Analysis and Testing of Koornstra-type Induced Exposure Models

Induced exposure models postulate a structure for accident data which permits the\nestimation of two factors: exposure and proneness. Since information on exposure\nis needed in order to assess the accident risk of different driver, vehicle, and\nenvironmental situations and since reliable exposure data is expensive to collect,\ninduced exposure models hold the promise of a rich exposure data source which is\nperfectly matched to the accident data to be analyzed. This paper assesses the\nvalidity of the postulated structure of one induced exposure model: the Koornstra\nModel. The Koornstra Model was chosen for analysis and testing because it\nappeared to have the most potential for usefulness based on :\na) previously reported favorable results (in limited testing);\nb) universal applicability;\nc) damaging criticism of certain other models; and\nd) a rich and well posed model structure.\nThis paper analyses and tests the Koornstra Model from three different points of\nview:\n1) Is it based on reasonable assumptions?\n2) Does the model provide a significantly better fit to accident data than\na simpler model which does not permit exposure or proneness to be\nestimated?\n3) How do the exposure estimates provided by the model compare with\nthose from externally collected data?\n