An Analysis of the Impact of Log-Linear Regression on the Estimated Learning Curve Parameters
Abstract
This research had three main objectives. First, was to determine whether bias existed in the estimate of the learning curve parameters, as calculated by popular learning curve programs. Second, was to compare the fitting techniques of ordinary and weighted least-squares, with and without bias removed. Third, was to test the normality and constant variance assumptions of the average unit cost data. When using the least-squares fitting technique to fit production lot data to the unit formulation of learning curve theory, log-linear regression biases the estimate of the first unit cost parameter on the high side, resulting in overestimating program costs. The bias was shown to be almost exclusively a function of the variance in the production cost data. Factors which would reduce the bias were investigated. An approximation to the bias reduction factor proposed by Ilderton was used with excellent results. Keywords: Cost estimates, Learning curves, Least squares method, Curve fitting.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1988
- Accession Number
- ADA201471
Entities
People
- Tom Tracht
Organizations
- Air Force Institute of Technology