Revisiting “Investigating Causal Effects of Software and Systems Engineering Effort" Using New Causal Search Algorithms

Repeating the earlier Investigating Causal Effects of Software and Systems Engineering Effort study, but with newer algorithms, Causal Learning (Causal Discovery and Causal Inference) was applied to discover and quantify the causal relationships among variables in the COCOMO and COSYSMO cost estimation models' calibrating datasets.

Results include identifying which project factors versus organizational factors are more likely to affect costs and schedule and thus, relative to the other variables in the dataset, are more likely to be effective targets for intervention to control costs.

As with the earlier study, USC researchers performed the actual research with coaching from SEI researchers, who did not have access to the data.

Instead of bootstrapping and analyzing the resulting distribution of edges, as in the earlier study, the Best Order Score Search (BOSS) algorithm was applied directly (i.e., without bootstrapping). Also, the new Markov Checker was applied to evaluate the resulting graph for consistency with the original dataset. (Note: The Markov Checker has since improved to do additional testing that extends beyond what is reported in this presentation.)