SMABS 2004 Jena University
SMABS 2004 Home Organization About Jena Sponsors Links Imprint / Contact SMABS Home

European Association of Methodology

Department of methodology and evaluation research

Jena University

Contributions: Abstract

Composite links have proven useful in generalized linear models, for instance for handling missing data, for modeling censored and ordinal responses and for linear odds models (Thompson and Baker, 1981; Rindskopf, 1992). When the linear predictor includes continuous or discrete latent variables, the use of a composite link allows simple specification of complex models.
In cumulative models for clustered ordinal responses (possibly with right-censoring), we can specify latent class-specific thresholds, thresholds depending on continuous multilevel random effects or on arbitrary functions of covariates and response categories. Zero-inflated Poisson models can be generalized to multilevel settings by including random effects in either component of the mixture model. These and other latent variable models with composite links can be estimated by maximum likelihood using adaptive quadrature in gllamm.