Many statistical techniques (ordinary least squares regression, ANOVA, ANCOVA) are based on normally distributed dependent variables. However, data that are not normally distributed are commonly encountered. These include binary outcomes (whether a pregnant woman went to the prenatal clinic or not), count data (number of children a woman had) and categorical data (type of car a person buys). Generalized linear models are appropriate for all of these types of outcomes, and for normally distributed responses. When there is clustered data, repeated measures, or longitudinal data, the correlations among observations on the same unit need to be taken into account.
This two-day workshop will cover uses of the Generalized Linear Model for cross-sectional data and GEE (Generalized Estimating Equations) for clustered, repeated measures or longitudinal data. Logistic regression, Poisson regression and loglinear models will be discussed. Instruction emphasizes when to use each method, as well as interpretation of SAS output and checking for model appropriateness. There will be a morning lecture and an afternoon hands-on lab, using SAS Proc Genmod to analyze real data sets.
Please make check payable to CSCAR-University of Michigan, or give the University of Michigan shortcode to be billed. Send check to CSCAR, 3550 Rackham Bldg., University of Michigan, 915 E. Washington St., Ann Arbor, MI, 48109-1070.