Kjetil Moløkken

Eliminating Over-Confidence in Software Development Effort Estimates

Previous studies show that software development projects strongly underestimate the uncertainty of their effort estimates. This overconfidence in estimation accuracy may lead to poor project planning and execution. In this paper, we investigate whether the use of estimation error information from previous projects improves the realism of uncertainty assessments. As far as we know, there have been no empirical software studies on this topic before. 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 how to complete the uncertainty assessment process. The remaining nine teams (Group B) were instructed to apply a history-based uncertainty assessment process. The main results is that software professionals seem to willing to consider the error of previous effort estimates as relevant information when assessing the minimum effort of a new project, but not so much when assessing the maximum effort!

Does Use of Development Model Affect Estimation Accuracy and Bias

Objective. To investigate how the use of incremental and evolutionary development models affects the accuracy and bias of effort and schedule estimates of software projects. Rationale. Advocates of incremental and evolutionary development models often claim that use of these models results in improved estimation accuracy. Design of study. We conducted an in-depth survey, where information was collected through structured interviews with 22 software project managers in 10 different companies. We collected and analyzed information about estimation approach, effort estimation accuracy and bias, schedule estimation accuracy and bias, completeness of delivered functionality and other estimation related information. Results. We found no impact from the development model on the estimation approach. However, we found that incremental and evolutionary projects were less prone to effort overruns. The degree of delivered functionality and schedule estimation accuracy, on the other hand, were seemingly independent of development model. Conclusion. The use of incremental and evolutionary development models may reduce the chance of effort overruns.