Effects of social identity on responses to mass emergency evacuation (ESRC project)

Find out how we are investigating the effects of social identity on public responses to emergency mass evacuation.

The project

The ESRC research project sought to develop a novel approach to emergency mass evacuation based on the social identity perspective. We argue that this approach can help explain the conditions determining whether emergency evacuation behaviour is un-coordinated and individualised (and hence collectively ineffective and dangerous) or co-ordinated and co-operative (and hence more efficient for the crowd as a whole).

Our hypothesis, research and studies

Early studies suggested that crowds panic when faced with danger and a limited exit. Later work suggested that behaviour would instead be limited by social rules and affective ties, so people were more likely to help others rather than compete to escape.

Our project has sought to help explain gaps in existing theoretical models – such as why it is that people in emergencies take risks to help strangers.

Our initial hypothesis was that mutual concern, helping and co-ordination were more likely when crowd members shared a common identity.

The personally selfish behaviour associated with 'mass panic' may occur if there was no shared identity.

Three kinds of studies were carried out:

  • interview studies with survivors
  • computer visualisation (virtual reality) experimental studies
  • room evacuation experimental studies.

We also managed to begin an analysis of the London bombings of July 2005.

Finally, we started a survey study of (over)crowding and evacuation at a Brighton beach party in 2002, based on some of the interview findings. 

Find out more about our studies below. 

  • Comparative interview study 

    We recruited 21 survivors of 11 real and perceived emergencies. We carried out in-depth interviews, subjecting the data to qualitative and quantitative analysis.

    The results led to an elaboration of our hypothesis and a more dynamic account of psychological responses to emergency evacuation and disasters.

    First, we found no evidence for events which could be characterised as ‘mass panic’. This supported the literature, according to which panic is extremely rare.

    Second, we found that, while our data included different crowd events, there was a common experience of uniting in response to the emergency.

    In crowds where there was already some common identity – such as football fans – the sense of unity broadened to cover all those affected by the danger, not just supporters of the same team. We were able to show that, while a common identity did indeed predict mutual helping and concern (and even self-sacrifice to help strangers on some occasions), a common identity itself is an emergent function of the experience of an emergency or disaster.

    You can find out more in the following publications:

  • Visualisation study

    One of the aims of the present project was to develop an experimental simulation that was both engaging and ethically sound. We therefore compared two types of experimental simulation to determine which was more suitable for further development: (i) a visualization (‘virtual reality’) design, and (ii) a room evacuation design.

    The visualisation developed by the team in Nottingham was an animation of a crowd evacuation from an underground railway station, and modelled on a computer game (using similar graphical techniques and user interface).


    The ‘task’ facing the user is to evacuate the station as soon as possible, while at the same time facing bottlenecks caused by the rest of the crowd. The user also has to make decisions about whether to stop and help four people who are apparently injured. Within this design, we were able to vary key dimensions, such as the appearance of the characters in the evacuation, and the number of other evacuees, and to enhance the urgency of exit through a varying ‘danger’ indicator.

    Participants’ identity was varied by a vignette at the beginning of each trial which cast them either as group members or individuals in an aggregate crowd. We then looked at the number of times ‘injured’ characters were helped (or not) and the extent to which participants pushed characters out of the way. A post-test questionnaire assessed the role of shared identity, feelings towards others, and intentions to help.

    The results

    72 people took part in the first experiment. While it turned out to be difficult to get people to think of themselves as group members (versus individuals) in the way we intended, there was nevertheless a correlation between feelings of psychological ‘groupness’ and the amount of helping. That is, the more people saw themselves as group members, the more likely they were to stop and help fallen characters – even though such action delayed their own exit. This result was replicated in a student project on 40 participants using a different vignette. (Thanks to Becky Powell for this work.) In both cases, the more people saw themselves as group members, not only did they help more, but on the subjective (questionnaire) measures they cared for others more and expressed a greater desire to help.

    A third experiment in which 62 people took part eliminated potential problems with the design but weakened the identity manipulation further. There was more helping in the ‘group identity’ than the ‘personal identity’ condition but this difference was not statistically significant. 40 people took part in a fourth experiment, which was carried out as a student project. (Thanks to Andy Hardwick for this work.) This study used a different vignette and identities to the previous ones (‘football supporters’ instead of ‘students’). Those in the ‘group identity’ condition displayed significantly more helping behaviour than those in the ‘personal identity’ condition. There was some support for the idea that this behaviour was partly caused by positive feelings towards ingroup members.

    In each of these studies, while there was some evidence in support of the role of shared identity on the helping and questionnaire measures, there was no pattern in the ‘pushing’ data. Observations of participants in these studies, as well as at a public exhibition at the Royal Society, led us to conclude that there was a problem with the on-screen instruction at the beginning which explains how to push other characters. This could be read as an encouragement to push in what might be perceived as a ‘game’. The visualisation was modified three times within the project, each time achieving greater realism. But it was beyond the scope of the current project to address this problem in the instructions through further changes to the software.


    Overall, the visualisation studies support the conclusion that, where there is a strong sense of collective identity, there will be mutual concern and helping. Importantly, people with a strong shared identity in these studies tended to help the fallen character even though this meant delaying their own exit. Where the sense of shared identity was weak, on the other hand (either through our deliberate manipulation or because the manipulation was weak), there was less mutual concern and less helping. The significant results that were found were therefore in line with the self-categorisation account of mass emergency evacuation behaviour.

    In terms of the aim of developing an experimental simulation that was both engaging and ethically sound, our conclusion is that the visualisation method has more potential than the room evacuation method. There was more psychological engagement with the visualisation than the room evacuation method: participants took the visualisation more seriously overall. The visualisation also included more measures (more opportunities for helping versus personally selfish behaviours) than the room evacuation.

    For details of studies and results, see:

  • London bombings: 7 July 2005

    The tragedy of the 7July 2005 London bombings offered an opportunity for further testing of our hypotheses.

    In comparison to the earlier interview study, there would be less reliance on memories of distant events, and more comparability of experiences. Also, the availability in the media and internet of so many eye-witness accounts meant that we expected a much bigger dataset.

    We have so far gathered 141 newspaper accounts, more detailed secondary accounts (from websites, the GLA enquiry and magazines) from over 30 eye-witnesses, and a substantial primary data-set (14 e-mail accounts and 12 interviews).

    For our preliminary analysis, see our report for user groups

    For details of studies and results, see: 

You might also be interested in: