Sociological Forum, Vol. 13, No. 2, 1998 A Test of the Emergent Norm Theory of Collective Behavior B. E. Aguirre,2,3 Dennis Wenger,4 and Gabriela Vigo5 Objective: The paper uses the timing of evacuation behavior of occupants of the World Trade Center at the time of the explosion of February 26, 1993, to ti test predictions from Emergent Norm Theory. Method: It uses ordinary least square multiple regression analysis to examine data from a survey done in the first week in May 1993 of 415 people who worked at the World Trade Center. Results: The theory's predictions regarding the additive effects of size of group and preexisting social relationships on the timing of evacuation are supported. However, the findings document important and unexpected interaction effects of these two variables on the effects of perceived threat, resources, and cooperativeness on the timing of evacuation. Conclusion: The results augment the theory by showing the continued importance of enduring social relationships as determinants of collective behavior. Enduring social relationships are not only useful to differentiate collective behavior from institutionalized behavior but also specify the dynamics attending the occurrence of collective behavior. KEY WORDS: disaster; collective behavior; evacuation timing; emergent norm. INTRODUCTION Emergent Norm Theory (ENT) posits that nontraditional, collective behavior emerges from the crucible of a normative crisis (Turner, 1964; Turner and Killian, 1972/1987). A precipitating event occurs that, depending upon how the event is collectively interpreted by the participants, creates a normative crisis. The crisis destroys, neutralizes, or no longer allows the 'The authors would like to acknowledge the helpful comments of Professor Ralph Turner. Department of Sociology, Texas A&M University, College Station, Texas 77843. 3To whom correspondence should be addressed. 4Hazard Reduction and Recovery Center, Texas A&M University, College Station, Texas 77843. 'Hazard Reduction and Recovery Center, Texas A&M University, College Station, Texas 77843. 301 0884-8971/98/0600-0301$15.00/0 e 1998 Plenum Publishing Corporation 302 Aguirre et al. traditional normative guidelines to be collectively defined as appropriate guides for action. The crisis creates a sense of uncertainty and urgency forcing people to act even as they create meaning through symbolic interaction processes. They interact and create a new, emergent normative structure that guides their behavior. This new normative structure is explained by the theory as a product of the milling and keynodng process that the participants undergo as they attempt to define the situation, propose cues for appropriate action, and try out alternate schemes of social action. According to ENT, collective behavior occurs as people are forced by the crisis to abandon their previously established conceptions regarding legitimate ways of acting. In contrast to other theories of collective behavior such as contagion, ENT assumes the presence of heterogeneous actors with different backgrounds, perceptual abilities, and motives about what is going on, what should be done to respond to the crisis, and who should do it. ENT assumes that collective behavior is not irrational but social, normative behavior (Tierney, 1980). ENT offers a symbolic interactionist explanation of collective behavior. It uses concepts such as the "definition of the situation" reminiscent of W. I. Thomas and H. Blumer's earlier contributions (for a thorough discussion of the symbolic interactionism background of ENT, see McPhail, 1991:61-102). ENT is often used in the sociology of disasters. It emphasizes the transformations of the normative structure of communities impacted by hazards (Gillespie and Perry, 1976). In the most recent version of ENT (turner and Killian, 1987:9-11; see also turner, 1996), the emergent norm is defined not as a precise rule guiding collective action but as an emergent revised definition of the situation. It originates in out-of-the-ordinary "extramundane" social situations in which people may come to feel their emergent collective behavior is feasible, timely, permissible, necessary, or duty-bound behavior. Their collective action is seen as appropriate. For this revised version of ENT, the search for a dominant revised definition of the situation occurs in either of two social contexts, whether preexisting or emergent social relationships. Thus, social relations are used in conjunction with norms to identify collective behavior and differentiate it from conventional behavior. The emergence of social relationships as well as norms characterizes the "pure" form of collective behavior. There are "mixed" forms in which one of the two types of emergence take place, and there is conventional behavior in which neither occur (Weller and Quarantelli, 1973). The present paper tests predictions from this revised version of the theory. It shows the importance of enduring social relations in instances of collective behavior. This test focuses on one of the three types of collective behavior identified by ENT, the solidaristic compact acting crowd. It is a collectivity of Emergent Norm Theory 303 people in close physical proximity and exhibiting a division of labor and a desire to change its external environment (Turner and Killian, 1972/1987:79-95). We lack information on crowd dynamics that would allow us to measure the extent to which the mostly work-related groups of persons evacuating as a result of the massive explosion that rocked New York's World Trade Center on February 26,1973, became solidaristic crowds. Nevertheless, it is plausible to assume that such transformations occurred. The evacuees were in close physical proximity, were socially bounded and limited by preexisting sociocultural arrangements, experienced the effects of the explosion and resulting smoke, and had to adjust to it. The bombing at the World Trade Center (WTC) and the subsequent evacuation of its buildings created compact crowds that enacted nontraditional and emergent collective behavior and provided us the opportunity to test predictions derived from ENT These predictions cannot be tested through the direct observation of norms. Instead, the test depends on the observation of behavior Tierney, 1980) with known temporal attributes. As in other instances of social organization, collective behavior in crisis settings has a normative organization that structures it. We now turn to the predictions derived from ENT regarding precrisis and crisis factors that would have influenced the timing of the initiation of new, nontraditional normative behavior involved in evacuating the WTC buildings. This is followed by the methods used in this research, the extent to which the predictions are supported by the findings, and the implications of the results for the theory. PREDICTIONS FROM THEORY The present test's emphasis on ENT does not preclude the possibility that some of the predictions derived from it can also be derived from other explanations not included in this test. Rather, our approach to theory testing is a multiple testing strategy in which a number of complementary hypotheses are derived from ENT and tested (Brewer and Hunter, 1989:37). The predictions derived from ENT involve both social structure and social interaction. Social Structure Size of Group and Social Relationships The size of collectivities has long being used to differentiate the collective behavior of large collectivities from the dynamics of small group Aguirrc et al. 304 interaction. Given the nature of the milling process assumed by ENT, people in large groups should take longer to organize and mobilize on the basis of their emergent social organization than people in smaller groups. Weller and Quarantelli (1973), while theorizing about the collective behavior of large collectivities, do not specify the direction of the effects of social relationships, whether enduring or emergent, on the timing of collective behavior. It is unclear whether enduring preestablished social relations would delay it. Two opposite effects can be anticipated. It is plausible to argue that people who are detached from others in their groups will join the collective behavior quicker than those who have preestablished social relationships with others in their group. No social tie holds them to their groups, and they can act as free agents. It is also plausible to claim that prior social relationships among participants facilitate communication among them during the milling process (as they interact to find a solution to their collective problem) and may facilitate their early collective behavior. This is the case since according to ENT such communication is essential to normative emergence as well as to people's ability to act concertedly during a crisis. Enduring social relationships facilitate the sharing of a repertoire of experiences and a common vocabulary that enable people to communicate easier in problematic situations and to mobilize. Social Interaction ENT emphasizes the nature of the communication that is occurring within the milling process in large collectivities. Milling is the concept used in ENT to describe the social interaction that takes place among participants in a crisis setting as they attempt to define the situation, propose and adopt new, appropriate norms for behavior, and seek coordinated, collective action to find a solution to shared a problems (Turner and Killian, 1972/1987). From the perspective of the theory, anything that facilitates communication among the participants in a crisis setting can and often does facilitate the emergence of new norms structuring collective behavior. However, the theory is unclear about how emergent or enduring social relationships impact the symbolic interaction processes that are hypothesized to produce the emergent norm. Keeping in mind this absence in ENT, it is still possible to identify four testable propositions: 1. The greater the search for meaning in the milling process focuses upon defining the situation as serious, the quicker should be the mobilization of people and the initiation of collective behavior. Similarly, the greater Emergent Norm Theory 305 the degree of perceived danger, the quicker should be the emergence of norms and the mobilization of the participants. 2. While not explicitly or systematically addressed by ENT, it is plausible to derive from its treatment of the milling process the prediction that resources available to groups responding to crises make for more efficient and effective collective response. However, they also have the effect of increasing the complexity of the collective deliberation of the groups, thus delaying the emergence of crisis norms. The prediction is of a positive association between the extent of resources available to groups and the amount of time it takes people in them to begin acting collectively. 3. The greater the extent to which the search for meaning inherent in the milling process focuses upon proposed cues for emergent action, such as discussions among people as to what to do, the longer it should take them to agree on a course of action such as joining the evacuation. ENT argues that the appearance of unanimity among the participants in incidents of collective behavior accompanies the emergence of dominant norms. According to this theory, once a dominant norm emerges group members disagreeing with it keep quiet out of fear of group censure. Their acquiescence with the dominant norm masks disagreements among them about how to proceed and produce the perception of unanimity typical of crowds. The prediction is that the presence of opinions among group members about social practices reflecting agreement among them—such as discussions as to what needs to be done and helpful behavior among group members—typifies social groups in which a dominant definition as to how to behave has emerged. It should be associated with delayed participation of the members of these groups in collective behavior. 4. ENT does not consider in any systematic way the impact, on the emergent norm, of intergroup interaction in instances of collective behavior. Brown and Goldin (1973:150-163) used E. Goffman's dramatist theory to refer to it as a process of proselitization among groups in a gathering. Such groups compete in their attempt to impose their collective definition of the situation on the gathering. Such proselitization is part of the milling and keynoting that occurs prior to the emergence of the dominant norm. Thus, the prediction is that people's timing of evacuation behavior is impacted by the presence of other groups nearby that may have alternative definitions of when to begin to behave collectively. In the present case, intergroup proselitization takes time and should delay the start of evacuation. The next section presents the methods used to test these predictions derived from ENT 306 Aguirre et al. METHOD The information comes from a survey done in the first week in May 1993 of people who worked at the World Trade Center (WTC) during the explosion of February 26, 1993. It is supplemented by field work and a content analysis of New York Time articles on the incident published during the first month after the event. Field Work In early March 1993, one of the authors went to the WTC and established contact with authorities in it and the City of New York responsible for disaster response. A subsequent site visit, during the second week in April and the first week in May, involved six researchers representing the disciplines of emergency medicine, engineering, and sociology. We interviewed representatives from 43 agencies involved in the incident as well as 23 victims, mostly people who were at or near the garage where the explosion occurred. We used these interviews to prepare some of the questions used in the survey instrument and to help us interpret the statistical findings. Survey The World trade Center is a complex of seven buildings, including six office structures and the Vista Hotel. The vast majority of the approximately 25,000 tenants reside in one of the two 110 floor towers of the WTC. The survey focused on the response of those people who were in the towers at the time of the explosion. The two towers are large, complex social systems. For purposes of sampling, we thought of them as communities of approximately 13,000 residents. Respondents were selected using a random sample of floors stratified by height. Initially nine floors were randomly selected in each tower and 690 questionnaires were distributed to management representatives through the various firms and offices in these floors. These firms representatives distributed and collected the completed questionnaires. Three hundred and sixty-three questionnaires (53 percent) were returned. Two hundred and fifty-four questionnaires were from Tower 2 and 109 from lower 1. Subsequently, in an attempt to increase the size of the sample in Tower 1, additional floors were randomly selected. Eighty-six questionnaires were distributed; 60.4% were returned. Overall, the total sample includes 415 respondents. Two hundred and fifty-four respondents are from Tower 2 and 161 from Tower 1. The response rate in Tower 2 is 307 Emergent Norm Theory 61% and 44% in Tower 1. The overall response rate is 53.4%. Because of the theoretically derived interest in studying collective behavior processes specified in ENT, responses from 35 respondents who were alone at the time of the explosion are dropped from this analysis. We do not have information on the people who opted not to participate in the study. Our assumption is that an unknown proportion were new employees who were not present during the explosion and thus could not answer the questions. Others were on vacation or professional travel and did not have the time available to answer the questions. Still other nonresponses occurred among employees of companies that, unbeknown to us, decided to curtail participation in the study. A consequence is that we do not know the implications, if any, of these nonresponses for the theoretical conclusions made in this paper. Variables The dependent variable, Minutes, (mean = 56, SD = 42) presents the number of minutes after the explosion (12 PM) that it took the respondents to join the evacuation (not to exit the buildings). Sixty-eight percent of the respondents had joined the evacuation during the first hour after the explosion and 91% had done so after two and a half hours. Thirty-three respondents with scores higher than 150 minutes were assigned consecutively higher scores beginning with 151. In this way we preserved their relative rankings while leaving most of the distribution of scores undisturbed. This was done to normalize the distribution of scores in this variable and to solve the presence of six outliers in the ordinary least squares (OLS) multivariate regression analysis with very high untransformed scores in Minutes. The following concepts were used to generate predictors of the timing of the initiation of evacuation: size of the evacuating groups, the scope and extent of the respondents' social relations, perception of the threat, resources, and measures of social interaction during the crisis. Large Group is a dichotomous variable, scored one for collectivities of twenty or more people who collectively moved to a stairway to begin existing the building and zero otherwise. Forty-two percent of the respondents evacuated in large groups. In comparison to smaller groups in which face to face interaction is possible, large groups necessitate different processes of communication, different techniques for coordinating the collective behavior of people in them, and different styles of leadership. Social Relationship (mean = 8, SD = 3) is an ordinal level variable ranging from 0 if the respondents did not know anyone in the groups in 308 Agulrre et al. which they found themselves to 11 if they knew everyone in the groups very well. In order to simplify the presentation of the results, both size of evacuating groups and extent of social relations are combined in one multiplicative term (BIGSOC, mean = 3.3, SD = 4.4, range 0-11). It measures the extent to which respondents were in large groups and knew everyone in those groups very well. The OLS regression analysis includes the multiplicative terms of each of the other predictors (see below) with Bigsoc. This was done to determine how social relationships and size of evacuating groups will affect the patterns of social interaction involved in the emergence of dominant norms, in the present case joining the evacuation of the WTC. This test of ENT uses four measures of the threat as experienced by the respondents. These include whether they reported that they were in danger immediately after the explosion (Danger, 1 = no danger whatsoever, through 11 = life threatening danger, mean = 5.3, SD = 3.2) nothing serious had happened (Serious, 0 = no, 1 = yes, 22%), presence of smoke in their rooms (Smoke, 1 = no smoke, 4 = thick, dense smoke, mean = 2.17, SD = .92), and injuries (Injured, 0 = no, 1 = yes, 15%). Measures of resources used in this test are respondents' contact with formal Rescuers (0 = no, 1 = yes, 65%); knowledge of others on their floors with emergency medical training (EMT, 0 = no, 1 = yes, 22%); and respondents' Familiarity with WTC, measuring whether or not they had previous evacuation experience at the WTC and had worked for more than two years at the WTC. It ranges from one for respondents with no previous evacuation experience and less than three years of employment, to three for respondents with evacuation experience and more than two years of work at the WTC (mean = 2.3, SD = .8). The test of ENT also includes two indicators of social interaction of the respondents during the crisis. The first is Cooperativeness. It measures the extent to which respondents were in groups in which people discussed what needed to be done and were helpful to others as opposed to self centered. It ranges from 0 (no one discussed and all were self-centered) to 5 (everybody discussed and were helpful) (mean 3.2, SD = 1.5). The second taps whether or not respondents received Information and Guidance from friends, office personnel, and others near them in the aftermath of the explosion. It ranges from 0 to 3 for respondents contacted by people in the three categories (mean = 1.4, SD = .98). The number of firms in the floors in which the respondents worked (Firms, 0 = one, 1 = two or more, 43%) is included in the OLS regression models as a proxy of the presence of multiple work groups. It is used to test intergroup proselitizing's hypothesized effect of delaying the initiation of evacuation. 309 Emergent Norm Theory Other predictors are included in this test as statistical controls. They may be associated with the probability of initiating evacuation and the social relationships and patterns of social interaction of the respondents: the population of employees of the floors where respondents worked (Size, mean = 114, SD = 49); the floor number, indicating its location in the buildings (Floor, mean = 51, SD = 26); and respondents' Age in years (mean = 37, SD = 9) and gender (Female, 0 = male, 1 = female (56%). Information available in the survey and not otherwise used in the statistical analysis allowed the imputation of missing values (Anderson et al, 1983) in six variables: Social Relationships = 11 scores assigned; Large group = 6; Helpful = 16; Previous = 8; Look for Others = 6; Serious = 17. We also used pairwise deletion of cases in order to preserve sample size. Results are mostly invariant when listwise deletion of cases is used. Statistics OLS multiple regression is used to model the timing of the initiation of the evacuation (the results are replicated when logistic regression is used; not shown, available upon request). The assumptions of OLS regression are met. Showing the absence of significant multicollinearity, all tolerance values of the predictors are above .77, except Floor (.66). Plots of standardized residuals do not show marked deviations from normality. There are no multivariate outliers in the OLS analyses. Moreover, its DurbinWatson statistic (1.51), a test for sequential correlation of adjacent error terms, is statistically insignificant. We do not attempt to exclude variables from the models on the basis of their degree of statistical significance. Our intent is to use them to test theoretically derived predictions. Thus, we do not report standardized regression coefficients. The effort is not to show the relative importance of the predictors. Instead, we report unstandardized partial regression coefficients. The unstandardized regression coefficients have the advantages of allowing ease of comparison of the same predictors between the two OLS models we present, a particularly important task since we wish to stress the impact of interaction terms (with Bigsoc) on the effects of each of the predictors. They also show the number of minutes before initiation of evacuation (Minutes) per change in original units of each of the predictors after controlling for the other predictors in the equations. Finally, by preserving the original scales of the predictors, we allow the calculation of the impact on the dependent variable of combinations of values in the predictors which we do not perceive as relevant at present. 310 Aguirre et al. The following pages present a description of the terrorist event that precipitated the unscheduled and unplanned evacuation of the WTC and the results of the tests of the predictions. RESULTS Description of Event Evacuation from the WTC buildings, including the two 110 story towers, began immediately after a van packed with high explosives exploded on the B-2 level of the underground parking garage at approximately noon time. The blast created a crater that was approximately 170 feet wide, 180 feet long, and seven stories deep. The explosion killed six people and injured an unknown number of others. It ignited fires that produced massive amounts of thick, black smoke. The smoke entered the ventilation system, stairways, and elevator shaft of the building, and within 5 minutes, people who were over 40 stories above the garage began experiencing smoke in their work areas. The explosion destroyed both the primary and backup electrical systems for the building. As a result, the public address system did not function, there were no official evacuation orders issued by recognized emergency and building officials, and tenants could not ascertain with ease what had occurred. Because of the lack of electricity, all 250 elevators in the complex stalled. More than 25,000 people in the buildings were forced into only three stairwells and us.ed them as escape routes. These stairways are very narrow and people had to virtually stand toe-to-toe during evacuation. Although successful, the evacuation was hindered by the lack of lighting, very limited means of communication among authorities and among authorities and tenants, thick smoke, the great height that many respondents had to traverse, and limited number of exit routes. Despite these difficulties, panic flight did not take place, and more than half of the evacuees had left the buildings by 3:00 PM. The majority of the respondents joined the evacuation during the first hour after the explosion. Nevertheless, a sizable minority chose to remain in the office for varying lengths of time before they evacuated. Within our sample of respondents who were in their office at the time of the explosion, 30% reported that they initially decided to remain in their offices. More than half of the respondents were female. The majority were contacted by official rescuers. Their average age was 37. The average respondent had worked in the Word Trade Center for five and a half year, worked in one-firm floors with average populations of 114 workers, and in floors high above the ground (Floor, mean = 52). Thirty-six percent had experienced Emergent Norm Theory 311 previously an emergency evacuation at the World Trade Center. A minority of the respondents (15% sustained injuries sometime during the incident. The 380 respondents were in the presence of others at the time. Many were in sizable groups of people. Indeed, almost half of the respondents evacuated in Large Groups of 20 people or more, with whom they interacted as they moved together to the exits. The majority knew the people around them rather well (Social Relationship, median = 9). The majority of people near the respondents discussed what needed to be done (Discuss, median = 4). A great deal of milling occurred. Twentysix percent of the respondents sought information and advice from other people in the area, 25.5% tried to phone for help and/or information, and 11.3% turned to the media for information. Only 8.7% stated that they immediately left the building without engaging in any confirmation behavior. Most respondents (89%) perceived people near them as helpful rather than self-centered. Respondents described the behavior of others near them as rational as opposed to irrational (96%), calm as opposed to excited (60%), cooperative as opposed to competitive (98%), and orderly as opposed to panicked (86%). A small percentage (12) of the respondents received instructions from friends (Friend). Fifty-six percent did so from office personnel. Most respondents (73%) Looked for Others, although unfortunately we do not have information on the identity of those they searched for. The majority (76%) thought that something serious had occurred (Serious). Indeed, respondents report that very few people continued working as if nothing had happened (mean of 1.6 in a scale of 1-5). The explosion created a rather anomic condition for the tenants. Most respondents knew that something unusual had happened. They felt the explosion or were alerted by a loss of power and loss of their computers and phone services. However, it simply was not clearly evident to them what had occurred. As a result, different definitions of the situation were discussed. For example, while 33.7% claimed that they immediately knew that there had been an explosion, a larger number (40.5%) stated that they did not know what had occurred. The majority of the respondents perceived moderate levels of Danger (median = 5). Analyses indicate that the level of perceived danger was higher for women than men and increased over the course of their presence in the buildings in the aftermath of the explosion. Respondents in Tower 1 perceived higher levels of danger than did those in Tower 2. This is also true of respondents in the lower floors, closer to the site of the explosion. 312 Aguirre et al. Analysis of the Timing of Initiation of Evacuation Behavior Two models (see Table I) present the results of the OLS multivariate regression analyses. The first is an additive effects model that includes the effects of statistical controls and theoretically specified predictors. The second model also includes the interaction terms of all the predictors with Bigsoc. It tests the theoretically specified interaction effect of the size of evacuating group and extent of social relationship on the other predictors of the timing of evacuation. Results show that it explains more of the variance of the timing of their evacuation. As shown in Table I, the overall fits of the additive and interactive models as measured by Rz (the percent of the variance in Minutes explained by the predictors in the models) are .19 and .30 (p < .000), respectively. On the average, as shown by the standard error of estimate in the additive model, predicted Minutes scores will deviate from the actual scores by 38.8 minutes. A. The result of this test supports the predicted effect of the size of evacuating groups on lengthening the time of the initiation of evacuation behavior. Respondents in large groups took 6.7 minutes longer (P = .12) to initiate their evacuation. The results also support and clarify the prediction regarding the effects of Social Relationship. As shown in Table I, the more people respondents knew in their evacuating groups and the better they knew them the longer it took them to initiate their evacuation (b = 1.24 minutes, p < .10). The results of this test support Weller and Quarantelli's (1973) claim regarding the importance of the effect of preexisting social relationships on collective behavior. In the present case, such enduring social relationships delayed the start of evacuation behavior. The web of social relationships of respondents and members of their groups worked against the espousal and adoption of norms supporting individual, competitive flight behavior, and for the adoption of cooperative behavior that delayed their exiting the buildings. People who know others well tend to concern themselves with their fates. Crowds of known people inhibit individualistic solutions in favor of a shared norm. These results support Mawson's (1980) affiliative model of escape behavior rather than competitive panic models of escape (see Sime, 1983, for an empirical comparison of these two models; contrast it to ENT's claim that emergent norms occur in solidaristic as well as individualistic crowds in which people do not help each other). Bigsoc represents the effect of these two variables on the timing of evacuation. This multiplicative interaction term is not statistically significant when included in the additive model (not shown). However, it is highly 313 Emergent Norm Theory Table I. Timing of the Initiation of Evacuation: Additive and Interactive Modelsa Predictors Large Group Social Relationship Additive 6.T (4.4)C 1.24 (.71) Bigsoc Danger -.99 (.75) -.76 (5.46) I Smoke I Injured -3.87d (2-4) 10.0* (5.0) 11.6fc 16.2* (5.4) I Familiarity (5.4) -2.6* (.89) .63* (.18) -8.0 (6.8) 3.1* (1.2)C -4.8 (2.8) .88 (.60) -18.3* I BMT 1.03 (.88) -13.96* (5.75) I Rescuers 14.5 (11.5) -17.2* 1' Serious Interactive (6.8) .58 (1.4) (6.1) .60 (1-17) 27.8* (6.9) -3.52* (1.26) .20 (2.78) I 3.76 (3.4)C -1.14 (.69) Cooperativeness I Information and Guidance 1.35 (1.4) .65 (2.2) I Firms 12.4* (4.6) I Population I -.12* (.04) 3.95* (1/7) * -.96* (.32) -.70 (2.6) .49 (.52) 1.33 (5.8) 3.17* (1.2) -.18 (.05) .02* (.01) Aguirre et al. 314 Table I. Continued Predictors Floor Number Additive .32 (.10) .10 (.13) .05C -(.06) .24 -.86* (.37) .18* (.07) -18.4* (5.5) 4.06 (1.06) I Age (.03) I Female -5.74 (4.48) I Constant R2 Standard error F P Interactive 6 39& .19 38.6 4.75 < .0000 97.4fc .30 36.8 4.25 < .0000 "Unstandardized regression coefficients rounded to the nearest hundreth. Standard errors of the is in parentheses. b p < .05. c p < .10. *p < .12 "I: interaction with Bigsoc. statistically and substantively significant in the interactive model (b = -17.2 minutes, p < .05). We now discuss how it affects the effects of the other predictors of the timing of evacuation. 1. According to ENT, the transformation of people's interpretation of the meaning of their environment brought about by the crisis is an important determinant of their subsequent behavior. As predicted, if the respondents perceived the situation as serious, they began evacuating somewhat sooner than if they did not (b = -8 minutes, not significant). The opposite is true, however, if they were in large groups and if they knew more people more intimately in these groups (b = 3.1 minutes, p < .05). For them, the greater the extent to which the search for meaning inherent in the milling process focused upon defining the situation as a serious crisis demanding an out-of-the-ordinary response, the longer it took them to mobilize and initiate their evacuation. Three other predictors tap elements of the threat experienced by the respondents and help us elaborate this finding. As shown in Table I, the greater is the degree of perceived danger generated by the crisis, the quicker Emergent Norm Theory 315 is the mobilization of respondents. However, this is the case only in the interactive model (b = -2.6 minutes, p < .05). In it, people who perceived more danger tended to initiate evacuation earlier. However, the reverse is true if they were in large evacuating groups and as they knew more people more thoroughly in these groups (b = .63 minutes, p = .05). A similar pattern is found when a measure of the smoke entering their room is used. As predicted, the greater the smoke was the quicker they began evacuating (b = -4.8 minutes, p < .10). Somewhat the opposite is true for those in large evacuating groups who knew more people intimately (b = .88 minutes, but not statistically significant). Finally, respondents who were injured as a result of the explosion tended to initiate their evacuation quicker than those who were uninjured (b = -18.3 minutes, p < .05), and this is true irrespective of the social context in which they found themselves. In sum, an enhanced sense of threat as precursor to protective behavior is mediated by the effects of social relationships. These results corroborate N. Johnson and his colleagues' finding about the importance of pro-social behavior in evacuations. They show that people exposed themselves to great personal risk to try to rescue or otherwise help their friends and known others. 2. ENT recognizes that the process of symbolic interaction in instances of collective behavior centers in part on the identification of skills, past experiences, and other instrumentalities among the participants. These elements of the situation are the resources employed by people to respond to the challenge they face (McCarthy and Zald, 1997). Their use takes time and slows down the enactment of collective behavior. We find partial support for this prediction. The results of this test support the prediction of a positive association between the extent of resources available to groups in the amount of time it takes people in them to begin evacuating. Contact with rescuers is characteristically an important source of information, assistance, and other resources for people faced with the need of evacuating. Such contact slowed down the beginning of evacuation behavior; respondents who came in contact with Rescuers delayed their start of the evacuation (b - 11,6 minutes, p < .05). This is also the case for respondents who knew others on their floor with EMT (b = 27.8 minutes, p < .05). However, the opposite is true (b = -3.52 minutes, p < .05) if they were in large groups and as they knew more people more thoroughly in these groups. Respondents' Familiarity with the WTC is not a statistically significant predictor in the additive model. It facilitates the early start of evacuation of respondents who were in large groups and knew others well in these evacuating groups (b = 1.14 minutes, p < .10). A possible explanation of these unexpected findings (with 316 Agnirre et al. EMT and Familiarity) is that large groups of people with enduring social relationships are able to ascertain and utilize available resources more quickly and effectively. 3. The results of this test partially support the prediction that the greater the extent to which the search for meaning inherent in the milling process focuses upon proposed cues for emergent action (such as discussions among people as to what to do and their helpfulness), the longer it should take them to agree on a course of action such as joining the evacuation. In support of this prediction, Cooperativeness is positively associated with the timing of evacuation (b = 3.95 minutes, p < .05). However, it is negatively associated with it (b = -.96 minutes, p < .05) for respondents who were in large groups and to the extent that they knew everyone well in those groups; they tended to join the evacuation earlier than their counterparts. Enduring social relationships and people who are perceived by the respondents as helpful and engaged in deciding what to do facilitate early collective behavior such as joining the evacuation. Contrary to the prediction, whether respondents received Information and Guidance from friends, office personnel, and others near them in the aftermath of the explosion is not statistically significantly positively related to the initiation of evacuation. 4. Results support the prediction from Brown and Goldin's (1973) elaboration of dramatist theory. Respondents who worked in floors with more than one firm started their evacuation later than their counterparts (b = 12.4 minutes, p < .05) in floors with one firm. This is also true for respondents in larger evacuating groups in which they knew everyone (b = 3.17 minutes, p < .05). Finally, statistical controls are important predictors of the timing of evacuation and have unexpected relationships with the extent of preexisting social relationships in the evacuating groups. Both the size of the population of floors (b = .02 minutes, p < .05) and the location of the floors in the buildings (b = .05 minutes, p < .05) are statistically significantly related to delaying the timing of the initiation of evacuation only for respondents who were in large groups and to the extent that they knew others well in their groups. Indeed, the additive pattern is for people in floors with large populations to join the evacuation quickly (b = -.12 minutes, p < .05) and to delay it with higher location in the buildings (b = .32 minutes, p < .05). Another pattern occurs with the age and gender of the respondents. While neither is statistically significant in the additive model, in the interactive model older respondents and female respondents tended to initiate their evacuation more quickly than younger and male respondents, respectively. However, the opposite is true if they were in large evacuating groups and to the extent that they knew people well in these groups. We can only 317 Emergent Norm Theory speculate that such social contexts provided these categories of respondents (also true for respondents in higher floors, b = .05, p < .10) with a greater sense of safety, so that they would delay joining the evacuation of their buildings. CONCLUSION We are indebted to the authors and critics of ENT for providing us with sufficiently clear guidance for the conduct of this test. Without their work, our own work would not have been possible. Ours is only one partial test based on one incident. The findings we report are tentative and should be replicated, hopefully with better data employing random, representative, and larger samples. It would be preferable to have shorter time delays between the occurrences of the collective behavior event and the data collection. We do not know to what extent respondents' recollections of the events may have become imprecise due to the weeks that had passed between their evacuation behavior and the collection of the information. Moreover, the present test depends on survey data that does not fully capture the processual, dynamic emergence of tentative definitions of the situation such as when to evacuate. Despite these problems, our hope is to qualify and augment elements of ENT's explanation of collective behavior and to contribute to the development of theory in the field of collective behavior. ENT is very much centered on people's perception and interpretation of events and on the milling and keynoting process inherent in instances of collective behavior. In its support, we have shown that the transformation of people's interpretation of the relative safety of their environment brought about by the crisis is an important determinant of their collective behavior. Moreover, engaging in milling, as reflected in helpful behavior and discussions about what needed to be done, has the predicted effect of delaying evacuation behavior. This is also true of the impact on the timing of evacuation of proselitization between work groups. Our finding regarding the effect of resources in delaying the initiation of evacuation also supports a prediction derived from ENT It underscores the value of recent conceptualization of resources in the social movement literature for understanding mobilization in instances of collective behavior. Results also document the validity of Weller and Quarantelli's structuralist argument regarding the occurrence of collective behavior. The extent of preexisting social relationships among respondents impacts the timing of evacuation behavior and the associated processes of social interaction. Symbolic interaction processes involved in the emergence of domi- 318 Aguirre et al. nant norms are embedded in these structures and must be understood in their terms. Social relationships, whether emergent or enduring, are not only helpful to differentiate collective behavior from institutionalized behavior. Enduring social relationships especify the social interactions that are associated with the emergence of a dominant norm in an instance of collective behavior, be it risk taking, the use of resources, or cooperativeness. We have shown that a threat as precursor to protective behavior is mediated by the effects of the social relationships among the people experiencing it. It bears repeating that these findings replicate N. Johnson and his colleagues' finding about the importance of enduring social relationships in eliciting pro-social behavior in crisis evacuations. ENT needs to be expanded so as to incorporate these results in a more specified explanation of the emergence of dominant norms in specific types of collective behavior. A renewed emphasis on preexisting social organization would make ENT a more general theory worthy of a prominent place in the sociological specialty of collective action/behavior. Such needed theoretical elaboration would recognize that the emergence or creation of norms in instances of social interaction is a common occurrence. Consequently, normative emergence cannot be used as the sole differentiating criterion for the identification of collective behavior (Wrong, 1994:50; see also Gusfield, 1994:67-69). "(N)orms grow in unplanned fashion out of ongoing interaction" (Wrong, 1994:49; social relationships create habits among participants that inform social expectations that have the potential to become new, latent norms guiding behavior. An appreciation of preexisting social organization would facilitate studying the linkage of the latent norms that may come to guide collective behavior and the norms that guide the collective action of social movements As Wrong recognizes (1994:53), echoing R. E. Park, some latent norms are adopted by new social actors and continue in operation through time, becoming institutionalized. These relatively well-known albeit unsolved difficulties in ENT would require for their resolution and the continued value of a symbolic interaction approach to collective behavior greater emphasis on the sociocultural context in which instances of collective behavior occur. Fortunately, such contextual specification is offered in Brown and Goldin's (1973) interpretation of E. Goffman's writings. From this perspective, social situations or culturally meaningful scenes bringing people together are analytically differentiated from encounters, or episodes of face to face interaction. Both are differentiated from the precipitating event that provides the topic for the ensuing symbolic interaction and the emergence of a shared definition of the situation and consensus for collective action. 319 Emergent Norm Theory We hope to have shown that the often heard argument by the critics that ENT could not generate testable hypotheses is incorrect. The rebirth of ENT is needed. Due to a number of reasons often unrelated to its scientific value, the last two decades or so has seen its relative popularity decline vis a vis the various variants of the resource mobilization approach to collective action (Aguirre, 1994; Killian, 1994; Aguirre and Quarantelli, 1983). The hoped-for reconsideration of ENT seems timely. New emphases on culture, collective memory, symbols the mass media, to name a few in the study of social movements, as well as the dramatic, seemingly unorganized, and unexpected transformations in Eastern European countries beginning in 1989, provide a more propitious intellectual context for it. REFERENCES Aguirre, B. E. 1994 "Collective behavior and social movement theory." In R, R. Dynes and K. J. Tierney (eds.), Disasters, Collective Behavior, and Social Organization: 257-272. Newark: University of Delaware Press. Aguirre, B. E. and E. L. Quarantelli 1983 "Methodological, ideological and conceptual theoretical criticisms of the field of collective behavior: A critical evaluation and implications for future studies." Sociological Focus 16:195216. Anderson, A., A. Basilevsky, and D. P. J. Hum 1983 "Missing data: A review of the literature." In Handbook of Survey Research: 415-494. New York: Academic Press. Brewer, John and A. Hunter 1989 Multimethod Approach. A Synthesis of Styles. Beverly Hills, CA: Sage, Brown, Michael and Amy Goldin 1973 Collective Behavior: A Review and Reinterpretation of the Literature. Pacific Palisades, CA: Goodyear Publishing Company. Gillespie, David F. and Ronald W. Perry 1976 "An integrated systems and emergent norm approach to mass emergencies." Mass Emergencies 1:303-312. Gusfield, J. R. 1994 "The reflexivity of social movements: Collective behavior and mass society theory revisited." In Enrique Larana, Hank Johnston and J. R. Gusfield (eds.), New Social Movements. From Ideology to Identity: 58-79. Philadelphia, PA: Temple University Press. Johnson, Norris 1987 "Panic at 'The Who Concert Stampede': An empirical assessment." Social Problems 34:362-373. Johnson, Norris and W. E. Feinberg 1994 "The emergence of competitive flight from the Beverly Hills Super Club fire." Unpublished manuscript presented at the ASA Annual Meetings, Los Angeles, CA. Johnston, D. M. and N. R. Johnson 1988 "Role extension in disaster: Employee behavior at the Beverly Hills Club fire." Sociological Focus 22:39-51. Killian, Lewis M. 1994 "Are social movements irrational or are they collective behavior?" In R. R. Dynes and K. J. Tierney (eds.), Disasters, Collective Behavior, and Social Organization: 273-280. Newark: University of Delaware Press. Kreps, Gary 1983 "The organization of disaster response: Core concepts and processes." International Journal of Mass Emergencies and Disasters 1:439-464. 1985 "Disaster and the social order." Sociological Theory 3:49-65. McCarthy John and Mayer Zald 1997 "Resource mobilization and social movements: A partial theory."In S. M. Buechler and F. K. Cylke, Jr. (eds.), 320 Social Movements: Perspectives and Issues: 149-171. California: Mayfield Publishing Company. McPhail, Clark 1991 The Myth of the Madding Crowd. New York: Aldine. Mintz, Alexander 1951 "Nonadaptive group behavior." The Journal of Abnormal and Social Psychology 46:150-159. Mawson, Anthony R. 1980 "Is the concept of panic useful for scientific purposes?" In Second International Seminar on Human Behaviour in Fire Emergencies 208-211. Washington DC: National Bureau of Standards. Moscovici, Serge 1994 "Three concepts: Minority, conflict, and behavioral style." In S. Moscovici, A. Mucchi-Faina, and A. Maass (eds.), Minority Influence: 233-251. Chicago, IL: Nelson-Hall Publishers. Norusis, Marjja 1994 SPSS. SPSS Advanced Statistics, 6.1. Chicago, IL: SPSS, Inc. Sime, Jonathan D. 1980 "The concept of panic." In D. Canter (eds.), Fires and Human Behavior: 6381. New York: John Wiley & Sons. 1983 "Affiliative behaviour during escape to building exists." Journal of Environmental Psychology 3:21-41. "Original publication date. Aguirre et al. Sorensen, John H. 1991 "When shall we leave? Factors affecting the timing of evacuation departures." International Journal of Mass Emergencies and Disasters 9:153-166. Tierney, K. 1980 "Emergent norm theory as theory: An analysis and critique of Turner's formulation." In M. D. Pugh (ed.), Collective Behavior: A Source Book: 42-53. St. Paul, MN: West Publishing Company. Turner, Ralph 1964 "Collective behavior." In R. E. L. Paris (ed.), Handbook of Modern Sociology: 382-425. Chicago, IL: Rand McNally. 1996 "Normative emergence in collective behavior and action." Mobilization 1:1-14. Turner, Ralph and Lewis Killian 1987 Collective Behavior. (1972*) Englewood Cliffs, NJ: Prentice-Hall. Weller, Jack and E. L. Quarantelli 1973 "Neglected characteristics of collective behavior." American Journal of Sociology 79:665-685. Wrong, Dennis 1994 The Problem of Order. Cambridge, MA: Harvard University Press.