Logistic Regression and the Generalized Linear Model

This chapter explains why binary dependent variables require a different statistical model than do continuous ones. It discusses the difference between logistic regression and ordinary least-squares and odds and logged odds. The chapter introduces continuous variable, which is called as logit that is log of the odds. It shows how to compute a percentage change in the odds. The chapter also explains how to model several predictors in multiple logistic regression. It examines the nature of the generalized linear model, and how many models can be subsumed under it. When the response variable is binary or polytomous, either naturally occurring or by design of the investigator as in the case of the number of car accidents, a new type of statistical model is required. These models, generally known as generalized linear models, effectuate a transformation on the binary or polytomous response such that the new transformed response can be deemed more or less continuous.