Description
This course introduces methods for analyzing correlated categorical data such a commonly occur due to repeated measurement or other forms of clustering. The course begins with a review of some older, established methods for matched pairs, including McNemar's test. The main focus, though, is on two types of models. One type models the marginal distributions, with parameter estimation most commonly handled with generalized estimating equation (GEE) methodology. The other type generalized linear mixed models, uses random effects to represent the mechanism by which within-cluster correlation occurs. For each type, focus is mainly on logit models for binary responses but with some discussion of ordinal responses. Brief discussion will also be given of other mixture models for handling overdispersion for such data, such as beta-binomial models and related quasi-likelihood methods. Several examples will be shown, with implementation using SAS. The presentation will be at a low technical level, with emphasis on concepts rather than technical details of model fitting. However, attendees should have some background in generalized linear modeling, especially logistic regression.
|