This volume helps readers understand the intuitive logic behind logistic regression through nontechnical language and simple examples. The Second Edition presents results from several statistical packages to help interpret the meaning of logistic regression coefficients, presents more detail on variations in logistic regression for multicategory outcomes, and describes some potential problems in interpreting logistic regression coefficients. A companion website includes the three data sets and Stata, SPSS, and R commands needed to reproduce all the tables and figures in the book. Finally, the Appendix reviews the meaning of logarithms, and helps readers understand the use of logarithms in logistic regression as well as in other types of models.
Logistic regression deals with the ceiling and floor problems in modeling a binary dependent variable by transforming probabilities of an outcome into logits. Although probabilities vary between 0 and 1, logits or the logged odds of the probabilities have no such limits—they vary from negative to positive infinity. Many other transformations also eliminate the ceiling and floor of probabilities. A number of functions define S-shaped curves that differ in how rapidly or slowly the tails approach 0 and 1. The logit transformation in logistic regression has the advantage of relative simplicity and is used most commonly. One other transformation based on the normal curve that appears in the published literature is worth reviewing.
Another Way to Linearize the Nonlinear
Probit analysis transforms probabilities of ...