Score test for testing etiologic heterogeneity in two-stage polytomous logistic regression

Karagülle, Saygın
Two-stage polytomous logistic regression was proposed by Chatterjee (2004) as an effective tool to analyze epidemiological data when disease subtype information is available. In this modeling approach, a classic logistic regression is employed in the first level of the model. In the second level, the first-stage regression parameters are modeled as a function of some contrast parameters in a somehow similar spirit of an ANOVA model. This modeling also enables a practical way of estimating the heterogeneity in the probabilities of occurrence of different subtypes given a certain covariate set. However, the only way of testing for significance of the heterogeneity is the Wald test, so an alternative test has yet to be developed. In this context, the aim is to develop a score test and examine both the asymptotic and finite sample properties of the test. The simulation results showed that a minimum average expected subtype frequency, depending on the number of disease subtypes and total sample size, must be attained for the asymptotic distribution of the score test to hold. For the cases in which it is implausible to make asymptotic distribution assumption, through an extensive Monte Carlo simulation study, use of permutation test-based critical values were suggested.
Citation Formats
S. Karagülle, “Score test for testing etiologic heterogeneity in two-stage polytomous logistic regression,” M.S. - Master of Science, Middle East Technical University, 2013.