• Prof. Armando Teixeira Pinto – Faculdade Medicina Universidade do Porto
  • FCUL (DEIO) – Campo Grande – Bloco C/6 Piso 4 – Sala 6.4.31 – 14:30h
  • Quarta-feira, 30 de Janeiro de 2008

Abstract: Biomedical research often involves the measurement of multiple outcomes in different scales (continuous, binary and ordinal).  The common approach to this type of data is to model each outcome separately ignoring the potential correlation between the responses. We describe and contrast several full likelihood and quasilikelihood multivariate methods for non-commensurate outcomes. We present a new multivariate model to analyze binary and continuous correlated outcomes using a latent variable. We study the efficiency gains of the multivariate methods relative to the univariate approach.
For complete data, all the models give consistent estimates for the parameters and the gains in efficiency by adopting a multivariate approach are negligible in most contexts.  In situations of missing data, the univariate approach may lead to biased results. Extensions of the multivariate approaches are proposed for these situations. Real data examples illustrate the methodology.