Categorical Data Analysis: Models for Binary, Ordinal, Nominal, and Count Outcomes

Instructor(s): Scott Long, Sociology, Indiana University

This workshop examines the most important regression models for binary, ordinal, nominal, and count outcomes. Learning how to interpret complex, nonlinear models is the primary objective of this class. The first two days are devoted to understanding fundamental issues of estimation, testing and assessing fit of nonlinear models. Basic concepts and notation are introduced by reviewing the linear regression model. Within this familiar context, the method of maximum likelihood estimation is presented. These ideas are then used to develop the logit and probit models for binary outcomes. A variety of practical methods for interpreting the nonlinear models are presented. Statistical testing and assessing fit are also illustrated with a series of real-world examples. The last three days focus on models for nominal, ordinal, and count outcomes. The ordinal model is presented as a series of binary models that are simultaneously estimated with constraints. The methods of testing and interpretation presented for the binary model are extended to ordinal models. Next, the multinomial logit model is presented. While conceptually this model is a simple extension of binary logit, the large number of comparisons involved make this model difficult to interpret. Graphical methods are introduced to address this difficulty, along with a series of particularly useful statistical tests. The last day deals with models for count outcomes, including Poisson regression, negative binomial regression, and zero modified models. At the minimum, you should have a strong background in linear regression.

Dates: June 9-13

Location: Ann Arbor, MI

Fees:  Member: $1,600;  Nonmember: $3,200

Application is considered incomplete unless accompanied by fee payment. Cancellation less than 21 days prior to the workshop is subject to a $100 late withdrawal fee.

Last Updated: 2008-02-11