• 作者： 林妙香
• 作者服務機構： 中央研究院統計科學研究所
• 中文摘要： 本文先行摘述研究單維檢視法優缺點，繼而就文獻尚未顧及的層面評估單維檢視法判斷試題單維的正確性。本文研究設計擬似探討第一類型誤差，亦即就符合二參數常態肩形曲線的模擬資料比較單維檢視法的效度。本文所使用的資料涵蓋不同的試題與能力相關的範圍及測驗長度，而所評估的單維檢視法來自因素分析理論，領域抽樣理論，Stout氏基本單維論，以及試題適合度考驗法。 依據研究發現，本文建議使用單維檢視法考量原則如下：1)就領域抽樣理論而言，若欲選一組試題滿足單維及高同質性，則試題相關平均指標優於alpha係數；2)就斯皮爾曼單一共同因素理論之單維定義而言，主軸分析法優於主成份分析及最大概法，因其所抽取因素特徵值類型近似母群參數類型；3)若由試題滿足最基本單維數考量，則Stout氏的DIMTEST併隨因素選題法優於先驗選題法及其他單維檢視法；4)若欲檢視個別試題符合二參數常態肩形曲線符合的程度，當題數大於或等於20，可先後以 才統計量及標準化殘餘值進行二階段檢視程序。
• 英文摘要： This study reviewed the performance of various measures of unidimensionality by using simulated
  data sets which were in compliance with the two-parameter normal ogive model conditions. The
  investigated measures were based on homegeneity, factor analysis, item fit statistic, and Stout's DIMTEST
  methodology. The simulated data sets varied in terms of the item-ability correlation range and test length.
  Based on the overall results of this empirical review, the following suggestions are offered regarding
  use of different unidimensionality measures. To arrive at a unidimensional and homogenous set of items,
  an index based on average inter-item correlation is better than one based on the coefficient alpha. In
  determining whether a data set satisfies an assumption of unidimensionality in terms of a single common
  factor model, the principal axis method performs much better than do the other methods in terms of
  examining the extracted eigenvalue pattern based on tetrachoric correlation matrices. To ensure that
  test items hold essential local independence, Stout's DIMTEST factor procedure is preferred not only
  to the prior knowledge procedure, but also to other unidimensionality measures. To assess whether an
  individual item fits the two-parameter normal ogive model, a two-stage procedure consisting of goodness-
  of -fit by chi-square statistic and root mean square standardized residual may be used when n?20.
• 中文關鍵字： --
• 英文關鍵字： --