Advanced Topics in Maximum Likelihood Estimation
- David Darmofal, University of South Carolina
- Christopher Zorn, Pennsylvania State University
This is the second course offered in models that rely on maximum likelihood as a primary method of estimation. This course will be devoted to the analysis of longitudinal data: that is, data that vary both across units and over time.
The first section of the course covers methods for making inferences with repeated observations data, focusing mostly on the theory and estimation of models for panel and time-series cross-section data. Topics covered include fixed effects, random effects, dynamic panel models, random coefficient models, models for spatial dependence, models for qualitative dependent variables, and panel attrition.
The second section of the course will cover methods and models for survival data. Survival data record the length of time until some event occurs, for example, the termination of a cabinet government or the time until an unemployment spell ends. Because time-to-event occurrence is an important feature of these kinds of data, methods suitable to duration data are sometimes called "event history analysis." This portion of the course will consider a wide variety of survival methods, including some non-traditional models for categorical data.
Prerequisites: A course in maximum likelihood estimation and/or generalized linear models (such as probit, logit, and event counts) at the level of Long's Regression Models for Categorical and Limited Dependent Variables or Faraway's Extending the Linear Model with R is a prerequisite for this course. Course participants should also be familiar with the multiple regression model in matrix form.
Fees: Consult the fee structure.
Tags: probit, logit, logistic regression, duration models, hazard models, panel data, generalized linear models, spatial dependence, qualitative dependent variable, event history analysis, survival analysis