The midterm will cover the material in lectures 1-13 prior to Fall Break.

Course description:

This course covers the basic theory, methodology, and application of generalized linear models. Brief background will be provided on standard approaches to frequentist inference, but the focus will be Bayesian methodology. The course will contain an introductory overview of Bayesian methods for analysis of binomial, count, categorical, and event time data. Extensions of univariate GLMs to accommodate longitudinal, clustered, and multivariate data will also be covered. A particular emphasis will be on latent variable methods and their use in hierarchical GLMs. Methodology will be motivated by case studies and applications, primarily using biomedical data. Some discussion of advanced topics, including model selection and averaging, joint modeling of data having different scales, and approaches for order restricted inference will be incorporated.

Textbook:

The course will not use a textbook but notes will be available through the website and outside readings will be assigned. Good reference textbooks include "Generalized Linear Models" by McCullagh & Nelder (Chapman & Hall) and "Generalized Linear Models: A Bayesian Perspective" edited by Dey, Ghosh & Mallick (Marcel Dekker, Inc)

Student Responsibilities:

Assignments: Outside reading and problem sets will be assigned regularly. Students will be evaluated on their class participation and ability to respond to questions about the covered material. [15% of final grade ].

Mid-term Examination: An in-class closed-book examination will be given covering the first 1/2 of the course material [25%].

Project: Students will be expected to write up [15%] and present [15%] results from a complex data analysis project.

Final examination: There will be an out of class final examination [30%].

Computing:

Computational algorithms will be discussed without focusing on specific software or packages. Students will be expected to have adequate background in computing to implement algorithms (e.g., Markov chain Monte Carlo) out of class without further instruction.

Datasets:

1. Smoking and childhood obesity : data consist of repeated body weights and indicators of obesity for a random subsample of girls from the Collaborative Perinatal Project (CPP).

Variable key

2. DDE and Preterm Delivery Dataset: Data from the Longnecker et al. (2001, Lancet) study relating maternal serum concentration of the DDE metabolite DDE and preterm and small-for-gestational-age babies at birth. [Variable key: x=dde dose, y=indicator preterm birth, z1-z5 = potential "confounders"]