The construction of parallel forms of a test designed to measure human relations insight

The primary objective of this study was to construct two parallel forms of Supervisor's Problems, a test designed to measure the human relations Insight of industrial supervisors. Criteria established for the resulting parallel forms included approximately equal means, equal standard deviations and equal intercorrelations. Additional objectives included satisfactory difficulty levels in all test items and test form reliability suitable for purposes of individual prediction. A related objective was to determine the ability of each test form to discriminate between various levels of education in selected behavioral science courses. The original form of Supervisor's Problems (Form AA) contained eight case problems and fifty test questions. Twelve additional case problems were selected from the twenty-four originally prepared by Armstrong Cork Company, and questions were developed for each additional case. A lengthy rough draft was written which contained all twenty cases and one hundred ninety-one questions. This draft was administered to three psychology graduate students and a personnel department supervisor from a chemical company. On the basis of item responses and general comments from this group, certain items were eliminated and others were revised. All remaining items were examined by the investigator and a member of the psychology faculty in terms of content and construction. Some additional items were revised or eliminated. The resulting twenty case problems and one hundred seventy-five items were divided for convenience of administration and further analysis into two forms. Supervisor's Problems Forms YA and YB. These forms were not intended to be parallel. Preliminary forms YA and YB were administered to ninety-seven undergraduate students enrolled in psychology and management courses. On the basis of obtained test data, an item analysis was performed. For purposes of this analysis, forms YA and YB were combined and treated as one test. The mean difficulty level of all items was 0.637, the average biserial correlation between the test items and total score was 0.312, and the estimated average interitem correlation was O.O97. The item analysis data were used to eliminate or revise unsatisfactory items and subsequently to construct parallel forms of Supervisor's Problems. These forms were designated XA and XB, each containing eight case problems and fifty questions. Both forms XA and XB were administered to an additional group of ninety-four students, and a separate item analysis was performed on the data from each test form. For Form XA the mean difficulty level of the items was O.566; for Form XB O.578. The average item-total score correlation on Form XA was O.298; on Form XB 0.307. Estimated average interitem correlation on Form XA was O.089; on Form XB 0.094. The test score distributions for the two forms did not depart significantly from normality. A reliability index was computed for each item. These indices appear in Appendix E along with other data specific to each item. The Kuder-Richardson Formula 20 reliability coefficient computed for Form XA was 0,642 for Form X3 0.662 Product-moment correlation between scores on the two test forms, or parallel form reliability was 0.494, significant at the .0l level of confidence. Students to whom forms XA and XB had been administered ware separated into three groups depending on the number of semester hours they had completed in selected behavioral science courses, mean scores were computed for each group on both forms of Supervisor's Problems. An F test showed no significant difference between the means of the three groups for either form of the test.