Short course on mixed models / multilevel modelsshop

Short Course on

Hierarchical Data Analysis

EHPS, Maastricht, the Netherlands, August 15, 2007

Geert Molenberghs (Hasselt University, Diepenbeek, Belgium)

Short course description

Hierarchical, or multi-level data, arise in a variety of psychometric, social-sciences, and biomedical contexts. Based on the Belgian Health Interview Survey, an application will be presented with a design that involves systematic sampling, stratification, weighting, and in particular clustering induced by three-stage sampling.

Without neglecting the other design aspects, the focus will be on the clustered, i.e., three-level nature of the data. Three modes of analysis will be considered: (1) naïve, which implies that the design is not taken into account; (2) design-based, following traditional survey sampling methodology; (3) model based, involving hierarchical models (mixed-effects models, multi-level models).

Two key illustrative outcomes will be considered, the (logarithm of) a respondent’s body mass index (LNBMI) and whether or not a respondent has a dedicated general practitioner (SGP). This choice enables us to focus on the methodology for continuous and binary outcomes. In the continuous case, the linear mixed model will be given a prominent place. In the binary case, emphasis will be on generalized estimating equations and generalized linear mixed models.

The methodology will be illustrated using the SAS procedures (1) MEANS, REG, and LOGISTIC, (2) SURVEYMEANS, SURVEYREG, and SURVEYLOGISTIC, (3) MIXED, GENMOD, GLIMMIX, and NLMIXED. This apparent focus on SAS notwithstanding, the emphasis will be more on general principles and on modeling concepts, not so much on mathematical detail, nor on software Furthermore, a perspective will be given on the use of other packages, including STATA, SPSS, SUDAAN, and MLwiN.

Method of instruction:  The learning event will take the form of a half day short course, planned for the morning of August 15, 2007.

The course will be explanatory rather than mathematically rigorous. Emphasis is on giving sufficient detail in order for participants to have a general overview of frequently used approaches, with their advantages and disadvantages, while giving reference to other sources where more detailed information is available. Regarding software, general concepts and interpretation are more important than the precise implementation detail.

Language: English

Prerequisites: Throughout the course, it will be assumed that the participants are familiar with basic statistical modeling, including linear models (regression and analysis of variance), as well as notions of logistic regression.

Students may want to consult the two textbooks below. However, this is certainly not a formal requirement.

Literature:

·     Copies of the transparencies, used in the course, will be provided.

·     Verbeke, G. and Molenberghs, G. (2000) Linear Mixed Models for Longitudinal Data. New York: Springer-Verlag.

·     Molenberghs, G. and Verbeke, G. (2005) Models for Discrete Longitudinal Data. New York: Springer-Verlag.

For more information regarding the course, please contact:

Professor Geert Molenberghs

Center for Statistics, Director

Universiteit Hasselt

Agoralaan 1

B-3590 Diepenbeek, Belgium

geert.molenberghs@uhasselt.be