Analysis of a case-control study with differential measurement error
by
Gary Glonek
The University of Adelaide

This talk describes the analysis of a case-control study, undertaken by the Adelaide University Road Accident Research Unit, on effect of speeding on serious car accidents. By the nature of study, it is to be expected that large errors would occur in the measurements of speed for the crash vehicles but not for the controls. In general, it is known that estimates based on ordinary logistic regression can be biased in the presence of differential measurement error with attenuation in the regression coefficients being an often cited consequence.

Because replication was not available in the present study, the measurement error variance cannot be estimated and the approach taken was that of sensitivity analysis. Perhaps surprisingly, it was found that the estimates of increased risk were very stable across a wide range of values for the measurement error variance. The method of analysis will be described and an explanation for the apparent insensitivity to the measurement errors provided.

Date received: April 29, 2002