Accession Number : ADA187680

Title :   Analysis of Learning Curve Fitting Techniques.

Descriptive Note : Master's thesis,

Corporate Author : AIR FORCE INST OF TECH WRIGHT-PATTERSON AFB OH SCHOOL OF SYSTEMS AND LOGISTICS

Personal Author(s) : Avinger, Charles R

PDF Url : ADA187680

Report Date : Sep 1987

Pagination or Media Count : 88

Abstract : This thesis was basic research on learning curve fitting techniques. The unit formulation of the learning curve was fit using two parametric techniques, ordinary least-squares and weighted least-square, and two nonparametric techniques, called median slope and mean slope. Comparisons were made between the techniques in four data cases, equal and unequal lot sizes with normally distributed error terms and equal and unequal lot sizes with triangularly distributed error terms. The Cauchy error term distribution was tried but rejected. The fitting techniques were compared on their estimation of the formula parameters, the dispersion of the data points around the fitted line and their predictions of future costs. Analysis showed that in cases of equal lot sizes, ordinary least-squares has the most bias, but still is a good predictor of future costs. In cases of unequal lot sizes the weighted least-squares and the ordinary least-squares techniques performed well. Ordinary least-squares estimated the parameters with the least bias but had prediction errors that increased with the unit numbers. Weighted least-squares estimated the parameters with the most bias and had prediction errors that, on average, started high and turned low as the unit number got larger. The nonparametric techniques did a better job of predicting future cost, in that the mean predictions were closer to the population costs. However, these techniques had wider ranges of predictions. Because this was basic research, there are many areas for further research. Keywords: Cost estimates, Theses.

Descriptors :   *COST ESTIMATES, *CURVE FITTING, *LEARNING CURVES, *MANAGEMENT PLANNING AND CONTROL, BIAS, CAUCHY PROBLEM, COSTS, DATA BASES, DISTRIBUTION, ERROR ANALYSIS, ERRORS, FITTINGS, FORMULATIONS, LEARNING, LEAST SQUARES METHOD, MEAN, METHODOLOGY, NUMBERS, PARAMETRIC ANALYSIS, POPULATION, PREDICTIONS, SLOPE, WEIGHTING FUNCTIONS, NONPARAMETRIC STATISTICS, THESES

Subject Categories : Economics and Cost Analysis
      Numerical Mathematics

Distribution Statement : APPROVED FOR PUBLIC RELEASE