Pairwise comparisons for parallel profile models with mixed effects (Q2913238)

A method for constructing approximate simultaneous confidence intervals for level differences in a mixed effects nonlinear parallel profile model is suggested and their statistical properties are studied and compared by Monte Carlo simulations.NEWLINENEWLINEThe considered nonlinear mixed effect model is useful for modelling and data analysis which typically arise in pharmacokinetics and/or in general class of growth processes, when the (nonlinear) group mean functions are possibly parallel, however, depend on random effects of the the individual subjects. In such situations, a pairwise comparison of level differences is of particular interest.NEWLINENEWLINEThe suggested approach is based on a novel iterative estimation procedure, which in first step uses the EGLS procedure (estimated generalized least squares) for estimation of the fixed effects parameters (of the nonlinear mean functions) in an approximate model (approximated by the first-order Taylor expansion), then estimates the variance-covariance components, and finally estimates the fixed effect parameters representing the level differences. The author considers and compares two estimators for the level differences. Based on such estimators of the approximate model, Tukey's method is used to construct simultaneous confidence intervals for pairwise comparison.