Regression Models for Categorical Outcomes: Specification, Estimation, and Interpretation (Amherst, MA)


This workshop deals with the most important regression models for binary, ordinal, nominal, and count outcomes. While advances in software make it simple to estimate these models, the effective interpretation of these nonlinear models is a vexingly difficult art that requires time, practice, and a firm grounding in the goals of your analysis and the characteristics of your model. The workshop begins by discussing the general objectives for interpreting results from any regression model and considers why these objectives are more difficult in nonlinear models. Concepts of estimation, testing, and identification are introduced in a quick review of the linear regression model. These ideas are used to develop the binary logit and probit models. Advanced methods of interpretation are introduced using Stata's margins command, with detailed examples on how to compute and interpret average marginal effects, the distribution of effects, and related methods. Concepts from the binary model are used to develop the multinomial logit model for nominal outcomes, followed by the development of several models for ordinal outcomes. Finally, the workshop will cover models for count data, including Poisson regression, negative binomial regression, and zero modified models.

Prerequisites: Participants must be thoroughly familiar with linear regression. While familiarity with Stata is recommended, the labs provide step-by-step instructions for those new to Stata.

Participants should read the detailed information on the workshop webpage.

If you have questions, feel free to contact the instructor at jslong at indiana dot edu.

Fee: Members = $1600; Non-members = $3000

Tags: probit, logit, logistic regression

Course Sections