This paper identifies mathematical models that can be used to identify potentially influential measures on accidents. Modelling is necessary because of the underlying noise found in any set of data. The noise tends to obscure relationships among measures, and the modelling process helps to see what measures have some explanatory power. Different generalised linear models (GLMs) can be used, and they make different assumptions. It may be necessary to consider a number of approaches and models in order to find the 'best' model for the data being analysed. A 'goodness of fit' statistic can be used to identify which model explains most of the variation in the data. An example of a simple GLM illustrates how models can be used to identify relationships between accidents and available measures. Specifically, the example showed how an apparent relationship reversed, once influencing measures had been controlled for. This is clearly desirable when the data contain interacting measures. The use of path models is a different and more holistic approach. It recognises that an outcome, such as an accident, depends on many influences. The immediate influence, such as driving behaviour, may influence accident liability, but the behaviour may be influenced by an attitude that in turn may vary with age and sex. Modelling the 'reality' and paths of influences on driving may help to gain a better understanding. Further, it may help to identify which measures should be targeted because they not only influence the final accident risk but can be influenced via appropriate road safety campaigns. For the covering abstract see ITRD E0903020.
Samenvatting