抗作假人格迫选测验中瑟斯顿IRT模型的影响因素

Influencing factors of Thurstonian IRT model in faking-resisting forced-choice questionnaire

使用蒙特卡洛(Monte Carlo)模拟研究的方法,探讨应用瑟斯顿IRT模型对抗作假迫选测验计分时需满足的编制条件,考察了测验所测特质个数(2或5个)、每维度包含陈述数量(10或20个)、单维配对题目比例(0或20%)和正负向陈述间配对题目比例(0、20%或40%)对模型估计的影响.结果如下:1)特质个数对模型估计有显著影响,测验所测特质个数越多,模型估计越准确;2)陈述数量影响模型估计,测验包含陈述数量越多,模型估计越精确,且当特质个数较少时,陈述数量的影响更大;3)测验中单维配对题目的比例基本不影响瑟斯顿IRT模型的估计精度;4)测验中加入一定比例(约20%)的正负向陈述间配对题目可提高模型的估计精度,且特质个数较少时,该因素的影响更大.最后,在研究结果的基础上给出了开发抗作假人格迫选测验的建议.

A Monte Carlo simulation was used here to compare performance of Thurstonian IRT model across a variety of faking-resisting forced-choice designs. In total, 24 conditions were examined by crossing the following factors： 1） number of trait （2 or 5）; 2） items per trait （10 or 20）; 3） percentage of pairs composed of items on the same dimension （0 or 20%）; 4） percentage of pairs composed of items keyed in opposite directions （0, 20% or 40%）. Larger number of traits was found to lead to more accurate estimation of model parameters, and more precise recovery of trait score. Items per trait had a positive impact on application of Thurstonian model, with a larger amount indicating more precise parameter estimation and better true trait score recovery. It had greater influence when there were not enough traits. The percentage of item pairs composed of items on the same dimensions had no significant influence on model estimates. Including a percentage of pairs with items keyed in opposite direction （20%） could greatly improve accuracy of parameter and trait score estimation. The factor impact was greater when the number of traits measured in the questionnaire was small. These results have important implications for how faking-resisting forced-choice questionnaires should be designed in the future.