Fitting Learning Curves with Orthogonal Polynomials
Technical rept. Apr-Oct 1986
ARMY RESEARCH INST FOR THE BEHAVIORAL AND SOCIAL SCIENCES ALEXANDRIA VA
Pagination or Media Count:
A FORTRAN program that generates orthogonal polynomials is listed. These polynomials can be used to analyze learning curves into components that can then be tested separately for significance planned comparisons. The program is appropriate to situations in which the data are obtained from equally spaced intervals, as is naturally the case when the independent variable is trials, and also to situations involving unequal intervals e.g., when data are available from the first, second, fourth, and eighth trials only. A demonstration of the component analysis is provided for both the equal interval and unequal interval situation. These analyses result in best-fit curves that reflect only those components that have proved significant. An appendix lists all orthogonal polynomials for situations with equal intervals and up to 20 trials. This listing greatly extends those currently available in the statistical literature, which provide all polynomials only for situations with up to 7 trials.
- Numerical Mathematics
- Statistics and Probability