-
Authors:
高清辉
-
Information:
厦门大学,厦门
-
Keywords:
Reliability estimation; Cronbach’s α; MacDonald’s ω
信度估计; 克隆巴赫α; 麦克唐纳ω
-
Abstract:
Coefficient Cronbach’s α is the most widely used for estimating reliability. Researchers have
used the index extensively in the papers that need to report its reliability, but many scholars have also
questioned it. The α coefficient is used as reliability and must satisfy the “essential tau-equivalence”
assumption. This assumption is too strict and difficult to meet, and its violation may lead to α
overestimating or underestimating the reliability. Using Cronbach’s α to estimate internal consistency is
inappropriate. The acceptable lower bound to the reliability of a test is often set empirically, and there is
no precise standard. Researchers increase the α value by deleting items, which may also lead to a decrease
in the actual reliability of the scale. Although these problems exist, α has been widely used in related
research for a long time. This is due to the following reasons: many research fields involve reporting
reliability coefficients, researchers have not been taught how to use α correctly for a long time; in addition,
standard statistical software has the function of calculating α, which is convenient for calculation; the
editors also have requirements for reporting α in the paper. McDonald pointed out that α is a particular
case of MacDonald’s ω, and ω becomes α when the “essential tau-equivalence” is satisfied. ω is better
than α when the “essential tau-equivalence” cannot be satisfied in reliability estimation. However, the
calculation of ω must use confirmatory factor analysis (CFA), which is challenging to implement in the
pre-computer era. Researchers have gradually started to use ω instead of α in their research, and more and
more people have accepted ω. However, whether MacDonald’s ω should be used instead of Cronbach’s
α, there is still a heated debate in the academic community. Opinions on ω mainly focus on the fact that
the actual values of ω and α are not significantly different in the calculation. That ω may show more
estimation failures when the sample size and overall reliability are small. It is unreasonable to abandon α
or be unwilling to move forward and stick to α. In future research, use McDonald’s omega and Cronbach’s
alpha for reliability estimation to coexist for a long time and complement each other.
在众多信度指标中,克隆巴赫α (Cronbach’s α )系数最为常用,但α 系数在使用中常常忽视重要的“基本τ 等价”假设,导致α 系数并非真正的信度;并且存在着误用与滥用、阈值不确定等问题,受到很多批评,被长期广泛应用是因其易于计算、易于理解等特性。麦克唐纳ω 克服了α 系数的上述缺点,信度估计上明显优于α 系数,但麦克唐纳ω 也存在计算要求较高等缺点,尚无法取代克隆巴赫α 。两个信度指标ω 与α 将长期共存,在研究报告中应同时提供ω 与α 的数值,互相补充。
-
DOI:
https://doi.org/10.35534/pc.0408114
-
Cite:
高清辉.取代还是共存——信度估计系数阿尔法与欧米伽[J].中国心理学前沿,2022,4(8):954-961.