Taking account of system interactions in modelling road accidents

This paper argues that modelling of accidents and injuries not based on the broadest possible conceptualization of the accident generation process is likely to suffer from unintended biases. Further, careful consideration must be given to the objectives for which the modelling is designed. The analysis recognizes that the form modelling can take is constrained by data availability. Alternative modelling techniques are discussed along with the advantages and weaknesses of each. Finally, it is recommended that, when possible, a battery of techniques be used, to guarantee needed information is obtained while assuring the results are not simply a biased product of the technique used.