|Title||Fallacies and Biases When Adding Effort Estimates|
|Afilliation||Software Engineering, Software Engineering|
|Publication Type||Proceedings, refereed|
|Year of Publication||2014|
|Conference Name||Proceedings at Euromicro/SEEA|
Software professionals do not always clarify what they mean by their effort estimates. Knowing what is meant by an estimate is, however, essential when adding individual effort estimates from a work breakdown structure to find the estimated total effort. Adding the most likely instead of the mean effort of a set of cost elements may result in substantial underestimation of the total effort. In a survey of forty-four software companies we found only two companies that clarified the meaning of their estimates and had a proper method for adding these estimates. The other companies typically added single point estimates without clarifying what they added or with types of estimates likely to give too low estimates of the total effort. We examine the effect of improper addition of estimates and find, for the studied contexts, that summing the most likely effort estimates would lead to a substantial under-estimation of the most likely total effort. We also find that the use of the PERT-method, which provides a proper statistical basis for adding effort estimates and is used by many software companies, is likely to underestimate the effort in software development contexts. This is caused by, we argue and illustrate with empirical data, people's tendency towards providing too narrow minimum and maximum effort intervals. We outline a method that, we believe, better ensures that proper estimates of the total effort are produced.