Longitudinal data with binary and ordinal outcomes frequently arise in social and health sciences, e.g. due to stratified sampling and repeated measurements. Often, random effects are included to account for within-subject correlation, and valid inference relies on untestable model assumptions. It is important to gain knowledge about the robustness of estimation to violations of the underlying assumptions and develop techniques to detect such misspecifications.
Furthermore, statistical inference for these model classes are typically based on maximum likelihood procedures that are computationally burdensome as they involve numerical approximations of intractable integrals.
Simulations studies are used to investigate robustness of random effects models against model misspecifications and to investigate the relative performance of standard maximum likelihood and composite likelihood methods for correlated binary and ordinal data.
Keywords: Longitudinal data analysis; Random effects models
Biography: Nina Breinegaard is a PhD student at the Department of Biostatistics, University of Copenhagen.