12 Apr module 06: generalized linear models
- module 6: generalized linear models
07/09/2020 - 11/09/2020
1:45 pm - 3:15 pm
Mon 07/09: 1:45 pm - 5 pm | online
Tue 08/09: 1:45 pm - 5 pm | online
Wed 09/09: 1:45 pm - 5 pm | online
Thu 10/09: 1:45 pm - 5 pm | online
Fri 11/09: 1:45 pm - 3:15 pm | online
week 1 | register now
This course elaborates on generalized linear models by extending linear models with continuous outcomes to models with discrete outcomes (binary and count data). These are so called generalized linear models. In addition, different techniques on how to build a good model are discussed.
The course contains 4 topics: (1) overview of generalized linear models, (2) logistic regression, (3) Poisson regression and (4) model building. Each topic consists of a combination of a theoretical session and practical hands-on exercises in R in which participants can try out the theory by practical examples. In the first topic, we repeat the characteristics of linear models, which are a subgroup of generalized linear models. We explain why generalized linear models are needed and go deeper into the characteristics of these models (e.g. exponential family, link function, variance function, Maximum Likelihood). Second, we explain 2 important generalized linear models in detail: logistic regression and Poisson regression. Link functions, estimation, inference and goodness-of-fit of these models are discussed and examples are given. In addition, we discuss variants of these models such as Exact, conditional and Firth’s for logistic regression and Zero-inflated and Hurdle models for Poisson regression. Finally, we talk about good model building procedures including forward, backward and stepwise model building. Methods to compare models are discussed (likelihood ratio tests, AIC) and we review the issues of interactions, confounding and multicollinearity.
You will need a laptop with R (or Rstudio) installed on it. Also install the following R packages before the course starts: catdata, reshape, MASS, pscl, effects and rms.
- hypothesis testing
- linear regression
- confidence intervals
- analysis of variance
- basic knowledge of programming and fitting linear models in R
Dr. Emmanuel Abatih is a post-doctoral research fellow at Ghent University and he works as the FLAMES coordinator for UGent and as a statistical consultant for FIRE. He worked for the Institute of Tropical Medicine (ITM) in Antwerp, as a post doc assistant on topics including: space-time analysis, diagnostic test evaluation, transmission dynamic modeling and risk analysis. He has supervised/co-supervised over 30 masters and 7 PhD students. He has experience with R, python, SATSCAN, SAS and STATA.