Modeling the Demand for Spare Parts
Estimating the Variance-to-Mean Ratio and Other Issues
Mathematical models are commonly used to study the performance of the Air Force's spare parts supply and repair systems. But accurate evaluations of supply policies are not possible without accurate models of the demand process for spare parts, and models that understate the variability in the demand process will bias evaluations in favor of policies that assume accurate predictions of part failures. This Note examines the model for part failures used in The RAND Corporation Supply System model, Dyna-METRIC. The ability to predict levels of parts failures is strongly affected by at least two types of uncertainty: about the number of failures that will occur assuming the model is correct, and about the adequacy of the model as an approximation of the demand process for spare parts. The author suggests that a model that allows more variability, such as a negative binomial model, would be more appropriate for dealing with the first type of uncertainty, and the second type can be accommodated, in part, by using models with more parameters.