Factor Tree Copula Models for Item Response Data.

Area of Science:

• Statistics
• Psychometrics
• Machine Learning

Background:

• Factor copula models offer interpretability for item response data but struggle with conditional independence violations.
• Truncated vine copula models handle complex dependencies but can lack interpretability.
• Existing models face limitations in balancing interpretability and capturing residual dependencies.

Purpose of the Study:

• To introduce a novel factor tree copula model for item response data.
• To combine the strengths of factor copula and truncated vine copula models.
• To develop robust methods for modeling complex dependencies in item response data.

Main Methods:

• A hybrid model, the factor tree copula, is proposed, integrating factor and truncated vine structures.
• A truncated vine structure is applied to residuals conditional on latent variables.
• Model selection algorithms are developed for choosing appropriate factor tree copula models.

Main Results:

• The factor tree copula model demonstrates improved interpretability and fit compared to individual approaches.
• The model effectively captures residual dependencies while maintaining parsimony.
• Simulation studies confirm the methodology's validity and performance.

Conclusions:

• The factor tree copula model provides a powerful and flexible framework for item response data analysis.
• This approach offers a robust solution for handling complex dependence structures.
• The model is effectively illustrated through the analysis of Post-Traumatic Stress Disorder data.