Statistical methods for developing and distinguishing multinominal response models in the traumatological analysis of simulated automobile impacts.

Simulated car- to- car side impacts, designed for the analysis of traumatological aspects, involve two sets of variables. Predictors include exogenous biomechanical factors as well as anthropometric variables, such as age. The response is measured a scale of injury scores and is thus multinominal A goodness- of- fit test based on bootstrapping flexible, nonparametric alternatives to a given parametric candidate model is discussed, and several graphical techniques that allow relatively simple comparisons of different models are presented.