ABSTRACT
Dichotomous outcome variables are common in substance use research (e.g., smoker vs. nonsmoker). This chapter describes the use of logistic regression analysis to model such outcomes. A latent variable approach is used to describe the logistic regression model. An observed dichotomous outcome variable can be thought of as having a continuous variable underlying it (Long 1997). The continuous latent variable is linearly related to the predictors, and is linked to the observed dichotomous outcome by a single cutpoint or threshold value for which larger values have a value of 1 on the observed dichotomous outcome and smaller values have a an observed value of 0. The purpose of logistic regression is to find a linear combination of predictors that maximizes the likelihood of obtaining the observed probability of the sample outcome. Logistic regression is illustrated using an example from a study investigating prospective predictors of successful smoking cessation in young adults.
