HKIER

摘要

在任何涉及自製或標準問卷的心理、教育、或社會研究，工具的信度及效度均是研究者極為關注的課題。傳統上為了解量表內項目間關係，研究者檢視信度（尤其是alpha值）、內在一致性、均一性及單度向性。不過這些概念大都含意未盡相同，且定義含糊，頗多研究者更誤以為alpha值與度向性有一一對應關係。本文引述文獻，重申問題向在，並用模擬研究法，展示高alpha值並不保證單度向，而低alpha值也不一定表示量表是多度向的。

Abstract

In psychological, educational, and social research involving self-constructed or standardized tests, the reliability and validity of the instruments are of great concern to the researchers. Customarily, reliability (particularly Cronbach's alpha), internal consistency, homogeneity, and unidimensionality are used, sometimes interchangeably and quite confusingly, to describe the relationships among the items in the same scale. However, there is a common misconception that alpha value and dimensionality have a one-one correspondence relation. The present article reviews the problems associated with using alpha as an indicator of unidimensionality. Furthermore, using simulated data, it is demonstrated that high alpha does not guarantee unidimensionality and low alpha does not imply multidimensionality.