Generalized Linear Mixed Models

Lecturer

Juha Alho

Credits

6 credits (op).

Type

Course 78005. Special course in advanced studies. The course can be extended to 8 op. by doing on empirical assignment or a theoretical study of some special aspect.

Contents

The course provides an introduction to methods used in causal research, when the outcome of interest displays dependencies across experimental units. A typical response might be a binomial “yes/no” or a Poisson distributed count. The dependencies arise when experimental units form groups (such as members of a family, repeated measures from the same individual, or measures from close locations). The dependencies are modeled via random latent factors that are shared by the experimental units. The term “mixed” refers to the presence of both fixed and random effects in the generalized linear model.

Linear mixed models can, in many cases, be analyzed using likelihood methods. In the case of generalized models this becomes tedious as the required marginal models must be evaluated numerically. A large number of approximations exist, but the results rely on asymptotics that may not always be tenable in practical applications. A simpler analysis is often possible via Bayesian formulations.

Computations will be carried out using R functions. New functions are rapidly being developed. In this course nlme ja MCMCglmm are particularly useful.

Goals

After the course a student will understand the strengths and limitations of the generalized linear mixed models as tools of causal research, and knows how to fit some such models using existing R functions.

Pre-requisities and Target Group

It is assumed that a student is familiar with linear regression and generalized linear models before taking the course. Similarly, it is assumed that the student is familiar with principles of maximum likelihood inference, and, preferably, with Bayesian methods. (During Period 1, Linear Mixed Models, 78003, will be lectured – in Finnish.)

The course is suitable for students taking courses in Statistics at the level of advanced studies. It is also appropriate for researchers working on their doctoral dissertation in applied fields, and for researchers working on statistical problems outside the university.

Lectures

Period 2,  Oct. 29 – Dec. 11, 2013.

Lectures are on Tuesdays and Wednesdays at 10-12.  

Location: Exactum B120.


Course requirements

Lectures and exercises (24 h)

For 6 op. final exam and exercises. For 8 op. add a deeper empirical assignment or a theory exam.

References

Galecki A. and Burzykowski T. (2013) Linear mixed-effects models using R: a step-by-step approach. New York: Springer.

Hadfield J.D. (2010) MCMC methods for multi-response generalized linear mixed models: The MCMCglmm R package. Journal of Statistical Software, Vol. 33, issue 2, 1-22.

Ilmoittaudu

Unohditko ilmoittautua? Lähetä meilillä nimesi, opiskelijanumerosi ja mille kurssille haluaisit päästä osoitteeseen johanna.hyytiainen(a)helsinki.fi.

Did you forget to register? Send your name, student number and the course you wish to participate in  to the address johanna.hyytiainen(a)helsinki.fi.