On nonlinear random effects models for repeated measurements