In the KDSI data (used for COCOMO), Boehm's conversion formula for modified software was used [5].
In addition, a regression was run on the primary input to the COCOMO models, delivered source instructions, or KDSI. The result was an R.sup.2 of 49.4 percent, which is not as good a fit as any of the COCOMO models.
The results were the following: Actual man-months = 54.1 + 0.084 (SLIM) (10.17) R.sup.2 / 92.8% Actual man-months = 11.4 + 0.195 (COCOMO-Basic) (10.82) 93.6% Actual man-months = 51 + 0.17 (COCOMO-Intermediate) (12.24) 94.9% Actual man-months = 40.3 + 0.181 (COCOMO-Detailed) (11.62) 94.4% Actual man-months = -203 + 0.451 (Function Points) (4.63) 71.8% Actual man-months = -175 + 0.426 (Function Count) (4.09) 66.2% Actual man-months = -94 + 1.67 (KDSI) (2.90) 44.5% Actual man-months = -102 + 1.68 (KSLOC) (2.83) 43.2%
Only KDSI and KSLOC did slightly worse, and these are not inputs to ESTIMACS.