This paper examines three methods for estimating the relative risk of a driver fatal accident as a function of the driver's blood alcohol level: nonparametric estimation using accident and control (or exposure) data; parametric estimation using accident data either alone or with a very small sample of control data; generalized linear models using accident and control data. The second method was investigated because in Australia control data are scarce in precisely those states where good accident data are available. If successful, such an approach would have allowed estimation of risk for cohorts for which extensive control data are unavailable. As shown the approach is full of difficulties. The third method is The third method is particularly easy to use and allows socially relevant hypotheses, such as age and sex differences in relative risk, to be tested. The methods are applied to accident data from New South Wales and control data from South Australia. (A)
Samenvatting