Abstract

Under the framework of Latent Variable Modelling, three statistical methods, namely, the free baseline method (FR), the Benjamini-Hochberg method (B–H), and the alignment method (AM), were applied to identify the noninvariant measurement parameters at the indicator level through a simulation study. Model noninvariance was manipulated through varying the locations, degrees and magnitudes of noninvariant parameters with different sample sizes. As a result, the B-H method was discovered to possess a higher power rate in detecting noninvariant parameters than the other two methods, and the AM method performed the best in controlling the type I error rates in the process of parameter estimation. Comparatively speaking, the FR method performed the worst in almost all situations due to the choice of a fixed noninvariant reference indicator. The suggestions on which method is better used for model noninvariance detection was discussed and elaborated.

摘要

在潜变量建模的框架下，应用自由基线法（FR）、Benjamini-Hochberg法（B-H）和对齐法（AM）三种统计方法来识别非不变测量参数。通过模拟研究的指标水平。通过改变具有不同样本大小的非不变参数的位置、程度和大小来操纵模型的非不变性。结果发现，BH方法在检测非不变参数方面比其他两种方法具有更高的功率率，而AM方法在参数估计过程中对I类错误率的控制效果最好。相对而言，由于选择了固定的非不变参考指标，FR 方法在几乎所有情况下的表现最差。