Sophia Rabe-Hesketh (University of California, Berkeley, and University of London)

Vendredi 10 octobre 2008 à 11h15, salle 5220:

Structural equation models (SEMs) combine measurement models for latent variables with regression models among latent and observed variables. When the units of observation are nested in clusters, for example students nested in schools, SEM is often extended to multilevel SEM by including cluster-level latent variables, in addition to unit-level latent variables. The generalized linear latent and mixed modeling (GLLAMM) framework accommodates a wide range of response types, an arbitrary number of hierarchical levels, measurement models at different levels, cross-level regressions among latent variables, survey weights, and missing data. Some of these features will be illustrated using data from the Program for International Student Assessment (PISA). The gllamm framework will also be compared with the traditional approach to multilevel SEMs which consists of specifying separate models for the within cluster and between cluster covariance matrices.

Slides (567 Ko, )