Covariate models for accident data

Abstract

The statistical analysis of accident data has historically relied on the mathematical precepts of classical discrete distribution theory Since the first test of the standard null ("pure chance") hypothesis on accident mortality data, relating to horse-kicks in ten Prussian Army Corps during 1875-1894 (von Bortkiewicz, 1898), various models have been advanced to explain departure from Poisson's (1837) law. The ostensibly separate hypotheses of "proneness" (termed "unequal liability" Greenwood and Yule (1920) and "false contagion" by others Bates, et al , (1952), and "contagion" have received extensive attention in the literature (Newbold, 1927, Anscombe, 1950, Bliss and Fisher, 1953, Neyman, 1939, Cresswell and Froggatt, 1963, Kemp and Kemp, 1965 and Kemp, 19670). However, despite considerable development of the associated statistical models, formidable problems of interpretation remain (Froggatt, 1968a).