Empirical likelihood confidence intervals for linear regression coefficients

Abstract

Nonparametric versions of Wilks′ theorem are proved for empirical likelihood estimators of slope and mean parameters for a simple linear regression model. They enable us to construct empirical likelihood confidence intervals for these parameters. The coverage errors of these confidence intervals are of order n-1 and can be reduced to order n-2 by Bartlett correction. © 1994 Academic Press Inc.