Logistic regression

Logistic regression models are members of the generalized linear models family. Like the more common member, linear regression, they aim to estimate the relationship between independent (explanatory) variables and one dependent (outcome) variable. The main difference is that, with logistic regression, the outcome variable is binary, meaning it can only present two different states such as dead or alive, yes or no, present or absent. This simple characteristic creates some specific differences in the interpretation of the results and opens many possibilities. Outcome variables can be transformed to become binary in certain situations. For example, if a nonlinear behavior is suspected in regard to a continuous variable, then the transformation into a two-categories variable can offer a better solution. Also, Likert-type scale variables (i.e., completely disagree to completely agree) can be dichotomized into agree versus not agree (taking the neutral level to not agree) or disagree versus not disagree (taking the neutral level to not disagree). Logistic regression is a simple and powerful method to build a classifier to be used in machine learning approaches. This entry reviews the rationale for using logistic regression models and then discussing building the model and interpreting results.