瑟斯顿模型是一种IRT模型，使用该模型对人格迫选测验进行计分可以解决传统计分方式导致的自模式数据问题，然而，以往的研究在构建迫选测验时没有考虑人格迫选测验的抗作假效果，在配对时未匹配社会称许性。本研究采用模拟研究的方法，在匹配题目社会称许性的基础上，探讨题目数量、特质间相关、配对陈述间载荷差异、社会称许性与载荷间相关等因素对瑟斯顿IRT模型拟合及估计精度的影响。研究考察的指标包括模型收敛、模型拟合、参数估计及测验信度，结果发现各因素均会对瑟斯顿模型的应用效果产生一定的影响。具体来说：①题目数量对模型应用的影响最大，题量越大，模型估计越准确，模型稳定性越高；②配对陈述间的载荷差异越大，模型收敛越好，模型估计越准确，测验信度越高；③特质间相关对模型参数估计和特质分数估计的影响存在不一致的地方，需进一步探讨；④社会称许性与载荷间相关越大时，模型收敛越差，模型估计越不精确。本研究可以为抗作假的人格迫选测验的开发提供科学的依据和参考。

Forced-choice questionnaires have been controversial because their traditional scoring methods result in ipsative data. For addressing this problem, the Thurstonian model, an item response model, has proved helpful. However, previous studies didn’t consider social desirability when pairing the items and thus render the questionnaires vulnerable to fake. The present study was a simulation study in which we considered social desirability when pairing the items and compared the application of the Thurstonian model under different conditions. Twenty four conditions were examined by crossing the following factors:(a)items per trait(10,20 or 40);(b)the trait correlation (0.2 or 0.6);(c)correlation between social desirability and factor loading(0.3 or 0.6);(d)differences of factor loadings within the same pair(20%>0.2 or 60%>0.2). The study shows that these factors will influence the application of the model. Specifically, the results are as follows: (1)Items per trait had the biggest impact on the application of the Thurstonian model, with a larger amount indicating both better model estimates and stability.(2)Larger differences in factor loadings of statements within the same pair led to better model convergence and estimates.(3)Impacts of correlation between traits on parameter recovery and Latent trait score recovery were inconsistent, so this impact should be further examined.(4) Increased correlation between social desirability and factor loadings worsened model convergence and estimates.