Testing the Statistical Significance of Linear Programming Estimators

Linear programming-based estimation procedures are used in a variety of arenas. Two notable areas are multi-attribute utility models (LINMAP) and production frontiers (Data Envelopment Analysis, DEA). Both LINMAP and DEA have theoretical and managerial advantages. For example, LINMAP treats ordinal scaled preference data as such in uncovering individual level attribute weights, while regression treats these preferences as interval scaled. DEA produces easy to understand efficiency measures which allow for improved productivity benchmarking. However, acceptance of these techniques is hindered by the lack of statistical significance tests for their parameter estimates. In this paper we propose and evaluate such parameter significance tests. Two types of tests are forwarded. The first examines whether a model’s fit is significantly reduced when an explanatory variable is deleted. The second is based on generating a standard deviation or distribution for the parameter estimate using nonparametric jackknife or bootstrap techniques. We demonstrate through simulations that both types of tests reliably identify both significant and insignificant parameters. The availability of these tests, especially the relatively simple and easy to use tests of the first type, should enhance the utilization of linear programming-based estimation.