A model for multivariate binary data with covariates based on compatible conditionally specified logistic regressions

• Main Author: Liu, Ying
• Format: Others
• Language: English
• Published: 2009
• Online Access: http://hdl.handle.net/2429/5380

Rather than construction of a multivariate distribution from given univariate or bivariate margins, recently several papers seek to promote the development and usage of a simple but relatively unknown approach to the specification of models for dependent binary outcomes through conditional probabilities, each of which is assumed to be logistic. These recent proposals were all offered as heuristic approaches to specifying a multivariate distribution capable of representing the dependence of binary outcomes. However, they are limited in scope, for they all describe some special patterns of dependence. This thesis is concerned with a model for a multivariate binary response with covariates based on compatible conditionally specified logistic regressions. With this model, we allow for a general dependence structure for the binary outcomes. Three likelihood-based computing methods are introduced to estimate the parameters in our model. An example on the coronary bypass surgery is presented for illustration. === Science, Faculty of === Statistics, Department of === Graduate