Evaluating the bias of alternative cost progress models: tests using aerospace industry acquisition programs

Abstract

This study evaluates the quality of cost estimates produced by each of four cost progress models--a random walk model, the traditional learning curve model, a production rate model (fixed-variable model), and a model incorporating both learning curve and production rate effects (Bemis production rate adjustment model). Emphasis is on assessing the level of bias associated with these models and determining the influence of various factors on model performance. Findings indicate, on average, the learning curve and Bemis models underestimate unit costs, while the random walk and fixed-variable models overestimate unit costs. Different factors are evaluated to determine their significance in explaining variations in the bias of their significance in explaining variations in the bias of unit cost predictions and relationships between the significant variables and model cost prediction bias are described Findings indicate the Bemis model is superior to the other cost progress models because it exhibits the least bias and is not significantly influenced (in terms of bias) by variations in the factors considered.