Conditional Independence Models for Two-Way Tables

Abstract

Christiansen analyses the data as a 2 × 2 × 4 contingency table with the three variables gender, age, and health concern. In the presence of the missing frequencies, he then fits a log-linear model with a zero second-order interaction term. This approach appears to smooth the main feature in the data. The latter will lead to the key conclusion that the younger males are over-concerned, in relation to the remainder of the sample, regarding how healthy they are. The chapter shows how to investigate quasi-independence hypothesis. In the presence of fixed zeroes, algebraically explicit estimates for the row and column effects do not typically exist.