Poster - Causal Models for Software Cost Prediction and Control
November 2020 • Poster
Michael D. Konrad, William Nichols, Robert W. Stoddard, David Zubrow
This poster describes CMU SEI's collaboration with other researchers to apply causal learning to learn how to control costs in software development and sustainment.
Software Engineering Institute
Estimating and controlling software project costs would benefit from causal knowledge of what drives program costs; however, much of software engineering research reports results from multiple regression, which focuses on correlation rather than causation. For three years (FY18-20), this CMU SEI project investigated the application of causal discovery as an alternative approach to conducting software engineering research, resulting in the identification of direct causes (instead of spurious causes) of program costs, schedule, and quality. This knowledge supports (1) program investments in the enablers of program success and reduction in software vulnerabilities; and (2) corrective and preventive actions on those factors more likely to place a program on track to achieve its objectives.