Coping with a Wider Range of Outcomes: Demonstration of the Generalized Linear Model Platform in SPSS Statistics

• Activity Number: 259
• Type: Contributed
• Date/Time: Monday, August 1, 2011 : 2:00 PM to 3:50 PM
• Sponsor: Section on Teaching of Statistics in the Health Sciences
• Author(s): Letao Sun*+ and Hongwei Yang
• Companies: University of Kentucky and University of Kentucky
• Keywords: regression ; generalize linear model ; link function ; SPSS ; multilevel

Abstract

A common problem shared by researchers using regression analysis in health, social and behavioral sciences is that the outcome fails to be normally distributed. This includes data that are continuous, discrete, censored, and a combination of all these types. To appropriately analyze such data, link functions are utilized to link the mean of the outcome to the linear predictor, a linear combination of all predictor variables. Such a modeling framework is known as generalized linear modeling (GLM). Featuring a user-friendly environment, SPSS provides a procedure (GENLIN) for GLM analysis with a variety of built-in link functions. This procedure enables applied researchers to harness the power of GLM without any syntax-based coding. This study examines link functions available in the SPSS GLM procedure with a primary focus on how to quickly help applied researchers get started in using the procedure to tackle a wider range of data with different types of outcomes. Examples of applications in health sciences are presented. A future expansion of this study is to analyze more complex data structures (multilevel, etc.) in the GLM framework using this cost-effective, easy-to-use program.