A Practical Guide to Variable Selection in Structural Equation Models with Regularized MIMIC Models.

Area of Science:

• Behavioral Science
• Psychology
• Statistics

Background:

• Sophisticated statistical models are increasingly used for complex behavioral data.
• Small sample sizes can limit the application of complex statistical models.
• Regularization is a method to penalize model complexity, effective in small n, large p settings.

Purpose of the Study:

• To integrate regularization within structural equation models (SEM).
• To evaluate the performance of regularized SEM in small sample, large predictor situations.
• To demonstrate the application of regularized SEM in empirical research.

Main Methods:

• Extending regularization principles from linear regression to SEM.
• Conducting a simulation study to compare regularized SEM with traditional SEM.
• Applying regularized SEM to empirical datasets in neuroscience and mental health.

Main Results:

• Regularized SEM outperforms traditional SEM estimation when dealing with many predictors and small sample sizes.
• The simulation study validates the effectiveness of the proposed regularization approach.
• Empirical examples showcase the practical utility of regularized SEM.

Conclusions:

• Regularization offers a viable solution for parameter estimation in SEM with limited observations and numerous predictors.
• Regularized SEM provides a powerful alternative to traditional methods in challenging data scenarios.
• The approach is applicable to diverse research areas, including neuroscience and mental health research.