Development of Mental Health Assessment System with CAT Technique and Its Application to Workplaces

（1）差异项目功能分析（Differential Item Functioning，DIF）对来自东亚、北美和南美大学生的流行病学研究中心抑郁量表（CES-D）数据进行跨文化/种族DIF分析。与北美和南美的人相比，日本人在控制了抑郁情绪的潜在特征后，有抑制积极情感表达的倾向。（2）项目反应理论（IRT）分析采用项目反应理论的广义部分信任模型（Generalized Partial Credit Model，简称IRT），以简明压力问卷（BSQ）和CES-D为材料，采用四个选项，检验两个选项之间的间隔是否相等。在心理学项目中，“经常”和“几乎总是”的反应间隔很窄。对于抑郁项目，“几乎从未”到“几乎总是”之间的范围较窄，这表明抑郁症状将更好地通过二分类型的测量进行评估。（3）心理健康计算机化自适应测验（CAT）系统的开发利用IRT的修正等级反应模型（Modified Graded Response Model）获得的BSQ应激症状项目的定位参数和判别参数，开发了CAT系统的初始版本。CAT算法采用限制贝叶斯方法进行项目选择，停止规则θ估计的标准误< 0.35。开发了反馈显示器，以在执行后立即显示结果。(4)A CAT系统的现场试验该初始CAT系统的现场试验显示：1）大约。10个条目足以估计被访者的水平; 2）平均反应时间为7秒; 3）具有较高区分度的条目更有可能被选择的趋势;因此4）区分度较差的躯体症状条目应设置为另一个CAT模块。这一制度的进一步完善，使职业心理健康工作有了更大的创新。

(1)Differential Item Functioning (DIF) AnalysesThe cross-cultural/ethnic DIF analyses were conducted on data of the Center for Epidemiologic Studies Depression Scale (CES-D) obtained from university students in East Asia, North and South America. Japanese have a tendency to suppress the expression of positive affect as compared to people in North and South America, after controlling the latent trait of depressive symptomatology.(2)Item Response Theory (IRT) AnalysesThe Generalized Partial Credit Model of IRT was employed to test whether the intervals between response alternatives were equal or not using data on the Brief Stress Questionnaire (BSQ) and CES-D, both with four-response alternatives. The interval between the responses of "often" and "almost always" was narrow for psychological items. The range between "almost never" through "almost always" was narrower for depressive items, suggesting that depressive symptoms would be better assessed by dichotomous type of measurement.(3)Development of Computerized-Adaptive Testing (CAT) System for Mental Health AssessmentAn initial version of CAT system was developed using location and discrimination parameters of the BSQ stress symptom items obtained by the Modified Graded Response Model of IRT. A CAT algorithm employed a restricted Bayesian method for item selection, and standard error of θ estimation < 0.35 for stopping rule. A feedback display was developed to show a result immediately after execution.(4)A field trial of the CAT systemA field trial of this initial CAT system revealed that 1)approx. 10 items were sufficient to estimate a respondent's level ; 2)the mean response time was 7 seconds ; 3)a tendency that item with higher discrimination would be more likely to be selected ; and thus 4)somatic symptom items having poor discrimination should be set for another CAT module. Further improvement of this system seems to make greater innovation in occupational mental health activities.