Identifying Missing Data Mechanisms in (2 x 2)-Contingency Tables

Consider the sample of two binary variables X and Y with some
missing structure within X or Y. The knowledge about the
corresponding values of the observed covariate allows to play
through all possible `originally' complete data sets. After
defining the notation, including some theoretical work, a test for
non--MCAR within the complete case table is presented. Simulating
all possible tables enables some testing on non--MAR. A simulation
experiment is used to illustrate this context.