Learning Curve and Rate Adjustment Models: An Investigation of Bias

Abstract

Learning curve models have gained widespread acceptance as a technique for analyzing and forecasting the cost of items produced from a repetitive process. Considerable research has investigated augmenting the traditional learning curve model with the addition of a production rate variable, creating a rate adjustment model. This study compares the forecasting bias of the learning curve and rate adjustment models. A simulation methodology is used to vary conditions along seven dimensions. The magnitude and direction of errors in estimating future cost are analyzed and compared under the various simulated conditions, using ANOVA. Overall results indicate that the rate adjustment model is generally unbiased. If the cost item being forecast contains any element that is not subject to learning then the traditional learning curve model is consistently biased toward underestimation of future cost. Conditions when the bias is strongest are identified.1