ABSTRACT

In confirmatory factor analysis (CFA), post hoc model modification (PMM) indexes are often used to adjust for possible residual correlations between items. Although the approach is useful for improving model goodness-of-fit, it requires an iterative, one-item-pair-at-a-time procedure that can be tedious and prone to error. This paper provides a didactic discussion in the form of a tutorial of a more efficient and practical alternative and its implementation using an R-based package. The tutorial contains (1) the Bayesian covariance Lasso (least absolute shrinkage and selection operator) approach as an alternative to the PMM method, and (2) the R package blcfa, which implements the Bayesian covariance lasso and directly interfaces with Mplus. It adopts a two-step approach by first estimating the entire residual covariance matrix, and then identifying the nonzero entries and seamlessly feeding them into Mplus. Two examples were used to illustrate package implementation.

摘要

在验证性因子分析 (CFA) 中，事后模型修正 (PMM) 指数通常用于调整项目之间可能的残差相关性。尽管该方法对于提高模型的拟合优度很有用，但它需要一个迭代的、一次一个项目对的过程，这可能很乏味且容易出错。本文以更有效和实用的替代方法及其使用基于 R 的包的实现的教程的形式提供了教学讨论。本教程包含（1）贝叶斯协方差套索（最小绝对收缩和选择算子）方法作为 PMM 方法的替代方法，以及（2）R 包blcfa，它实现了贝叶斯协方差套索并直接与 M plus接口. 它采用两步法，首先估计整个残差协方差矩阵，然后识别非零项并将它们无缝地输入 M plus。使用两个示例来说明包的实现。