Combining Independent Regression Estimators From Multiple Surveys

Abstract

Efficient generalized regression (GREG) procedures are proposed for the combination of comparable information collected independently from multiple surveys of the same population. In particular, for combination through alignment of estimated totals of common target characteristics, an efficiency improvement of Zieschang's original composite GREG method is developed, involving a correction of the GREG estimation for different effective sample sizes in a multiple-sample setting. The proposed method is nearly as efficient as the alternative method of Renssen and Nieuwenbroek for general sampling designs, and considerably more practical. Under broad sampling design conditions, the proposed method produces composite estimators that are equally efficient as the estimators of Renssen and Nieuwenbroek for common characteristics and more efficient for noncommon characteristics. Under some of these sampling design conditions, an equivalence of extended GREG estimation and extended optimal regression estimation is established.

KEY WORDS :

• Aligned estimates
• Composite estimation
• Generalized regression estimator
• Optimal regression estimator
• Weight calibration