应用项目反应理论对《中国士兵人格问卷》的项目分析

采用项目反应理论（IRT）对《中国士兵人格问卷》进行项目分析。计算机呈现中国士兵人格问卷（CSPQ）对100,523名适龄男性青年进行测验，随机抽取2676名任一维度标准分均低于70的定为合格组；将任一维度大于70分并经专业人员访谈不合格的274名定为不合格组；从精神病院抽取男性年龄相当的221名缓解期精神分裂症患者定为精神病组，并完成CSPQ测验。运用基于IRT的双参数Logistic模型进行分析；结果发现，区分度参数超过区间(0.30,4.00)的条目删除前后，被试的能力值与标准分均存在显著相关；精神病组的测验分数经IRT分析，图形曲线与不合格组有高度吻合。研究结果说明，在测验精度基本相同的条件下，应用IRT可以减少施测条目，提高测验效率，可在一定程度上更精确地区分被试的特质水平

Item Analysis of the Chinese Soldier Personality Questionnaire Using Item Response Theory

With the development of psychological and educational measurement, item response theory (IRT) is being increasingly used in psychological and educational testing areas. There have been many large-scale tests based on IRT, such as TOEFL and GRE. Further, there have also been many studies on the applications of IRT in personality questionnaires. However, many problems still remain, including model selection and determining the appropriate numbers of parameters. It is vital to conduct empirical studies in order to discuss the probability of applying IRT to this area. The Chinese Soldier Personality Questionnaire (CSPQ) was used in this study in order to discuss the possibility of applying IRT to personality questionnaires with large samples. In the study, 100,523 young individuals were tested using the computerized CSPQ. Subjects whose standard score in every subtest was less than 70 were considered to be normal, while the others were disqualified. Further, structural interviews were conducted by psychological experts in order to confirm the symptoms of those disqualified. If they indeed exhibited some symptoms, they were placed in the disqualified group. Additionally, 221 schizophrenia patients were tested using the same questionnaire, and they were placed in the schizophrenia group. From among the normal subjects, 2676 were randomly selected and placed in the normal group. The SPSS 11.5 software was used to test the unidimensionality of every subtest. Moreover, the BILOG-MG software was used to analyze the data with the two-parameter logistic model (2PL) based on IRT. Items with discriminating parameters beyond the range of (0.3, 4) were deleted as they provided little information that was relevant to testing. Moreover, before and after deleting these items, there was no noticeable change in the maximum test information and the correlations between IRT scores and standard scores. In the case of the standard scores, the distributions of the schizophrenia group were similar to those of the normal group, while in the case of the IRT scores, the distributions of the schizophrenia group moved towards those of the disqualified group. As shown in this study, IRT can improve the efficiency of testing with nearly the same precision. Further, IRT is more suitable for identifying the differences between varied groups, which is another advantage of IRT analysis. Therefore, as a new testing theory, IRT can and should be used in the analysis of personality questionnaires. However, personality questionnaires are not the same as cognitive tests, and different models of IRT (for example, nonparametric models) have not been compared in this study. Therefore, the best model for personality questionnaires should be discussed further