Assessment of Differential Prediction by Race for the USNA Classes of 1986-1990
Technical note Aug 1989-Feb 1990,
NAVY PERSONNEL RESEARCH AND DEVELOPMENT CENTER SAN DIEGO CA
Pagination or Media Count:
The USNA uses the Candidate Multiple CM, a composite of several measures, to evaluate applicants. This composite seeks to predict many aspects of USNA performance. While the USNA strives for equal opportunity, the CM has not been evaluated for fairness for some time. Toward that end, the CM was evaluated using the widely accepted regression model of test fairness. This model effectively states that a test is biased for a subgroup if it consistently over-or underpredicts the performance for members of that group. Using this model, procedures were used to statistically assess whether a prediction system based on the CM is fair to Blacks when the system for predicting academic performance is developed on a combined sample of Blacks and Whites. In essence, these procedures are aimed at determining whether a separate regression equation i.e., prediction system for members of the minority group would provide a more accurate and equitable prediction of their performance. These analyses revealed a statistically significant difference between separate prediction systems. This difference, while statistically significant, increases prediction by only about three-fourths of one percentage point. Thus, the use of separate equations does not provide a meaningful improvement in the prediction of academic performance. Finally, since Black academic performance scores were generally overpredicted when the combined-group regression equation was used, a combined-group regression equation does not adversely impact upon the Black minority. It is therefore recommended that the USNA continue using a single prediction system.
- Administration and Management
- Sociology and Law