TESTS OF INDEPENDENCE FOR BIVARIATE DATA WITH RANDOM CENSORING - A CONTINGENCY-TABLE APPROACH

Procedures are proposed for testing the hypothesis of independence between two discrete or discretized random variables, one or both of which may be randomly censored. The censoring mechanisms can be either independent (bivariate censoring) or identical (univariate censoring). The proposed tests reduce to goodness-of-fit tests for log-linear models applied to either complete or incomplete contingency tables. Three examples are analyzed for illustrative purposes.