NOTE: This video has some issues with the speed of the sound (it's not 100% in sync but is not off by a static amount). To flip through the slides yourself, please visit the course page (http://www.lisa.stat.vt.edu/?q=node/512).
LISA (Laboratory for Interdisciplinary Statistical Analysis) provided a series of evening short courses to help graduate students use statistics in their research. The focus of these two-hour courses was on teaching practical statistical techniques for analyzing or collecting data.
April 6, 2010:
Generalized Linear Models
Instructor: Mark Seiss
The third LISA mini course for the 2010 Spring semester focuses on appropriate model building using generalized linear models. While multiple linear regression models are common for Normal data, they are not appropriate for non-Normal data. This short course introduces attendees to linear models used for non-Normal data, with an emphasis on categorical response data such as binary or count.
The most common way to analyze a binary response (Yes/No or 0/1 outcomes) is the Logistic regression model, which is a linear model with a logit transform of the response mean. The most common way to analyze a count response (whole numbers) is the Poisson regression model, which is a linear model with a log transform of the response mean. The course will explain these two models in detail and how to interpret the results. The course will also work through examples of the application of each model using statistical package JMP and will explain all output that is produced.