LOGISTIC ANALYSIS OF STUDIES WITH 2-STAGE SAMPLING - A COMPARISON OF 4 APPROACHES

This paper discusses the analysis of two-stage studies where covariate s are missing or measured with error at the first stage of sampling an d are validated at the second stage in a subsample. Four recently deve loped approaches, the weighted pseudo-likelihood method of Flanders an d Greenland (1991), the pseudo-conditional likelihood methods of Bresl ow and Cain (1988) and Schill et al. (1993) and the maximum likelihood estimate obtained via the EM-algorithm (Wacholder and Weinberg, 1994) are reviewed, and some connections between them are established. It i s shown that, with respect to odds ratio estimation, case-control desi gns can be analysed as if first-stage sampling had been prospective. T he procedures are numerically compared with respect to asymptotic rela tive efficiency in a missing value setting.