Combining Inventory Estimates with Possibly Biased Auxiliary Information

Abstract

Abstract The usual method of combining sample data with auxiliary information is the familiar precision-weighted composite estimator. However, application of this estimator is straightforward only when the auxiliary information is unbiased. If this is not the case, and the bias is unaccounted for, then the risk of the usual composite estimator can be greater than that of the sample mean. We conjecture that investigators are often unsure of the possible bias in their auxiliary information. Accordingly, we develop an estimator which mimics the usual composite estimator when the auxiliary information is unbiased, yet dominates the sample mean even as the bias becomes large. In addition, unlike the usual composite estimator, the new estimator does not require one to specify the variance of the auxiliary information. The performance of both estimators, and the sample mean, is examined based on results of a simulation trial on actual sample data. For. Sci 36(3):693-704.