Indiana University Bloomington

Statistics S681
Model Comparison And Selection

Contact: Andrew Womack
Offered: Fall, 2015
Class Time: 11:15-12:30 PY 115
Class Days: M, W
Pre-Requisites: STAT 620, STAT 631, STAT 632 or consent of instructor
Algebra Required: Yes
Calculus Required: Yes
Syllabus: No Syllabus Avaliable
Keywords: Model Selection, Penalized Regression, Information Criteria
Description: This course is designed to provide an overview of techniques for the comparison of competing statistical models. Increasing the richness of a statistical model always provides better fit to the observed data, requiring a penalization for model complexity to keep the model from overfitting. This course focuses on two basic means of producing this complexity penalization in generalized linear models: penalized regression and information criteria. From the penalization perspective, the course covers the LASSO, LARS, SCAD, and related thresholding rules for simultaneous selection and estimation. The other focus of the course is information criteria such as AIC, BIC, and Mallows’s Cp, among others. If time permits, advanced topics concerning multiple comparison correction, locally adaptive shrinkage, and analysis when the number of predictors is greater than the number of observations will be covered. The course balances theoretical developments with practical implementation and comparison of methodologies across a variety of simulated and real-data examples
Substantive Orientation: Applied Statistics
Statistical Orientation: Theory and application of model selection criteria to generalized linear models.
Formal Computing Lab: No
Software: R