Cluster analysis is a statistical classification technique that can identify subgroups of people with similar profiles on a research measure or measures. Given its exploratory nature, cluster analysis has a somewhat controversial history in traffic safety research. It has been argued that the ability to produce different solutions using different methods reduces the credibility of cluster analysis as a useful measure for traffic safety research. Part of the problem has been the inconsistent selection of clustering techniques by researchers. This is true even though studies of biological data with known structures have identified and recommended particular methods to apply to particular types of data. In addition, many validation techniques have been developed to test selected solutions. This paper presents an overview of these issues, including two recent Australian studies that have included cluster analyses; one relating to seatbelt use and the other to speeding behaviour. In each case profiles of road-user subgroups were identified based on differences in self-reported road safety attitudes and behaviours. The studies also used factor analysis to first reduce the amount of variables entered into the cluster analysis. In this way multiple measures could be included. It is argued that this combination, factor analysis followed by cluster analysis, can provide a powerful exploratory tool for traffic safety research. (A) For the covering abstract of the conference see ITRD no 207828. The reprints are also available at the web - http://www.vti.se/pdf/reports/K18APart1.pdf; http://www.vti.se/pdf/reports/K18APart2.pdf and http://www.vti.se/pdf/reports/K18APart3.pdf.
Abstract