19 Logistic Regression: Basic Foundations and New Directions

Research Methods in Psychology


  1. Alfred DeMaris PhD

Published Online: 26 SEP 2012

DOI: 10.1002/9781118133880.hop202019

Handbook of Psychology, Second Edition

Handbook of Psychology, Second Edition

How to Cite

DeMaris, A. 2012. Logistic Regression: Basic Foundations and New Directions. Handbook of Psychology, Second Edition. 2:IV:19.

Author Information

  1. Bowling Green State University, Department of Sociology, Bowling Green, Ohio, USA

Publication History

  1. Published Online: 26 SEP 2012


Logistic regression is a regression-modeling technique that is optimum when the response variable is categorical. Binary logistic regression applies when the criterion is dichotomous. The probability of an event is modeled as a nonlinear function of the regressor set, although the equation is easily linearized via the logit link function. Exponentiated coefficients, known as odds ratios, are interpreted as multiplicative analogs of the slope coefficients in linear regression. Empirical consistency can be assessed via a formal test, and analogs of the OLS R2 are available to assess a model's discriminatory power. Multinomial logistic regression is an extension applicable when a qualitative response has more than two categories. On the other hand, ordered logit models are useful when multinomial responses consist of ordered values. Recent applications of logistic regression include propensity-score matching and fixed-effects logistic regression models. A propensity score is the estimated probability of receiving the treatment, given a subject's measured characteristics. Propensity scores are easily estimated using logistic regression. Treated and untreated subjects with comparable propensity scores can be compared as though they were randomly assigned to treatment groups, provided no unmeasured heterogeneity exists. With two or more measurement occasions per subject, however, fixed-effects logistic regression models can eliminate the threats to causal inference posed by unmeasured heterogeneity. The chapter concludes with suggested further readings.


  • logistic regression;
  • proportional odds models;
  • propensity scores;
  • fixed effects