Eliminating Over-Confidence in Software Development Effort Estimates

Minimum-maximum effort intervals are applied in the planning of software development projects in order to, among other things, determine the contingency buffer. Several studies suggest that judgment-based minimummaximum intervals are based on a systematic over-confidence in the accuracy of the effort estimates. In this paper, we investigate whether the possession by estimators of information about previous estimation error for similar projects reduces this over-confidence. Nineteen realistically composed estimation teams provided minimum-maximum effort intervals for the same software project. Ten of the teams (Group A) received no instructions about the uncertainty assessment process. The remaining nine teams (Group B) were instructed to begin the minimum-maximum effort interval assessment by recalling the distribution of estimation error for similar projects. We found that the recall of the error distribution of the Group B teams did have an impact, but mainly on the assessment of the estimated minimum effort, not on the maximum effort. We discuss reasons for this finding and provide recommendations.