Psychology Ph.D. Dissertations


Applying Neural Networking Techniques to Improve Performance and Turnover Prediction

Date of Award


Document Type


Degree Name

Doctor of Philosophy (Ph.D.)



First Advisor

Michael Zickar


Neural networking techniques were compared with standard regression techniques in a selection context. Neural networking models were hypothesized to outperform standard regression techniques in predicting turnover and six objective job performance metrics using a standard pre-employment assessment battery. Seven inputs, representing all independent variables, and seven outputs, representing the dependent variables, were assessed. A field sample (N = 1632) from the telecommunications industry was used. For regression models, ordinary least squares and logistic regression were used. A wide range of neural network models were tested. All neural network models examined in this study were supervised feedforward models with backpropagation of error. The number of hidden units in the neural networks were varied from 1 to 14. Three learning parameters of the networks were examined: .1, .2, and .3. Epochs of 100, 500, 1000, and 5000 were run on ten sub-samples, leading to a total of 1680 neural network analyses. Results indicated that neither regression techniques nor neural networking techniques consistently predicted turnover or job performance.