Stepwise Latent Class Models for Explaining Group-Level Outcomes Using Discrete Individual-Level Predictors.

Area of Science:

• Multilevel modeling
• Latent variable analysis
• Quantitative psychology

Background:

• Explaining group outcomes requires aggregating individual data and correcting for measurement errors.
• Discrete variables pose challenges for standard aggregation and error correction methods.
• Existing methods lack clear approaches for discrete individual-level predictors in multilevel analyses.

Purpose of the Study:

• To present a novel stepwise latent class analysis (LCA) approach for aggregating discrete individual-level predictors to the group level.
• To correct for measurement errors in aggregated group-level variables derived from discrete individual data.
• To apply and evaluate this method within multilevel mediation models.

Main Methods:

• Estimating a latent class model using individual-level discrete predictor scores to form group-level latent classes.
• Aggregating the individual-level predictor by assigning groups to these latent classes.
• Conducting group-level analysis with error correction for class assignments, applied to multilevel mediation models.

Main Results:

• The stepwise LCA approach effectively aggregates discrete individual data for group-level analysis.
• The method corrects for measurement error in aggregated group-level variables.
• Simulation studies demonstrate the approach's utility and compare it to existing methods.

Conclusions:

• Stepwise latent class analysis provides a robust framework for handling discrete individual-level predictors in multilevel research.
• This method enhances the accuracy of explaining group-level outcomes.
• The approach is applicable to complex mediation models with multiple individual and group-level variables.