Confirmatory and Exploratory Multivariate Modelling

1^st week: Regression Models for Categorical Dependent Variables (Logit, Probit, and Related Techniques)

February 25 – February 29, 2008

Andreas Diekmann and Ben Jann (ETH Zürich)

The course teaches statistical methods for the analysis of categorical dependent variables such as logistic regression and related techniques.

Upon completion of this course, the student should have acquired:
(1) Knowledge on the foundations of several methods for the analysis of categorical dependent data, along with the conditions under which their use is appropriate
(2) Skill in the estimation, specification and diagnostics of the models
(3) Hands-on experience with those methods through the use of appropriate software and actual data sets in the PC lab

The course will begin with a short primer on multiple linear regression, in which a continuous dependent variable is “explained” by two or more independent variables, and discuss the limits the application of linear regression to a dichotomous dependent variable, i.e. the Linear Probability Model (LPM). The course will then in depth cover more appropriate models for binary dependent data (e.g. labor market participation, owning a car, getting divorced, successfully selling an item on eBay, surviving a disease, going to vote, etc.) such as logistic regression or the Probit model and address topics such as model derivation from utility theory, estimation of parameters through maximum-likelihood, statistical inference and goodness-of-fit, interpretation of coefficients, and model diagnostics. Finally, a selection of related techniques for the analysis of categorical data (e.g. the multinomial logit for the analysis of traffic mode choice or the ordered logit for socio-economic status) and some advanced models (e.g. models for panel data or multilevel models) will be introduced.

Literature:

Long, J. Scott (1997). Regression Models for Categorical and Limited Dependent Variables. Thousand Oaks, CA: Sage.

under way...