Sankhya: The Indian Journal of Statistics

1993, Volume 55, Series B, Pt. 3, 297--309

REGRESSION WEIGHTING FOR MULTIPHASE SAMPLES

By

F. JAY BREIDT

And

WAYNE A. FULLER, Iowa state University, U.S.A.

SUMMARY. Estimation procedures for some three-phase designs are considered. For samples with a large number of control variables relative to the number of sampling units, we suggest estimators constructed in two steps. First, standard regression on data collected in all but the last phase of sampling provides initial estimates. Second, a regression weight generation procedure is applied to data collected in the final phase of sampling. Estimates from the first step are used as control totals. The generated weights lead to simple, internally consistent estimators. The methods are applied to a natural resource inventory conducted by the United States Forest Service in Alaska. The three phases consist of high altitude photography, low altitude photography and ground visits. Jackknife variances estimation is employed to compare the efficiency of the full multiphase estimator to the simplified estimator.

 Subject classification. 62D05

FULL PAPER.

This article in Mathematical Reviews.