In this project we propose the investigation of important issues related to the nonlinear mixed-effects modeling assuming elliptical errors, including the assumption of heteroscedastic and/or autoregressive structures as a generalization of the scale matrix from the models developed by the applicant Cibele Russo in her Doctorate Thesis and published in Computational Statistics and Data Analysis. The elliptical nonlinear models with mixed effects provide relevant alternatives for the modeling of longitudinal data, since it introduce the intragroup correlation and permit the obtaintion of robust estimates against aberrant observations and few sensitive to perturbations. However, a more sofisticated modeling on the involved scale matrix may provide a significative gain for the models when variability and dependence patterns among the measurements taken in the same experimental unit are observed. Moreover, validation and diagnostic techniques for the proposed models will be investigated, which provide important tools for the choice of models. The results will be applied to real data.
News published in Agência FAPESP Newsletter about the scholarship: