Generalized Linear Models


Timing
4 hours weekly (3+1) during Winter Terms.

Lecturer
Herwig Friedl .

Teaching Aims
To give more emphasis on statistical modelling. Starting from the linear regression framework, results and techniques in the field of generalized linear models will be developed. To improve the student's ability to apply the theory in exploratory data analysis and further in statistical modelling. To introduce statistical software packages like R.

Learning Objectives
Students should be able to analyse real data problems by their own. They should be also familiar with the concepts of exploratory data analysis and should find and verify relationships in the data by applying the ideas of statistical modelling.

Contents
(0) Linear Regression Models. (1) Box-Cox Transformation Family. (2) Exponential Family: Members, Maximum-Likelihood Estimates (Score, Information), Quasi-Likelihood Technique. (3) Generalized Linear Models: Parameterization, Maximum-Likelihood Estimation, Pearsons' Chi-Square Statistic, Deviance and Quasi-Deviance, Modelling Over- (Under-) Dispersion. (4) Logistic Regression: Binomial Models, Logit-Models, Log Odds Ratio. (5) Loglinear Models: Poisson Models, Multinomial Responses, Contingency Tables. (6) Random Effect Models: EM (Expectation/Maximization) Algorithm, Mixture Models, Gauss-Hermite Quadrature (Approximation), Nonparametric Maximum-Likelihood Estimation, Standard Errors.

Pre-requisites
Probability, Mathematical Statistics, Regression Analysis.

Literature

Download Homeworks:

Download Datasets:

Download R Macros and R Statements useful to analyze the data:


This page last modified May 8th, 2017 by Herwig Friedl (hfriedl@tugraz.at).