Fitting injury versus exposure data into a risk function

Abstract

A problem in safety research is how injury data from accidents, and damage data from laboratory experiments, should be concisely reduced into a single descriptive formula. The uncertainty that comes from variability of a statistical nature in biological subjects and in the physical course of events producing injuries and damages is well known. A less known source of uncertainty in the violence and loading data is that they most often are censored, that is, biased, the sign of the bias being known, but not its magnitude. The purpose of this report is to describe what censoring is and how censored data can be effectively analyzed. It is shown that the maximum likelihood method can be used for the fitting of a Weibull cumulative risk function to censored biomechanical data. It is found that the method is easily handled with the aid of available computer programmes. The three parameter Weibull distribution provides a result that gives a better representation of censored data than earlier methods used in biomechanics. Several of a selection of data sets have shown a marked right skewness, which speaks against the usual application of the normal distribution. For the covering abstract of the conference see TRIS 393117. (TRRL)