| Summary: | Approved for public release, distribution unlimited === The benefit of software cost estimation is universally recognized as one of the cornerstones of effective software project management and control. Despite the advances of computer-based estimation tools, their accuracy remains largely inadequate, and their utility among software development practitioners is limited. Consequently, the optimal estimation of software cost remains an elusive goal of most project managers. Central to this issue is the nature of the data on completed software projects that are incorporated into the organization's database of historical project results. This information forms the basis for both future project estimation and ex-post-facto assessment of estimation models. Actual project results are typically the data of choice for both the calibration and evaluation processes, despite the fact that these raw values disregard project inefficiencies such as initial size underestimation. This thesis challenges the notion that historical project results represent the preferred and most reliable benchmarks for future estimation purposes. Computer- based simulation is used to test a proposed strategy which capitalizes on an organization's learning experiences by neutralizing the cost excess caused by the initial undersizing, and that derives a posterior set of normalized effort and schedule estimation benchmarks. Analysis of the results indicates that normalization of the data leads to significantly improved project productivity, more optimal cost estimates, and provides the organization with increased potential for future cost savings.
|
|---|
