How accurately can we predict the costs of software development? This study evaluates the accuracy of four popular algorithmic models used for software cost estimation (SLIM, COCOMO, Function Points, and ESTIMACS). Using data from 15 large business data-processing projects, the researchers tested the models' ability to estimate effort ex post, providing a valuable empirical validation. One notable result is that Albrecht's Function Points effort estimation model showed good accuracy when validated against the independent data set, marking a significant finding for the model. The other models, which were not developed in business data-processing environments, demonstrated a need for calibration. Ultimately, the models tested did not sufficiently reflect the underlying factors affecting productivity. This suggests the need for further research to better understand and model the software-development process to enhance cost-estimating tools, thus, the study highlights the importance of ongoing research to enhance understanding in this area.
Published in Communications of the ACM, this paper addresses a critical issue in software engineering. By empirically validating software cost estimation models, it provides valuable insights for both researchers and practitioners in the field, enhancing the understanding of software development processes.