Inclusive classify-analyze can overcome the underestimation of simple regression parameters by the traditional classify-analyze approach in subsequent analysis of latent profile model, and manages to simplify the estimation of interaction among variables. The current study aimed to extend the inclusive approach to some multiple regression analysis subsequent to latent profile model. Monte Carlo simulation study was conducted to investigate which variables should be included in the latent profile model, and whether interaction should be considered in the subsequent analytic model. Data was generated based on regression model of a binary distal outcome to a three-profile variable and a binary predictor. Four commonly-used model scenarios were considered: multiple regression model, and regression models with moderator, mediator or both. The results confirm the necessity of adding distal outcome to the measurement model, and also suggest that, variables related to or interacted with the latent profile variable should be included in the latent profile model.Failure to include these variables would induce attenuation of the estimated effects. In addition, we found model which include interactions term is more robust in the subsequent analysis.%纳入式分类分析法能克服传统的分类分析法对后续一元回归模型参数的低估,发挥潜在类别模型的后续分析简化变量间交互作用的功能.本文进一步将纳入式分类分析法拓展至潜在剖面模型后续的多元统计分析中.通过蒙特卡洛模拟实验,比较各种纳入变量的方法思路与后续分析模型在四种常见的多元回归模型中参数估计的表现.结果发现,纳入式分类分析法所需纳入的变量取决于后续分析中与因变量、潜类别变量的关系,且后续分析使用含交互作用的模型更为稳健.
展开▼
