在社會科學的調查研究中，極端反應風格的存在經常會造成測驗品質的威脅。 本研究旨在發展一組新的試題反應理論（item response theory, IRT）模式，適用於 多向度測驗分析，並同時能納入極端反應風格的隨機效果。研究者擴展四種適用多 元計分的多向度試題反應理論（multidimensional IRT, MIRT）模式，並透過模擬研 究操弄多種不同情境，使用貝氏估計法檢驗模式參數的回復性。研究結果顯示：本 研究所發展出的四個新模式具有良好的估計品質。在試題參數估計方面，隨著向度 增加、樣本數增加與向度之間呈現高相關時，能提供良好的試題參數估計；在潛在 特質與權重參數估計方面，其樣本數、題數與向度增加時，受試者參數估計有不錯 的效果。此外，向度間的高相關也能提升各向度特質之估計。實徵研究分析顯示： 有納入極端反應風格的模式具有較佳的模式適配度，忽略極端反應風格時會造成潛 在特值估計的偏誤。最後，研究者提出若干建議供未來研究發展參考。

This study aims to develop a new set of multidimensional item response theory (MIRT) models for extreme response style (ERS). Four polytomous MIRT models were extended to incorporate the ERS effect in the item response functions. We conducted a series of simulations to assess the efficiency of the proposed models in terms of parameter recovery using Bayesian estimation. The results showed that a large sample size, a great number of dimensions, and high correlation between dimensions were associated with satisfactory item parameter recovery and that the latent trait and ERS weight parameters could be estimated precisely when the large sample size, the long test length and a great number of dimensions were used. In addition, as in multidimensional IRT models, the correlation between latent traits can provide precise latent trait estimation. An empirical example was provided to demonstrate the applications of the proposed model to real data analysis and the consequences of ignoring the ERS effect on persons’ parameter estimation were investigated. We closed this article by addressing several suggestions for future study.