A Comparison of Artificial Neural Networks and Multiple Regression in the Context of Research on Personality and Work Performance

Abstract

The current study provides an exposition of artificial neural network (ANN) methodology in the context of research on personality and work performance. We demonstrate some of the benefits and limitations of this methodology relative to multiple regression (MR) for conducting exploratory research. Using three data sets that each contained personality scores and measures of work performance, we compared the predictive accuracy of ANNs to both simple and complex MR equations. Across the three data sets, the neural networks performed as well or better than the MR equations on a relational measure of predictive accuracy but performed no better than the simplest regression equations on an absolute measure of predictive accuracy. Furthermore, through a combination of sensitivity analysis and graphical representations, we were able to identify the specific configural and nonlinear relationships that accounted for the superior performance of the neural networks with respect to the relational measure. The implications of the findings for researchers interested in applying ANNs to study organizational behavior are discussed.