Handling Missing Covariates in Conditional Mixture Models Under Missing at Random Assumptions.

Area of Science:

• Psychometrics
• Statistical modeling
• Developmental psychology

Background:

• Mixture modeling addresses unobserved population heterogeneity using latent classes.
• Conditional mixture models incorporate covariates but traditionally assume complete data.
• Existing methods like exogenous-x approach often use listwise deletion for missing covariates, reducing efficiency and potentially introducing bias.

Purpose of the Study:

• To present a modified joint likelihood approach for conditional mixture models with missing covariate data.
• To demonstrate the advantages of this new approach over the exogenous-x method.
• To provide a practical method for handling missing covariate data in mixture modeling.

Main Methods:

• A modified joint likelihood approach is proposed, allowing for nonnormal covariates.
• This method accommodates missing at random covariate data, retaining all cases.
• The approach is implemented and compared to the exogenous-x method.

Main Results:

• The modified joint likelihood approach yields lower bias and higher efficiency than the exogenous-x approach.
• It allows inference on parameters of the exogenous-x conditional mixture even with nonnormal covariates.
• The method is straightforward to implement in commercial statistical software.

Conclusions:

• The modified joint likelihood approach offers a superior alternative for mixture modeling with missing covariate data.
• This method enhances statistical power and reduces bias in psychological research.
• Recommendations for practical application in analyzing conduct problems are discussed.