40 Years of Parametric Effort Estimation- A Report Card on COCOMO-style Research

Oct1 '14

with Barry Boehm, Ye Yang, Jairus Hihn, from USC, Stevens Institute, JPL

The longevity of parametric effort estimation is remarkable. Decades after their invention, these methods are still both widely used and widely useful.

This research reviews the standard criticisms of this approach. We find that, contrary to common criticisms, (1) parametric estimation has not been superseded by more recent estimation methods; (2) it is not true that parametric estimation is no better than simplistic lines of code counts; (3) the old parametric calibration data is still relevant to more recent projects; (4) parametric estimation need not be expensive to deploy at some new site since these these methods can be tuned on very small sample sizes (in our experiments, a mere eight projects is enough); and (5) compared to other methods, parametric estimation is not unduly sensitive to errors in the size estimates.

Hence we conclude that, in 2015, is still valid and recommended practice to try parametric estimation before exploring other, more innovative methods. Also, it can be useful to augment parametric estimation with (a) some local calibration and (b) some column pruning (examples of those techniques are discussed in this research).