Robust Statistical Estimators for Use within Competitive Bid Data

Abstract

Location and scale estimates calculated from project bid data measures the consensus value from among the bidders along with their collective uncertainty about that value. Five robust statistical location and scale estimators are evaluated for their relative performance under competitive bidding conditions-small sample sizes with a high potential for spurious bids. Estimates are made for standard normal samples of three, five, seven, and nine observations using each estimator. Within each sample, one observation is systematically varied in a range of five standard deviations above and below the mean and the other observations are assigned their expected rank-order values. Sensitivity curves are used to illustrate the bias in each estimator caused by the placement of the errant observation. The median and normalized median absolute deviation were found to be the most effective of the five procedures analyzed for location and scale estimators, respectively. Applications of these robust procedures have use in improving legal guidance provided to the courts, enhancing procurement procedures, and providing more accurate bid evaluations. One application describes how robust statistics can be used to reconcile project bids.