ABSTRACT
The classical linear mixed-effects models of Chapter 2 involve four components: the population-mean terms describing the average time trends and covariate effects; the subject-specific terms describing the individual deviations from the population-mean terms; the distributions of the subject-specific parameters; and the overall measurement errors. These classical models depend on the crucial assumption that both the population-mean and subject-specific terms follow the linear model structure. In practice, the parametric forms of the population mean and subject-specific terms are often unknown, and a misspecified parametric family may lead to erroneous conclusions. We present in this chapter a class of nonparametric mixed-effects models to describe the following aspects from the data: (a) the differences in population time-trends among different subgroups and covariates, (b) the population and individual covariate effects on the outcomes, (c) the population and individual changes over time, and (d) the percentile curves of the outcomes. By extending the population-mean and subject-specific terms to nonparametric functions of time, the models of this chapter are more flexible than the classical linear mixed-effects models.
