| The data from graduate student applications at a large Western University were used to develop profiles of successful graduate students, as defined by cumulative graduate grade point average. Two statistical models were employed and compared, artificial neural networking and simultaneous multiple regression. The sample was divided into 3,097 masters and 805 Ph.D. students' university wide who entered between Fall 1990 and Spring 1994. In addition, Non-parametric statistics, including Mann-Whitney U and Spearman Rho were used to determine whether each individual variable was significantly related to graduate grade point average. Non-parametric statistics were used due to the violation of normal distribution by the dependent variable, graduate grade point average.;Results of the artificial neural network and regression models for master's students yielded similar results indicating that the combination of the college the student was applying for, marital status, gender, GRE verbal and analytical scores, and residency region of students could predict 10%--12% of the variance in graduate grade point average. Results of the two models for Ph.D. students were not as similar. The best of five artificial neural networks used the combination of GRE analytical, quantitative and verbal scores; gender, marital status, age, residency region, and citizenship continent to predict 24% of the variance in graduate grade point average. Multiple regression indicated that college, GRE verbal and analytical scores, marital status, and age could predict 9% of the variance in graduate grade point average. Caution should be taken when interpreting the results of the ANN Ph.D. predictive models because two of the five neural networks performed worse than the regression.;Because of having more confidence in the master's data set, the conclusion was that artificial neural networking and simultaneous multiple regression provide similar results in determining which student characteristic variables provide the best profile of students with high graduate grade point averages. A better understanding of the interpretation of artificial neural network outputs in these types of research questions is recommended. In addition, maintaining accurate and complete historical databases for analyses is necessary. |
