Improving Software Cost Estimating Techniques in Defense Programs

Abstract

As software becomes more ubiquitous in defense programs, there is a need to improve the accuracy and reliability of methods for estimating software size and cost. Historically, practitioners have used defined distributions in their estimating software to simulate likely outcomes. This research identifies new distributions of likely software costs and effective sizes through an analysis of Cost and Software Data Reports (CSDRs) as well as demonstrating the most appropriate distribution given certain program characteristics known at the genesis of the project. By utilizing various descriptive statistics and statistical tests, this research shows there are distributions that are more closely tailored to the actual qualities of a software program. In some instances, a broad and general distribution is sufficient; however, there are specific commodities, contractors, and system types that are distinctly different and require additional analysis. Overall, this research intends to equip practitioners with an arsenal of distributions and statistical information that will lead them to apply the best model to predict software size and cost, all with the goal of improving overall accuracy.