Nonparametric curve estimation with time series errors (Q811056)

A semiparametric model is considered for analyzing repeated measurement data. The author proposes a nonparametric approach for the estimation of the mean function, so that the model has the flexibility to fit a wide variety of data sets. To model serially correlated data which arise quite often in growth curve analysis, the error process is assumed to be part of a finite order moving average process. Under appropriate conditions on the smoothness of the underlying mean function, on the smoothing parameter and on the error process, it is shown that a sequence of local average estimators is pointwise and uniformly consistent in probability. It turns out that the rates of convergence are the same as in the case of independent errors. Moreover, for autoregressive time series error processes with a finite number of parameters, estimators of these parameters are considered. These estimators are based on the estimated residuals and may be useful for examining correlated structures in repeated measurement analyses.