A Comparison of Minimum Logit chi Square Estimation and Maximum Likelihood Estimation in RXCX2 Contingency Tables: Tests for Interaction

摘要

The exact small sample properties of twelve goodness of fit tests for interaction for 2x2x2 and 2x2x3 contingency tables are evaluated by enumeration and Monte Carlo sampling. Several modifications of the minimum logit chi square test, the Pearson chi square test, and the likelihood ratio test using an iterative maximum likelihood estimation procedure and three variants of a non-iterative minimum logit chi square estimation procedure are considered. Three criteria are used to evaluate the properties of the tests: (1) the adequacy of the tabular chi square distribution as an approximation to the exact level, (2) the power of the tests, and (3) ease of computation. The minimum logit chi square test (chi sub 13 superscript 2) based upon the non-iterative minimum logit chi square estimation procedure performs the best with respect to the three criteria. (Author)