利用混合模型探討英文詞彙測驗受測者之作答順序
No Thumbnail Available
Date
2017
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
研究第一語言及第二語言的研究者認為詞彙量對於語言能力有著重要的意義。而英文詞彙測驗(Vocabulary Levels Test, VLT)是一個廣泛應用在測試詞彙量的測驗，可有效確認受測者是否擁有足夠的詞彙量。由於英文詞彙測驗有10個題組的試題結構，文獻中已提出利用二參數英文詞彙測驗的依序試題模型(VLT-Sequential Model, VSM)，解決題組內題目間所存在的相依性的問題。當應用該模型來分析實徵資料時，出現配適不佳的狀況。除了考量該測驗中配適不佳的試題是否具有其特殊性之外，希望進一步地對於二參數VSM模型中所假設的全部受測者都是按照題目難易度順序作答的合理性進行檢定，故本論文將建構混合二參數VSM模型，進一步考慮也存在有按照題目順序作答的族群的可能性。在模擬研究中，使用邊際最大概似估計來估計參數，確認混合二參數VSM模型估計的有效性，並觀察當資料中同時存在有按照題目順序及題目難易度順序作答的受測者時的作答，卻僅使用二參數VSM模型，在參數估計上會有何影響。分析實徵資料3,000字等級的英文詞彙測驗中，確認二參數VSM模型配適度不佳的情況與按照題目作答順序無關，而在5,000字等級的英文詞彙測驗結果中發現，混合二參數VSM模型表現優於二參數VSM模型，更進一步，如果忽略其實存在有按照題目順序作答的這個族群，僅用二參數VSM模型來進行分析，將會低估這類受測者的能力。
Vocabulary knowledge is considered by both first-language and second-language researchers to be of great significance in language competence. Vocabulary Levels Test (VLT) is commonly used and shown effective in measuring learner’s vocabulary size. The test consists of ten item clusters, and two-parameter logistic VLT-Sequential Model(2PL-VSM) considers taking into account the dependency structure among items within a cluster in modeling VLT data. However, some clusters exhibit misfit while fitting the actual 3000-level data with a 2PL-VSM.The purpose of this study is to construct a two-parameter logistic Mixture VLT-Sequential Model(2PL-MVSM) that relaxes the assumption that the items within a cluster are answered in the order of their item difficulty, from easiest to the most difficult ones, and allows some examinees to answer these items in their order presented in the original test, i.e. item number. In the simulation study, we investigate the effect of ignoring the examinees answering items within a cluster by item number by comparing the estimation results from the 2PL-VSM to those from the 2PL-MVSM. In analyzing the 3000-level and 5000-level VLT data, the results show no necessity to include the additional class of answering-by-item-number examinees for 3000-level data, whereas 2PL-MVSM is shown to be superior over 2PL-VSM in the actual 5000-level data. Furthermore, the abilities of the examinees who answer the items by item number would be under-estimated if only 2PL-VSM is fit to the 5000-level data.
Vocabulary knowledge is considered by both first-language and second-language researchers to be of great significance in language competence. Vocabulary Levels Test (VLT) is commonly used and shown effective in measuring learner’s vocabulary size. The test consists of ten item clusters, and two-parameter logistic VLT-Sequential Model(2PL-VSM) considers taking into account the dependency structure among items within a cluster in modeling VLT data. However, some clusters exhibit misfit while fitting the actual 3000-level data with a 2PL-VSM.The purpose of this study is to construct a two-parameter logistic Mixture VLT-Sequential Model(2PL-MVSM) that relaxes the assumption that the items within a cluster are answered in the order of their item difficulty, from easiest to the most difficult ones, and allows some examinees to answer these items in their order presented in the original test, i.e. item number. In the simulation study, we investigate the effect of ignoring the examinees answering items within a cluster by item number by comparing the estimation results from the 2PL-VSM to those from the 2PL-MVSM. In analyzing the 3000-level and 5000-level VLT data, the results show no necessity to include the additional class of answering-by-item-number examinees for 3000-level data, whereas 2PL-MVSM is shown to be superior over 2PL-VSM in the actual 5000-level data. Furthermore, the abilities of the examinees who answer the items by item number would be under-estimated if only 2PL-VSM is fit to the 5000-level data.
Description
Keywords
混合模型, 英文詞彙測驗, 作答順序, Mixture model, Vocabulary levels test, Answering order