A CONDITIONAL MODEL FOR INCOMPLETE COVARIATES IN PARAMETRIC REGRESSION-MODELS

Incomplete covariate data arise in many data sets. When the missing co variates are categorical, a useful technique for obtaining parameter e stimates is the EM algorithm by the method of weights proposed in Ibra him (1990). This method requires the estimation of many nuisance param eters for the distribution of the covariates. Unfortunately, in data s ets when the percentage of missing data is high, and the missing covar iate patterns are highly non-monotone, the estimates of the nuisance p arameters can lead to highly unstable estimates of the parameters of i nterest. We propose a conditional model for the covariate distribution that has several modelling advantages for the E-step and provides a r eduction in the number of nuisance parameters, thus providing more sta ble estimates in finite samples. We present a clinical trials example with six covariates, five of which have some missing values.