Structural equation modelling: Adjudging model fit
Structural equation modelling: Adjudging model fit
复制标题
DOI：
10.1016/j.paid.2006.09.018
复制
发表时间：
2007-05-01
影响因子：
4.3
通讯作者：
Barrett, Paul
中科院分区：
文献类型：
--
作者：
Barrett, Paul
对于期刊编辑、审稿人和研究文章的读者来说，结构方程模型 (SEM) 拟合最近已成为评估方法中一个令人困惑且有争议的领域。可以确定两种模型拟合方法的支持者：那些严格遵守零假设显着性检验结果的人，以及那些忽略这一点而将模型拟合指数作为近似函数的人。两者都有各自行动方针的原则理由。本文认为，卡方精确拟合检验是 SEM 拟合的唯一实质性检验，但是，如果从解释理论的角度来看这些差异被认为是微不足道的，那么随着样本量的增加，它对与预期值差异的敏感性可能会产生很大的问题。另一方面，适当缩放的近似拟合指数不具有对样本大小的敏感性，但它们也不是模型拟合的“测试”。针对这一困境的建议解决方案是考虑在理论相关标准的预测准确性方面接受一种解释模型而不是另一种解释模型的实质性“后果”。如果没有可供评估的，那么就建议不能对“竞争”模型进行科学上有价值的区分，这当然引出了一个问题：为什么首先要进行这样的 SEM 应用。 (c) 2006 Elsevier Ltd. 保留所有权利。
For journal editors, reviewers, and readers of research articles, structural equation model (SEM) fit has recently become a confusing and contentious area of evaluative methodology. Proponents of two kinds of approaches to model fit can be identified: those who adhere strictly to the result from a null hypothesis significance test, and those who ignore this and instead index model fit as an approximation function. Both have principled reasons for their respective course of action. This paper argues that the chi-square exact-fit test is the only substantive test of fit for SEM, but, its sensitivity to discrepancies from expected values at increasing sample sizes can be highly problematic if those discrepancies are considered trivial from an explanatory-theory perspective. On the other hand, suitably scaled indices of approximate fit do not possess this sensitivity to sample size, but neither are they "tests" of model fit. The proposed solution to this dilemma is to consider the substantive "consequences" of accepting one explanatory model over another in terms of the predictive accuracy of theory-relevant-criteria. If there are none to be evaluated, then it is proposed that no scientifically worthwhile distinction between "competing" models can thus be made, which of course begs the question as to why such a SEM application was undertaken in the first place. (c) 2006 Elsevier Ltd. All rights reserved.