This study develops closed-form maximum likelihood estimates for data from contingency tables with missing values on some or all variables. We derive the estimates for two-way contingency tables via loglinear models that are easy to specify and that generalize readily to multi-way tables with missing data on some or all variables. The loglinear models we propose have the advantage of providing information about the missing-data mechanism and ascertaining how assumptions about the missing-data mechanism affect inference. Results of the method for two-way contingency tables appear in Statistics in Medicine. This project is completed.
