在試題反應理論下的表現衰退模式

Item Response Theory Models Capturing Performance Decline

當考生在作答測驗時，可能因動機或時間不足導致表現衰退，進而影響分數表現之推論效度。本研究目的在於擴展與修正Jin與Wang（2014b）所提出的表現下降模式，用以偵測考生在作答過程中表現逐步衰退的現象，依據四種常用的多元計分的試題反應理論模型，本研究擴展成為四種改良模型，分別為表現逐步衰退評定量表模式（rating scale model with gradual performance decline, RSM-GPD）、表現逐步衰退部分計分模式（partial credit model with gradual performance decline, PCM-GPD）、表現逐步衰退一般化評定量表模式（generalized rating scale model with gradual performance decline, GRSM-GPD），及表現逐步衰退一般化部分計分模式（generalized partial credit model with gradual performance decline, GPCM-GPD）。透過模擬研究，操弄試題長度、受試者人數、反應量尺、完全努力率與衰退率，以檢驗模式參數回復效果。模擬研究結果顯示，整體而言各參數回復性表現良好，且受到考生人數及試題長度影響較大；當人數及題數提升時，參數回復性愈佳；而在模式間比較，可發現愈趨於簡易之模式，參數估計愈精準。此外，研究者也以閱讀成就測驗作為實徵分析的範例，並說明忽略表現衰退現象時對於考生表現推論之影響。最後，依據研究結果提出相關建議，以提供未來研究之參考。

During the administration of a test, examinees may experience a decline in interest or face time constraints, resulting in gradual performance deterioration. To address this phenomenon, this study modified the model proposed by Jin and Wang (2017b) to develop a class of polytomous item response theory (IRT) models that can capture this performance decline. The study developed four modified IRT models: a rating scale model with gradual performance decline, partial credit model with gradual performance decline, generalized rating scale model with gradual performance decline, and generalized partial credit model with gradual performance decline. These models were assessed for estimation quality through simulation studies. The results revealed that model parameters could be recovered satisfactorily and that increases in test lengths and sample sizes were associated with improved quality of parameter recovery. The study conducted an empirical analysis using Progress in International Reading Literacy Study data to demonstrate the practical application of these proposed models and to investigate the consequences of using inappropriate models for data analysis. The authors conclude by offering several recommendations for future studies in this area.