Variable Selection by Regularized Multiple-Indicators Multiple-Causes Modeling With Missing Data: A Monte Carlo Simulation

• In Event: Advanced Statistical Methods for Complex Data

Abstract

In the era of proliferating data as well as analysis techniques, traditional analysis methods are often coupled with machine learning. In particular, regSEM (regularized structural equation modeling), a combination of SEM (a traditional method) and regularization (machine learning), has been proposed as a variable selection method in the framework of SEM. Specifically, this simulation study focused on regularized multiple-indicators multiple-causes (regMIMIC) in the presence of missing data. A total of 48 Monte Carlo simulation conditions included missingness mechanisms (MAR, MNAR), missing rates (5%, 30%), missing data techniques (listwise deletion, k-NN), sample sizes (500, 2000), and analysis methods (MLE, LASSO, MCP). Scientific importance of the study was discussed as well as further research topics.

Authors

• Minjeong Rho, Korea National University of Education

• Jin Eun Yoo, Korea National University of Education