A Model Sampling Experiment to Evaluate Two Methods of Test Selection

摘要

Assignment of Army enlisted personnel to Advanced Individual Training following basic training is based on a battery of tests called the Army Classification Battery (ACB). Currently, the ACB scores are combined in pairs to form aptitude area scores each of which is considered predictive of performance for several Military Occupational Specialties (MOS) within one or more occupational categories. Optimal assignment of enlisted men to these specialties is based on these estimates of performance. As part of the Army research program to improve assignment of personnel, new test batteries to complement or replace existing test batteries are continually being developed. A procedure generally used in the development of test batteries involves (1) the formation of an experimental battery composed of a large number of tests--the test pool--and (2) the selection of some subset of the pool to form an operational battery. Selection of tests to predict a single criterion of job performance is relatively simple and straightforward; tests are usually sequentially selected to maximize the multiple correlation coefficient under the constraint that all previously selected tests are retained. However, the selection of tests to predict several criteria can be accomplished by several different methods. The present Research Memorandum describes the use of model sampling techniques to evaluate two of these test selection methods.