Inverse regression for longitudinal data (Q2249841)

The dimension reduction methods are central in statistical modelling, but their application faces difficulties in cases where the observed data are collected over discrete, possibly random time points. Such observed data are often termed ``longitudinal'' to be distinguished from the data that are densely observed over a period of time and are termed ``functional''. This particular paper deals with such longitudinal data and develops a sliced inverse regression (SIR) dimension reduction method for intermittently and sparsely measured longitudinal covariates. Asymptotic theory is also developed for the new procedure, and it is shown that, under some regularity conditions, the estimated directions attain the optimal rate of convergence. Simulation studies and data analyses, whereby the estimates from dense and sparse data are compared, are finally used to demonstrate the performance of the proposed method. Despite the fact that the proposed approach has been developed to address the difficulties imposed by sparse longitudinal data, it is also capable to handle dense data, both theoretically and practically.