A Technique to Help Choose between Alternative Functional Forms of the Regression Equation
Abstract
The cost analyst is often faced with the problem of choosing among several different functional forms of a regression equation. For example, a problem frequently encountered in empirical cost research is the choice between a linear and a log-linear equation. Ideally, theory tells the researcher which to choose. However, in the absence of any firm theoretical indication, the cost researcher must resort to some ad hoc procedure in choosing between the functional forms. If the appropriate functional form is not used the estimate may be biased and/or inefficient. A procedure often used is to run both functional forms and select the one with the largest coefficient of determination (R-square). Unfortunately, the R-square statistic of equations 1 and 2 cannot be meaningfully compared because the dependent variables in the two equations are different. Rao and Miller state, The specification of the model, the error terms, and the computation of R-square for these two equations are entirely different and provide no common ground for comparison of the relative performance of these equations on the basis of computation of R-square.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 1985
- Accession Number
- ADA161909
Entities
People
- Thomas P. Frazier
Organizations
- Institute for Defense Analyses