Emerging strategies in ogy. Behavior Research Methods, Instruments, & Computers, 31, social psychological research (pp. 117-155). Chichester, U.K.: Wiley. 557-564. Greenberg, J., Pyszczynski, T., Solomon, S., Rosenblatt, A., MacKinnon, D. P., Fairchild, A. J., & Fritz, M. S. (2007). Mediation Veeder, M., Kirkland, S., & Lyon, D. (1990). Evidence for terror analysis. Annual Review of Psychology, 58, 593-614. management theory: II. The effects of mortality salience on reactions Mawson, A. R. (2005). Understanding mass panic and other collective to those who threaten or bolster the cultural worldview. Journal of responses to threat and disaster. Psychiatry, 68, 95-113. Personality & Social Psychology, 58, 308-318. McDougall, W. (1920). The group mind. New York: Putnam. Greenberg, J., Solomon, S., & Pyszczynski, T. (1997). Terror man- Milgram, S. (1963). Behavioral study of obedience. Journal of Abnor- agement theory of self-esteem and cultural worldviews: Empirical mal & Social Psychology, 67, 371-378. assessments and conceptual refinements. In M. P. Zanna (Ed.), Ad- Mintz, A. (1951). Non-adaptive group behavior. Journal of Abnormal vances in experimental social psychology (Vol. 29, pp. 61-139). San & Social Psychology, 46, 150-159. Diego: Academic Press. Muir, H. (1996). Research into the factors influencing survival in air- Gross, D. E., Kelley, H. H., Kruglanski, A. W., & Patch, M. E. craft accidents. Aeronautical Journal, 100, 177-181. (1972). Contingency of consequences and type of incentive in inter- Osborn, C. Y., Johnson, B. T., & Fisher, J. D. (2006). After 9/11 at dependence escape. Journal of Experimental Social Psychology, 8, ground zero: The anxiety-buffering effects of worldview support of 360-377. the first anniversary of 9/11. Basic & Applied Social Psychology, 28, Haslam, S. A. (2004). Psychology in organizations: The social identity 303-310. approach (2nd ed.). London: Sage. Preacher, K. J., & Hayes, A. F. (2004). SPSS and SAS procedures Haslam, S. A., O’Brien, A., Jetten, J., Vormedal, K., & Penna, S. for estimating indirect effects in simple mediation models. Behavior (2005). Taking the strain: Social identity, social support, and the expe- Research Methods, Instruments, & Computers, 36, 717-731. rience of stress. British Journal of Social Psychology, 44, 355-370. Proulx, G., & Fahy, R. F. (2003, October). Evacuation of the World Haslam, S. A., & Reicher, S. D. (2006). Stressing the group: Social Trade Center: What went right? In Proceedings of the CIB-CTBUH identity and the unfolding dynamics of responses to stress. Journal of International Conference on Tall Buildings (pp. 27-34). Ottawa: Na- Applied Psychology, 91, 1037-1052. tional Research Council of Canada. Helbing, D., Farkas, I., & Vicsek, T. (2000). Simulating dynamical Quarantelli, E. (1957). The behavior of panic participants. Sociology features of escape panic. Nature, 407, 487-490. & Social Research, 41, 187-194. Johnson, N. R. (1987a). Panic and the breakdown of social order: Popu- Reicher, S. (2001). The psychology of crowd dynamics. In M. A. Hogg lar myth, social theory, empirical evidence. Sociological Focus, 20, & R. S. Tindale (Eds.), Blackwell handbook of social psychology: 171-183. Group processes (pp. 182-208). Oxford: Blackwell. Johnson, N. R. (1987b). Panic at “The Who Concert Stampede”: An Ripley, A. (2005, May 16). How to get out alive. Time, 165, 48-52. empirical assessment. Social Problems, 34, 362-373. Ross, E. A. (1908). Social psychology: An outline and source book. New Johnson, N. R. (1988). Fire in a crowded theatre: A descriptive inves- York: Macmillan. tigation of the emergence of panic. International Journal of Mass Sime, J. D. (1983). Affiliative behavior during escape to building exits. Emergencies & Disasters, 6, 7-26. Journal of Environmental Psychology, 3, 21-41. Johnson, N. R., Feinberg, W. E., & Johnston, D. M. (1994). Micro- Slater, M., Antley, A., Davison, A., Swapp, D., Guger, C., 14     Drury et al. Barker, C., et al. (2006). A virtual reprise of the Stanley Milgram Notes obedience experiments. PLoS ONE, 1, e39. doi:10.1371/journal .pone.0000039. 1. Shrek, copyright Dreamworks (2001); The Incredibles, copyright Still, K. (2000). Crowd dynamics. Unpublished doctoral dissertation, Pixar (2004). Department of Mathematics, University of Warwick, U.K. Available 2. A polygon is defined mathematically as a closed plane figure from Crowd Dynamics Web site: www.crowddynamics.com/. bounded by three or more line segments. In computer graphics termi- Strauss, A. L. (1944). The literature on panic. Journal of Abnormal & nology, the word polygon has a more specific definition. Here, the term Social Psychology, 39, 317-328. polygon is taken to mean a multisided object that can be filled with Tajfel, H., & Turner, J. C. (1979). An integrative theory of inter-group color or moved around as a single entity. (In most wireframe geometric conflict. In W. G. Austin & S. Worchel (Eds.), The social psychology meshes, computers use triangular polygons.) of intergroup relations (pp. 33-47). Monterey, CA: Brooks/Cole. 3. Here, we follow the social identity model in distinguishing between Turner, J. C. (1982). Towards a cognitive redefinition of the social physical, or aggregate, crowds (numbers of people who are simply physi- group. In H. Tajfel (Ed.), Social identity and intergroup relations cally colocated) and psychological crowds (people who define them- (pp. 15-40). Cambridge: Cambridge University Press. selves as having a common identity) (Drury et al., in press; Reicher, Turner, J. C. (1985). Social categorization and the self-concept: A so- 2001). cial cognitive theory of group behavior. In E. J. Lawler (Ed.), Ad- 4. Because the visualization filled the participant’s view like a real vances in group processes: Theory and research (Vol. 2, pp. 77-122). event, it was thought that the danger bar might draw attention away from Greenwich, CT: JAI Press. the scene and, hence, detract from psychological engagement. It was Turner, J. C. (1999). Some current issues in research on social iden- therefore not employed in this version of the experiment. tity and self-categorization theories. In N. Ellemers, R. Spears, & 5. Since most of the outliers were among the low identifiers, the dif- B.  Doosje (Eds.), Social identity: Context, commitment, content ference in mean pushing scores between high and low identifiers (5.29 (pp. 6-34). Oxford: Blackwell. and 35.97, respectively) prior to their exclusion was much greater before Turner, J. C., Hogg, M. A., Oakes, P. J., Reicher, S. D., & Wetherell, than afterward; the exclusion of outliers therefore made for a tougher test M. S. (1987). Rediscovering the social group: A self-­categorization of the hypothesis. theory. Oxford: Blackwell. 6. Note that self-report scales are in the reverse direction to those in Turner, J. C., & Oakes, P. J. (1997). The socially structured mind. In Study 2. C. McGarty & S. A. Haslam (Eds.), The message of social psychology 7. Other researchers wishing to make use of the USES should contact (pp. 355-373). Oxford: Blackwell. the authors. Van Zomeren, M., Postmes, T., & Spears, R. (2008). Toward an in- tegrative social identity model of collective action: A quantitative research synthesis of three socio-psychological perspectives. Psycho- (Manuscript received December 19, 2008; logical Bulletin, 134, 504-535. revision accepted for publication March 24, 2009.)