Testing genetic hypothesis on bivariate dichotomous twin data using repeated measures logistic regression
The purpose of this dissertation is to develop methods for testing if specific phenotypes are a result of genes and/or environmental influences using an analysis of bivariate categorical twin data. In the literature, several methods are available for analyzing twin data. These include: Logistic Regression method and Structural Equation Modeling (SEM). Recently the Generalized Estimating Equations (GEE) under the logit link has been suggested for modeling bivariate twin data, where data on two traits are simultaneously considered. A new measure for the effects shared by the two traits was proposed as an alternative to the cross-twin cross-trait odds ratio.
This work proposes and evaluates new measures for the two traits, which is termed, the Linked Trait Effect (LTE). Using this measure the bivariate logistic regression model is extended to include individual level covariates. A theoretical justification for the proposed covariate model is provided for an individual level dichotomous trait. The methodology is then compared to the existing bivariate regression approaches, both theoretically and in applications. The proposed methodology is then applied to data from two sources. Bivariate twin data on drinking and smoking with individual level covariates such as marital status and church attendance from the Mid-Atlantic Twin Registry (MATR) is analyzed.