The problem is estimating the effort required to complete a software project. The problem is difficult because of the shortage of data within the project, so a promising strategy is to use data from other projects. Work has been done on predicting defects on a given project using data from other projects. The research reported here describes the results from using these outside data approaches, which predict defects, to estimate effort.

Seven approaches are tested on eight datasets. Performance is measured using “standardized accuracy,” that is, 1 - (the ratio of the mean error using the approach divided by the mean error resulting from random guessing). Several of these approaches appear promising, but none are clearly superior. Also, for each dataset, the most effective approach is determined. Again, no approach shows clear superiority.

Future work would include other approaches and ways of modifying the defect prediction algorithms to estimate error. The paper is clear and the analysis is valid; but, clearly, this is work in progress.