|Title||The Ignorance of Confidence Levels in Minimum-Maximum Software Development Effort Intervals|
|Afilliation||Software Engineering, Software Engineering|
|Publication Type||Proceedings, refereed|
|Year of Publication||2014|
|Conference Name||ICSIE, Dubai|
|Volume||Lecture Notes on Software Engineering, 2(4)|
Software professionals are frequently asked to provide minimum-maximum effort intervals for a given confidence level. They may for example be asked to provide minimum-maximum intervals where it is 90% likely to include the actual use of effort. If their response is the interval from 800 to 1200 work-hours this should correspond to that it is 90% likely that the actual effort will be more than 800 and less than 1200 work-hours. This effort interval information is, amongst others, used in the planning and budgeting of software projects. In this paper we show that software professionals tend to ignore the confidence levels connected with the minimum-maximum effort intervals. As a consequence, the meaning of minimum-maximum effort interval is unclear and the use of such intervals questionable. The experiment used to document the ignorance of the confidence level is based on requesting one group of software developers to be 98% confident, and another group to be 80% confident when providing their effort intervals. In spite of a difference in confidence levels that should generate quite difference effort intervals, the actual intervals were almost the same. This finding challenge commonly recommended effort uncertainty assessment practices, e.g., those implemented in the PERT method, which are based on the assumption that software professionals are able to provide minimum-maximum effort that reflect the stated confidence levels. The finding does also challenge the explanation typically given for too narrow confidence intervals, i.e., that people are over-confident. Instead, we propose that a more likely explanation is that people ignore the confidence level when setting the minimum and maximum values.