CONDITIONAL LOGISTIC-REGRESSION WITH MISSING DATA

Methods for parameter estimation in conditional logistic regression analysis with missing data are compared in a simulation study. The data consist of a matching variable and two covariates, one of which has missing values. Sample size, percent of data missing, and the relationships between the variables are varied. Four methods are compared when the covariates are continuous: 1) Analysis based on only the complete cases, 2) Substituting the mean of the observed values for the missing value, 3) Regressing the missing value on the remaining variables in the stratum and 4) Adding a random component to the predicted value obtained from the regression. Four additional methods are compared in the situation when the variable with missing values is dichotomous. Linear regression methods are modified and two methods based on logistic regression are employed. Percent relative bias, confidence interval width and confidence interval coverage are compared. In general, the complete case method and the regression methods with an error term added perform the best. A maximum likelihood-based estimator is considered.