Eliminating Bias in Classify-Analyze Approaches for Latent Class Analysis.

Area of Science:

• Behavioral Science
• Statistics
• Psychometrics

Background:

• Latent class analysis (LCA) is increasingly used in behavioral research.
• Traditional LCA methods can lead to biased estimates when class membership is treated as known in subsequent analyses.

Purpose of the Study:

• To propose and evaluate a more inclusive latent class analysis (LCA) strategy.
• To reduce bias in analytic models that include latent class variables.

Main Methods:

• The proposed method involves incorporating additional variables into the LCA model to generate more accurate posterior probabilities.
• An empirical demonstration and a simulation study were conducted to assess the strategy's performance.

Main Results:

• The inclusive LCA strategy effectively reduces or eliminates bias in parameter estimates.
• The effectiveness of the strategy is dependent on sufficient measurement quality or sample size.

Conclusions:

• The proposed inclusive LCA method offers a more accurate approach for analyzing complex research questions in behavioral science.
• This strategy mitigates bias issues inherent in traditional LCA applications.