Time Schedule:
Norman Breslow
BIOST 573
Seattle Campus
Advanced topics in generalized linear models and the analysis of categorical data: overdispersion, quasilikelihood, parameters in link and variance functions, exact conditional inference, random effects, saddlepoint approximations. Credit/no-credit only. Prerequisite: BIOST 571 and STAT 582. Offered: jointly with STAT 573; alternate years; Sp.
Class description
This class explores methods for the analysis of data presented in the form of contingency tables or where the outcomes are discrete (binary, multinomial) or in the form of rates (cases and person-years of observation). The main text is Alan Agresti’s Categorical Data Analysis (2nd Edition, 2002) with supplemental material drawn from McCullagh and Nelder’s Generalized Linear Models (2nd Edition, 1989), Cox and Snell’s Analysis of Binary Data (2nd Edition, 1989) and journal articles. Topics will include a selection from: review of generalized linear models (GLMs) via quasi-likelihood; (robust) score tests; overdispersion; joint inference on means and variances; constructed variables for non-linear parameters in link function and covariates; bootstrapping GLMs; marginal vs log-linear models for contingency tables; hierarchical vs marginal models for correlated data; conditional logistic regression for matched sets; exact inference based on conditional likelihoods and associated computing algorithms; sample survey (Horvitz-Thompson) vs NPMLE methods for two-phase stratified sampling designs.
Student learning goals
Recognize situations where the routine application of logistic, Poisson or other standard regression modeling techniques are likely to lead to incorrect inferences
Adapt standard software to accommodate non-standard aspects of the data
Understand the relationship between popular techniques of categorical data analysis and the fundamental principles of likelihood and estimating equations
Use these principles to develop appropriate methods of analysis for situations not covered in the standard texts
General method of instruction
Lectures, homework assignmentss and discussion.
Recommended preparation
Biostatistics 571 and Statistics 582
Class assignments and grading
Selected problems from Agresti's CDA; analysis and interpretation of data using methods discussed in class; derivation of simple statistics from first principles
This course is graded C/NC. To receive credit, students are expected to attend class regularly, complete homework assignments satisfactorily and present results of a term project both orally and in writing.
