中文摘要：

GIRM（Generalizability in Item Response Modeling）是一种将概化理论GT和项目反应理论IRT相结合后计算概化理论中方差分量的一种方法。当GIRM方法下θp和βi的抽样分布与GIRM方法中的MCMC先验分布一致时,GIRM方法对方差分量估计具有较高的准确性。为了进一步检验GIRM方法对IRT参数分布形态的敏感性,研究在将MCMC先验分布固定的情况下,探讨不同IRT参数分布形态下GIRM方法的适用性,并将所得结果与传统GT方法相比较。结果表明：（1）在各种参数分布形态下,采用GIRM方法估计IRT模型的参数是可行的;（2）GIRM方法在被试能力参数为标准正态分布时对σ2（p）估计的准确性高于传统GT方法,但在均匀分布和偏态分布下略差于传统GT方法;（3）GIRM方法在题目难度参数为偏态分布情况下对σ2（i）的估计准确性显著差于传统GT方法;（4）两种方法对于σ2（pie）估计的准确性在任何参数分布形态下都大致相当,优劣并无统一规律。

英文摘要：

Generalizability in Item Response Modeling （GIRM）is a combination of the Generalizability Theory （GT）and the Item Re- sponse Theory（IRT） ,which is used to estimate variance components. Under GIRM method,if the distributions of parameters θp and βi, which are simulation parameters on the IRT, accord with the prior distributions of MCMC method, GIRM approach estimates the variance components accurately. To investigate the performance of the GIRM approach under different distributions of parameters 0p and βi, this simulation study varys parameters distributions while the prior specifications in MCMC method stay fixed, Based on the simulated data, this study explores the applicability of GIRM approach, and compares the results of GIRM approach with that of traditional GT ap- proach. Conclusion shows that the GIRM approach can be used to estimate the parameters of IRT under different sampling distributions of IRT parameters. The estimation of σ2 （p） by GIRM approach is better than traditional GT approach when the distribution of Op is standard normal,but the result is opposite when the distribution of θp is uniform or skewed. However, when the distribution of βi is skewed,the estimation of σ2（i） by GIRM approach is worse than traditional GT approach. But even so,the performance of GIRM ap- proach is similar to the traditional GT approach in the estimation of σ^2 （p/e） under different situations.