理性刪除克漏字測驗效度的重新審視：驗證性因素分析方法的應用

此研究旨在證實容易編寫的理性刪除克漏字測驗（rational cloze test）可以很有信度及效度地用來測試課堂上學生的各種不同閱讀技能。此研究藉由驗證性因數分析方法（confirmatory factor analysis），重新探討Bachman（1982）所提出的主張：理性刪除克漏字測驗是可以用來測試課堂上學生的各種不同閱讀技能，尤其是銜接理解能力及連貫理解能力。713 位來自臺灣北部四所大學的同學，在研修一般性英文課程的課堂上，填答了Bachman（1985）所編寫一份的理性刪除克漏字測驗。其結果顯示本研究的數據與整體特質加四個細項特質的二階模式比較吻合；也就是說，理性刪除克漏字測驗是可以用來測試課堂教師想要考的各種不同閱讀技能，如文法能力、跨子句的銜接及連貫理解能力、跨句子的銜接及連貫理解能力、及篇章文字線索外的推論能力，而且這四個細項能力可以用二階整體特質解釋。

In response to the demand for quality tests that are easy to construct and in view of the cohesive and coherent competence that is important to reading comprehension and writing development, the present study was motivated and conducted, in the hope that its results could prove informative as to whether easily constructed rational cloze tests could be customized for classroom language teachers’ testing objectives with high reliability and construct validity. Employing the confirmatory factor analysis (CFA), the current study aimed to investigate whether additional evidence can be obtained to support Bachman’s (1982) claim that the rational cloze test can be designed by classroom language teachers to measure an array of reading skills as intended, specifically including cohesive and coherent competence. In this study were a total of 713 participants, all college students, taking the general English courses at four universities in northern Taiwan. The participants’ dichotomously-scored responses to a rational cloze test constructed by Bachman (1985) were analyzed through the CFA on the item-by-item level in the program Mplus, which provides a convenient mechanism to perform the CFA of dichotomous responses. The present study, partly following Bachman (1982), tested the general trait model, the four specific traits model, and the general plus specific traits model—the last comprising a general factor plus the four specific traits. The testing results of the present study have lent support to the hypothesized general plus specific traits model, stating that rational cloze tests can be designed to tap distinct and related language competence, such as syntactic competence, the cohesive and coherent competence that depends on contexts of across-clauses and across-sentences, and extra-textual inference, all of which could be explained by the second-order general language proficiency factor. Based on the findings of the present study, implications and recommendations were provided for future research and classroom language teachers, as well as language test constructors.