|
Abstract:
|
We consider integrating studies that collect the same high-dimensional repeated outcomes with different but overlapping sets of covariates. The primary goal of this integrative analysis is to estimate the effects of the overlapping covariates, while adjusting for study-specific covariates, through a marginal regression model. To estimate the effects of covariates of interest on the high-dimensional outcomes, we develop a divide-and-conquer procedure for statistical estimation and inference of regression parameters, which is implemented in a fully distributed and parallelized computational scheme. To overcome computational and modeling challenges arising from the high-dimensional likelihood of the repeated outcome, we propose to analyze small batches of data using Qu, Lindsay and Li (2000)’s Quadratic Inference Functions, and then to combine these estimators using a one-step meta-estimator in a similar spirit to Hansen (1982)’s Generalized Method of Moments. We show both theoretically and numerically that the proposed method is efficient and computationally fast. We develop an R package for ease of implementation.
|
