Selection Between Linear Factor Models and Latent Profile Models Using Conditional Covariances

Area of Science:

• Multivariate statistics
• Psychometrics
• Latent variable modeling

Background:

• Distinguishing between linear factor models and latent profile models is crucial for accurate data interpretation.
• Existing methods may lack the precision to differentiate between these related but distinct statistical structures.

Purpose of the Study:

• To propose a novel statistical method for selecting between K-dimensional linear factor models and (K+1)-class latent profile models.
• To provide a practical inferential approach for model discrimination based on conditional covariance properties.

Main Methods:

• The core of the method relies on the differential behavior of conditional covariances: constant under factor models, nonlinear under latent profile models.
• A data simulation approach was employed to evaluate the performance and error control of the proposed inferential method.

Main Results:

• The proposed method effectively differentiates between linear factor models and latent profile models.
• The inferential procedure exhibits acceptable error rate control in model selection tasks.
• Conditional covariances serve as a key diagnostic for distinguishing between the two model types.

Conclusions:

• The developed method offers a statistically sound and practical tool for researchers in psychology and vocational assessment.
• Accurate model selection enhances the validity of findings derived from latent variable analyses.
• The study highlights the importance of examining conditional covariance structures for model identification.