An Analysis of Uncertainty in Learning Curve Models

The dilemma with learning curves and actual costs is "fitting square pegs into round holes." Alternative regression techniques or further modifications to Y=aX^b may address discrepancies in modeling observed data with learning curves. Adequate data permits further analysis of the constants, variables, and factors that make up a learning curve model. The analysis involves trying a nonlinear solver on the traditional learning curve models (i.e., Wright's cumulative average theory and Crawford's unit learning curve theory) and changing the theoretical first unit cost (T1) to the actual first unit cost (A1). Mergers and acquisitions in the aircraft industry may have reduced certainty in learning curve models since human resources were dispersed to manufacturing facilities worldwide. The results indicated no significant differences between using a nonlinear solver over ordinary least squares. However, the nonlinear solver propagated more uncertainty into the learning curve models. The actual first unit cost (A1) almost rids cumulative average theory models' statistical advantage over unit learning curve theory models. Conversely, mergers and acquisitions reduced uncertainty in learning curve models. Research on program attributes should continue when the appropriate data is available. Cost analysts should request missing data from contractors.