Abstract
Previously published work on R&D time-phasing methods primarily addresses the suitability of various functional forms, such as Rayleigh and Weibull curves, to fit individual historical program profiles. Little guidance exists on how to select values of the Rayleigh or Weibull parameters for a program cost estimate or on how to measure accuracy of the resulting profile. In this study we present four quality metrics that model developers can use to evaluate budget-phasing methods. With metrics in hand, we demonstrate two ways to improve model accuracy. First, independent variables such as percent nonrecurring and number of development units cause a phasing profile to be more or less front-loaded and should be taken into account when developing an R&D budget. Second, to further improve predictive accuracy, we demonstrate some advantages of replacing a previously published approach of curve fitting large numbers of individual program profiles by an approach based on single-stage multivariate regression. Finally, a case study on military and intelligence satellite acquisition programs leads to new parametric schedule-estimating and time-phasing models.
