Accident Proneness
Abstract
The entry reviews five competing models for the occurrence of accidents: (1) pure chance, (2) true contagion, (3) apparent contagion, (4) spells, and (5) liability. The distributions for accident count data that they yield (e.g. Poisson, negative binomial, Neyman type A, short, and generalized Waring) can, however, arise from more than one model. Model identification, in order to modify behavior and hence reduce the number of accidents, has proved largely ineffective. Since 1985, research has moved from model detection to risk evaluation and analysis.