Cum Av or Unit? Is the Choice Between Cumulative Average and Unit Learning Curve Theory a Fair Fight?
Cullis, B.¹; Coleman, R.²; Braxton, P.^1
1Northrop Grumman - TASC; ²Northrop Grumman

The statistics of the regressions used to judge significance and "best fit" are inherently biased towards Cum Average because the Cum Average metric is smoothed by the very fact that it is cumulative. This paper will use generated data with and without error terms in an attempt to show the degree of this bias and will seek to generate some thumb rules or some insight to help the analyst decide between the two models.

It has long puzzled the authors as to whether both learning models can be true, and why most practitioners in some commodities, and some practitioners in most commodities, lean more toward the one than the other. It would be a grand project indeed to decide whether the one model or the other is best either globally or locally. The scope of that endeavor is, however, too great. In lieu of that project, we hope that by shedding light on this puzzling problem we can enable all practitioners to decide whether their preference is rational, and perhaps thereby bring some rigor to this problem.