Confirmatory Latent Class Analysis: Illustrations of Empirically Driven and Theoretically Driven Model Constraints
Confirmatory Latent Class Analysis: Illustrations of Empirically Driven and Theoretically Driven Model Constraints
复制标题
DOI：
10.1177/1094428117747689
复制
发表时间：
2018-10-01
影响因子：
9.5
通讯作者：
Bryan, Angela D.
中科院分区：
文献类型：
--
作者：
Schmiege, Sarah J.;Masyn, Katherine E.;Bryan, Angela D.
大多数以人为中心的方法的应用程序都依赖于数据驱动的类枚举方法。随着以人为中心的分析在组织研究中越来越受欢迎，可能会寻求验证性方法来提供更严格的理论测试并使复制工作正式化。验证性潜在类分析(LCA)是通过设置建模约束来实现的，但已发表的工作中潜在约束的类型多种多样，在评估模型适合性方面缺乏标准化。本文为模型约束的可操作性提供了一个全面的框架，并通过两个例子展示了验证性生命周期评价：(A)双样本方法(在探索性和验证性样本中分别为n=1,366和n=1,367)和(B)假设的潜在班级结构的验证性测试(n=1,483)。我们描述了阈值边界和/或等式约束在两种图解下的可操作性，以生成确证的潜在类结构，并解释了模型评估的方法和与替代模型的比较。验证性模型在双样本方法下得到了很好的支持，在假设驱动方法下得到了部分支持。我们讨论了在模型估计的不同点上的决策，并以未来方法论的发展结束。
Most applications of person-centered methodologies have relied on data-driven approaches to class enumeration. As person-centered analyses grow in popularity within organizational research, confirmatory approaches may be sought to provide more stringent theoretical tests and to formalize replication efforts. Confirmatory latent class analysis (LCA) is achieved through placement of modeling constraints, yet there is variation in the types of potential constraints and a lack of standardization in evaluating model fit in published work. This article provides a comprehensive framework for operationalizing model constraints and demonstrates confirmatory LCA via two illustrations: (a) a dual sample approach (n = 1,366 and n = 1,367 in exploratory and validation samples, respectively) and (b) confirmatory testing of a hypothesized latent class structure (n = 1,483). We depict operationalization of threshold boundary and/or equality constraints under both illustrations to generate a confirmatory latent class structure, and explain methods of model evaluation and comparison to alternative models. The confirmatory model was well supported under the dual sample approach, and partially supported under the hypothesis-driven approach. We discuss decision making at various points of model estimation and end with future methodological developments.