|Authors||T. Halkjelsvik and M. Jørgensen|
|Title||When 2 + 2 should be 5: The work effort summation fallacy in judgment-based estimation|
|Project(s)||Department of IT Management|
|Publication Type||Journal Article|
|Year of Publication||2019|
|Journal||submitted to a journal|
Estimates of work effort (e.g., work hours, man-months) for project tasks are often based on expert judgments, which in turn are used for calculating the total work effort, total costs, and duration of a project. Because actual work effort outcomes are realizations of probabilistic distributions (e.g., the work has a 20% chance of requiring more than 80 work hours), an estimate does not have an inherent meaning but can represent any value of the assumed outcome distribution, such as the most likely (mode) value. Only estimates reflecting the mean (i.e., expected value) can be added together into estimates of total work effort without changing the meaning of the estimate (the aggregate estimate is still an estimate of the mean). In Studies 1 and 2, we find that software professionals and companies provide estimates that vary substantially in meaning. This makes the aggregation of these estimates difficult. In Studies 3 and 4, we observe that a substantial proportion of software professionals either naïvely sum non-mean estimates to derive the total work effort, or provide total estimates that are otherwise incompatible with the probabilistic meaning of their estimates of individual tasks. In Study 5, we show that improper aggregation strategies can produce over- and underestimation and prediction intervals that are far too wide. The lack of a common understanding of an estimate, coupled with the tendency to aggregate estimates by naïve summation, is likely to produce biased cost estimates at the level of projects or project portfolios.