Generalized linear latent and mixed modelling
2 - 3 July 2004, Florence, Italy

Please note that no places are left at the short course


Department of Statistics "G. Parenti"
viale Giovanni Battista Morgagni 59
50134 Florence

Registration will start at 8.30 on Friday 2 July 9.30

Brief Description

Anders Skrondal and Sophia Rabe-Hesketh will present a two-day course on Generalized linear latent and mixed modelling on 2-3 July 2004 before the 19th International Workshop on Statistical Modelling.

Generalised linear mixed models, also known as hierarchical or multilevel models are useful for clustered data where observations on the same cluster cannot be assumed to be independent. Examples include longitudinal data, custer-randomised trials, surveys with cluster-sampling, genetic studies, meta-analysis and many other applications. The models are generalized linear models in which the intercept and some of the regression coefficients are allowed to vary randomly between clusters to represent between-cluster heterogeneity and induce within cluster-dependence. The random intercept and coefficients can be viewed as latent variables. Latent variables are also often used to represent the true value of a variable measured with error, a typical example being dietary intake (continuous latent variable) or diagnosis (categorical latent variable). Measurement models relating the measured variables to the latent variable can be used to investigate the characteristics of measurement instruments or diagnostic tests. These models are called factor or item response models if the latent variables are continuous and latent class models if they are categorical. Generalized linear latent and mixed models unify generalised linear mixed models and measurement models in a single response model. To allow for hierarchical dependence structures where clusters are themselves nested in higher-level clusters, latent variables can vary at different levels. The response model can be combined with a structural model to regress latent variables on other latent or observed variables varying at the same or higher level, giving (multilevel) structural equation models. An important application of structural equation models in biostatistics is regression with covariate measurement error. Structural equation models are also useful for modelling the dependence between different processes, for instance the response of interest in a clinical trial and the (non-ignorable) drop-out process. Maximum likelihood estimation of Generalized linear latent and mixed models (GLLAMMs) is implemented in a Stata program gllamm which also provides empirical Bayes prediction of the latent variables.

Provisional Course Outline

There will be lectures in the morning and computer labs in the afternoon. Approximate course outline:

Day I
  • Generalized linear mixed models
    • Generalized linear and latent response formulations for discrete response models
    • Random intercept models and intraclass correlation
    • Random coefficient models
    • Multilevel models
    • Brief overview of estimation methods
    • Empirical Bayes prediction
  • Latent class models
    • Exploratory latent class models
    • Mixture regression models
    • Nonparametric maximum likelihood
    • Classification
  • Applications in computer lab

Day II

  • Factor and item response models
    • Classical measurement models
    • Congeneric measurement models or unidimensional factor models
    • Multidimensional factor models
    • One and two parameter item response models
    • Latent scoring
  • Structural equation models
    • Multiple indicator multiple cause (MIMIC) models
    • Direct and indirect effects of covariates – item bias or differential item functioning
    • Recursive and non-recursive relationships
    • Multilevel structural equation models
  • Applications in computer lab

Daily Schedule

9.30-11.00 lecture I
11.00-11.30 break
11.30-13.00 lecture II
13.00-14.30 Lunch (not provided)
14.30-16.00 computer LAB
16.00-16.30 break
16.30-17.30 computer LAB

Prior Knowledge

The course assumes some knowledge of generalised linear models. Familiarity with Stata would be beneficial and experience with some command-based statistical package is essential.


  • 80 €

All participants will receive copies of the overheads used, a CD containing the data and programs used in the course and a copy of the book:

Skrondal, A. and Rabe-Hesketh, S. (2004). Generalized latent variable modeling: Multilevel, longitudinal and structural equation models. Boca Raton, FL: Chapman & Hall/ CRC Press.


Registration is possible from January 2004 through the IWSM 2004 registration form.

Back to main