為了解在評分者中介評量中，評分者是否會因為受試者潛在類別不同而有不同的反應風格，本研究旨在發展「多面向混合極端反應風格試題反應理論模式」，並檢驗其有效性。透過模擬研究的設計，在各種操弄情境下，使用貝氏估計法檢驗模式參數的回復性。模擬研究結果顯示：模式在試題參數、評分者參數、受試者能力參數、潛在類別的估計有良好的估計效果，且隨著量表點數、題項數、樣本數愈大，其估計效果愈好；但忽略潛在類別時，所有參數估計精準度皆會降低。而在實徵研究中使用泰國一所國際學校對學生的英文寫作測驗作為資料來源，以檢驗發展的模式在實務上之適配情形，結果顯示：在多面向混合極端反應風格概化部分計分模式有最佳的適配度。最後，研究者也提出相關建議供後續研究參考。

In order to understand whether raters in rater-mediated assessments exhibit different response styles on rating scales due to different latent classes of ratees, this study aims to develop a "many-facet mixture extreme response style item response model" and assess the efficiency of the proposed model. This study consists of two parts: simulation study and empirical study. The simulation study results show that the model has good parameter estimation for item and person parameters and that the precision of parameter estimates declines when the misleading model was used. In the empirical study, the data that were collected from writing assessment administered to an international school in Thailand were fit to several competing models to demonstrate the application of the proposed model. The empirical study shows that the many-facet mixture extreme response style generalized partial credit model provides the best-fitting model among the competing models. Moreover, the changes in rank order between the best-fitting model and it corresponding model without considering different response styles and latent classes are not trivial. Finally, the author addresses conclusions based on the results and provides suggestions for future study.