Subjective evaluations of individual performances by supervisors are subject to bias. It is important to correct for biases in order to more accurately measure the effects of specific variables on individual performance. This report develops statistical and econometric techniques for correcting biases in models of individual performance using a variant of the classical linear regression model. A multiscale model is proposed to deal with two types of bias: location bias when an individual's performance is systematically overestimated or underestimated, and scale bias when differences among individuals rated are exaggerated or minimized. Several specific multiscale estimating techniques are developed, including equal total variance, equal residual variance, maximum likelihood, and least squares. Finally, the multiscale estimators are applied to the problem of estimating the cost of on-the-job training in the military. The multiscale model can be applied to a wide variety of estimating problems where observations can naturally be categorized into specific subgroups.
Cooper, Richard V.L. and Gary R. Nelson, Analytic Methods for Adjusting Subjective Rating Schemes. Santa Monica, CA: RAND Corporation, 1976. https://www.rand.org/pubs/reports/R1685.html.
Cooper, Richard V.L. and Gary R. Nelson, Analytic Methods for Adjusting Subjective Rating Schemes, Santa Monica, Calif.: RAND Corporation, R-1685-ARPA, 1976. As of October 07, 2021: https://www.rand.org/pubs/reports/R1685.html