Much more than a prediction: Expert-based software effort estimation as a behavioral act

Traditionally, Software Effort Estimation (SEE) has been portrayed as a technical prediction task, for which we seek accuracy through improved estimation methods and a thorough consideration of effort predictors. In this article, our objective to make explicit the perspective of SEE as a behavioral act, bringing attention to the fact that human biases and noise are relevant components in estimation errors, acknowledging that SEE is more than a prediction task. We employed a thematic analysis of factors affecting expert judgment software estimates to satisfy this objective. We show that estimators do not necessarily behave entirely rationally given the information they have as input for estimation. The reception of estimation requests, the communication of software estimates, and their use also impact the estimation values — something unexpected if estimators were solely focused on SEE as a prediction task. Based on this, we also matched SEE interventions to behavioral ones from Behavioral Economics showing that, although we are already adopting behavioral insights to improve our estimation practices, there are still gaps to build upon. Furthermore, we assessed the strength of evidence for each of our review findings to derive recommendations for practitioners on the SEE interventions they can confidently adopt to improve their estimation processes. Moreover, in assessing the strength of evidence, we adopted the GRADE-CERQual (Confidence in the Evidence from Reviews of Qualitative research) approach. It enabled us to point concrete research paths to strengthen the existing evidence about SEE interventions based on the dimensions of the GRADE-CERQual evaluation scheme.