Naturally Occurring Noise: Experimental Economics & Stochastic Production Frontier Models

The University Of Central Florida Board Of Trustees, Orlando FL

Investigators

Abstract

The project examines the way in which controlled experimental data from human subjects can be used to improve the understanding and application of statistical estimators of latent data-generation processes. Controlled experiments can generate naturally occurring noise that can serve as a complement to the synthetic noise of standard Monte Carlo methods. In addition, those data will demonstrate the importance of using general statistical techniques for correcting for errors in variables problems that often are neglected in economics. In this manner, more natural measurements of data can be used to test general statistical estimators applied to human behavior and to generate sharper measurement tools for social scientists to apply in the field. Specifically, this methodology is applied to an investigation of the small-sample properties of several estimators for stochastic production frontier models commonly used in economics to see how well they are able to predict the latent performance frontier. This research exploits the complementarity between seemingly different empirical methods by highlighting the role of experiments as supplements to Monte Carlo studies for tests of statistical methods. By using experimental data from a range of controlled environments as a test-bed for statistical methods used in the field, we will have a better sense of the reliability of those statistical methods in field applications when the true underlying environment is unknown. Thus, the research contributes to a broader development within experimental economics to be much more concerned about the right statistical methods and in particular the right error specifications to be applied to such rich data. The use of natural noise from experiments, as a complement to Monte Carlo simulations, will be valuable for tests of any statistical method applied to data generated by human decision makers. Hence we will see how lessons from the controlled lab and observations from the uncontrolled field can be used together to make more reliable statistical inferences about human behavior.