Journal of Official Statistics, Vol.19, No.1, 2003. pp. 17–30
Confidence Interval Coverage Properties for Regression Estimators in Uni-Phase and Two-Phase Sampling
J.N.K. Rao, W. Jocelyn, and M.A. Hidiroglou
Abstract:Confidence intervals based on the normal approximation are widely used in the design-based approach. Hansen, Madow and Tepping (1983) noted that design-based intervals are inferentially satisfactory despite failures of assumed models. Dorfman (1994) studied confidence interval coverage associated with the sample linear regression estimator under two-phase random sampling using standard and new variance estimators, and concluded that the contention of Hansen et al. is not tenable. In this article we provide reasons for the poor performance under model failures and practical solutions to improve the coverage probability.
Copyright © Statistics Sweden 1996-2018. Open AccessISSN 0282-423XCreated and Maintained by OKS Group