3-p sampling and design versus model-based estimates

Abstract

The unadjusted and adjusted 3-p estimators suggested by L. R. Grosenbaugh (Soc. Am. For. Proc. 1963: 36–42. 1964) are examined as examples of design versus model-based estimators. This paper shows that the 3-p adjusted estimator has near optimal properties under a regression superpopulation model and is a member of the class of generalized regression estimators having desirable robustness properties. The variance estimators suggested by L. R. Grosenbaugh (1964, and For. Sci. 22(2): 173–176. 1976) also have bias-robust properties. The understanding that the model-based view lends will enable practitioners to consider alternative estimators and to have greater confidence in 3-p sampling as a practical tool.