Abstract: According to the literature review and references of psychological testing about foreign armies, a spatial ability item bank was developed based on the principles of item response theory (IRT) and theories about spatial abilities.
At first, we conducted a Paper and Pencil (P&P) test as a pilot study to explore the possibility of developing a picture assembling test using computerized adaptive testing (CAT) form. And correlation of this P&P test and task performance subscale of Soldier Performance Assessment Scale was analyzed. Then at the basis of the pilot study, a computer-administrated test (as a part of Nationwide Psychological Testing for Recruitment) was completed to modify and enlarge the item bank further. In this test, 7 different test forms, which were linked using anchor items, were developed and tested. Data was analyzed under 3 Parameter Logistic Model (3PL) using Bilog-MG software. Qualified items were selected to form the item bank, and a CAT form picture assembling test was developed. At last, we conducted validation test using subtest of block design in WAIS and scores of three curriculums as criterion.
From the results, we could see that the CAT form picture assembling test could satisfy the three assumptions of 3PL model, including unidimensionality, local independence and no speeding. And discrimination, location and guessing parameters, which were estimated using marginal maximum likelihood estimation (MMLE), were satisfying. So were ability parameters, estimated using Bayes expected a posterior estimation (EAPE). Furthermore, both test’s and items’ information functions were good. On the other side, item exposure can be controlled properly by strategy of using maximum information procedure and a-stratified method in this test. And on the basis of current item bank, using fixed-length stopping rule is appropriate for Nationwide Psychological Test for Recruitment. As for the validation study, we found that results of P&P form of picture assembling test had significant positive correlations with task performance subscale of Soldier Performance Assessment Scale. Scores of CAT form had significant positive correlations with subtest of block design in WAIS and examination scores of math and physics, but had no significant correlation with scores of Chinese.
However, a good CAT test need a good and large item bank to be based on, so the item bank in this study should be enlarged continuously

Key words: IRT, CAT, Spatial ability, Enlistees