Predicting human behaviour in emergencies

The outcome of an emergency is largely determined by the behaviour of the people involved. To improve the safety of buildings and to increase the effectiveness of response procedures and training programmes it is often necessary to predict human behaviour in emergency situations. There are several approaches which can be used to make these predictions, but not all had previously been systematically analysed and therefore their appropriateness for any given application was unknown. This thesis describes an analysis of approaches for predicting human behaviour in emergencies. The research focussed on approaches which could be used by human factors professionals to extend the contribution this systems-oriented and user-focussed discipline can make to managing risks and reducing danger. The investigated approaches were evaluated against criteria for judging their quality, including validity, reliability, resources, sensitivity and ethics. In research conducted to test the approaches, fire drills, virtual environments (VEs) and a new talk-through approach, in which participants describe the hypothetical actions they would take in an emergency scenario, demonstrated potential for predicting behaviour in emergency situations. These approaches were subsequently evaluated in a standardised comparison, in which each one was applied to analyse the behaviour demonstrated during an evacuation from a university building. The observed frequencies of behaviour produced by each approach were significantly correlated, as were the sequences of behaviour. All of the approaches demonstrated replicability. The resources required to apply each approach were relatively low, especially for the talk-through approach. Based on the findings from this research, and drawing upon previous work from the scientific literature, guidance was provided for selecting approaches and methods for behavioural prediction in emergency situations. The talk-through approach is suitable for use during the concept phase of a design as it is quick to implement and requires low resources. VEs and simulation tools are more appropriate for design activities when detailed CAD models become available. Fire drills can provide useful measures of human behaviour in evacuation scenarios, but require a physical representation of the building or environment under investigation. Fire drills, VEs and simulation tools can be used to inform emergency response procedures. Predictions from all of the aforementioned approaches can support the development of training programmes. This guidance was previously unavailable to human factors professionals and now serves both to inform design work and support the evaluation of existing evacuation procedures and protocols.