Kline Rex B
Department of Psychology, Concordia University, Montréal, Canada.
Int J Psychol. 2024 Dec;59(6):1293-1306. doi: 10.1002/ijop.13252. Epub 2024 Oct 2.
Consistent with reporting standards for structural equation modelling (SEM), model fit should be evaluated at two different levels, global and local. Global fit concerns the overall or average correspondence between the entire data matrix and the model, given the parameter estimates for the model. Local fit is evaluated at the level of the residuals, or differences between observed and predicted associations for every pair of measured variables in the model. It can happen that models with apparently satisfactory global fit can nevertheless have problematic local fit. This may be especially true for relatively large models with many variables, where serious misspecification is indicated by some larger residuals, but their contribution to global fit is diluted when averaged together with all the other smaller residuals. It can be challenging to evaluate local fit in large models with dozens or even hundreds of variables and corresponding residuals. Thus, the main goal of this tutorial is to offer suggestions about how to efficiently evaluate and describe local fit for large structural equation models. An empirical example is described where all data, syntax and output files are freely available to readers.
与结构方程模型（SEM）的报告标准一致，模型拟合应在两个不同层面进行评估，即整体层面和局部层面。整体拟合关注的是在给定模型参数估计值的情况下，整个数据矩阵与模型之间的总体或平均对应程度。局部拟合是在残差层面进行评估的，也就是模型中每对测量变量的观测关联与预测关联之间的差异。可能会出现这样的情况：整体拟合看似令人满意的模型，其局部拟合却存在问题。对于具有许多变量的相对大型模型而言，情况可能尤其如此，在这类模型中，一些较大的残差表明存在严重的设定错误，但当它们与所有其他较小的残差一起平均时，其对整体拟合的影响就被稀释了。在具有数十甚至数百个变量及相应残差的大型模型中评估局部拟合可能具有挑战性。因此，本教程的主要目标是就如何有效评估和描述大型结构方程模型的局部拟合提供建议。文中描述了一个实证示例，所有数据、语法和输出文件均可供读者免费获取。
