Population-level information for improving quantile regression efficiency

Abstract

Observational studies often rely on sample survey data for estimation, given the difficulty of obtaining exhaustive information for the entire population. However, the use of sample data can lead to a reduction in estimation efficiency due to sampling error. When certain population-level data are accessible, devising an effective strategy to integrate them into the underlying estimation process proves advantageous. This paper proposes a methodology based on empirical likelihood for conducting quantile regression analysis on longitudinal data while incorporating population-level information. Both theoretical analysis and numerical simulations demonstrate that the proposed approach outperforms estimation methods that do not leverage population-level data.