Posted by A Distribution‐Free Measure of the Significance of CER Regression Fit Parameters Established Using General Error Regression Methods
Models Track
Downloadable Files:
Abstract:
General Error Regression Methods (GERM) have earned a strong following in the cost estimating community as a means of establishing cost estimating relationships (CERs) using non‐linear functional forms. GERM has given rise to a wide variety of functional forms for CERs, but has so far lacked a means for evaluating the “significance ” of the individual regression fit parameters in a way that is analogous to the roles played by the t‐statistic in ordinary least squares (OLS) regression. This research attempts to remedy that situation by developing and describing an analogous “significance” metric for GERM regression fit parameters that is independent of the nature of the underlying error distribution.
The significance metrics developed herein are comparable across CERs regardless of the functional form of the regression equation or the underlying error specification.
Moreover, they require no distributional assumptions and they provide a collection of simple metrics by which to judge the “significance” of the individual regression fit parameters.
These metrics will be beneficial to anyone who uses GERM to develop CERs. They will enable cost modelers to judge whether or not cost‐driver variables used in CERs that are derived using GERM have any real impact on the mean result of the CER, and whether or not they contribute to the overall reduction in the CER’s variance. Those variables that, if excluded, would not substantially impact the mean result or significantly change the CER variance, can be removed from the cost model with little or no consequence.
This is often desirable because technical data collection can be difficult. Moreover, one less variable means one more degree of freedom, and that can be important when a small number of data points are available. So it is important to be able to eliminate from study those variables that are not significant cost drivers.
Author:
Timothy P. Anderson
Mr. Timothy P. Anderson is a technical manager for MCR, LLC’s Corporate Technical Directorate, and a professional cost analyst and operations research analyst with over fourteen years experience, primarily in the context of Department of Defense (DoD) weapon systems and national security space acquisition. Prior to arriving at MCR, Tim was a senior member of the technical staff at The Aerospace Corporation. His areas of interest are cost analysis, cost uncertainty analysis, operations research, and decision analysis. Mr. Anderson served for 20 years in the U.S. Navy and began working in the cost estimating field in 1994 while assigned to the Naval Center for Cost Analysis. From there he was assigned as a military professor at the Naval Postgraduate School, teaching cost estimation, operations research, and other technical courses. He retired from the Navy in June 2001. Mr. Anderson has a B.S. in Industrial and Operations Engineering from the University of Michigan, and an M.S. in Operations Research from the Naval Postgraduate School. He is a SCEA certified cost estimator/analyst, Program Chair for the Washington D.C. Area chapter of SCEA, an adjunct professor in the Systems Engineering/Operations Research Department at George Mason University, an adjunct professor of Engineering Economics and Cost Estimation for the Naval Postgraduate School (Distributed Learning Programs), and a frequent presenter of topics related to cost estimating and cost uncertainty analysis at forums including the Society of Cost Estimating and Analysis (SCEA), the Military Operations Research Society (MORS), the DoD Cost Analysis Symposium (DoDCAS) and the Space Systems Cost Analysis Group
(SSCAG).