Two‐Sample Problems for a Dichotomous Variable with Missing Data

Summary Two‐sample problems with dichotomous data are considered; probability models which generate missing data are proposed. The statistical techniques appropriate under each model for missing data are discussed and evaluated. Results indicate that inferences based on the estimation of the odds ratio p 1 (1 ‐ p 2 )/ p 2 (1 ‐ p 1 ) are more robust to non‐randomly missing data than inferences based on estimation of more standard measures such as p 1 ‐ p 2 or p 1 / p 2 .