A Bayesian Saturated Model Approach to Posterior Predictive Model Checks in Confirmatory Factor Analysis
A Bayesian Saturated Model Approach to Posterior Predictive Model Checks in Confirmatory Factor Analysis
复制标题
验证性因素分析中后验预测模型检查的贝叶斯饱和模型方法
DOI：
10.1080/00273171.2019.1700773
复制
发表时间：
2020
影响因子：
3.8
通讯作者：
Mintz, Catherine E.
中科院分区：
文献类型：
--
作者：
Zhang, Jihong;Templin, Jonathan;Mintz, Catherine E.
图1.单因素模型中条目1和条目2之间相关性的后验预测模型检验。实心垂直线表示相关性的最大似然估计的位置。实线密度表示饱和模型生成的相关性的后验预测分布，虚线表示单因素模型生成的相关性的后验预测分布，虚线表示双因素模型(过度指定)生成的相关性的后验预测分布。当实心垂直线(MLE)远离正确模型和错误模型的后验预测分布的中心时，模型拟合较差。饱和模型和替代模型之间后验预测分布的重叠越大，表明模型的拟合程度越好。
Figure 1. Posterior predictive model check of correlation between Item 1 and Item 2 in one-factor model. The solid vertical line represents the location of the MLE of the correlation. The density with solid line represents the posterior predictive distribution of the correlation generated by the saturated model, dashed line represents the posterior predictive distribution of the correlation generated by the one-factor model, and dotted line represents the posterior predictive distribution of the correlation generated by the two-factor model (overspecified). When the solid vertical line (MLE) is far away from the center of the posterior predictive distribution of the correct model and the mis-specified model, the model fit is poor. Greater overlapping of the posterior predictive distribution between the saturated and the alternative models indicates better model fit.