*New* CSSS Seminar: Copula-Based Latent Variable Modeling for Asymmetric Dependence (10/14/26)
Posted: 2026-09-25 11:28:44 ()
Latent variable models such as item response theory (IRT) models and structural equation modeling (SEM) are widely used in social science. Traditional latent variable models heavily rely on multivariate Gaussian assumptions to model dependence among multiple latent variables. However, the Gaussian dependence structures are limited to linear, symmetric, and tail-independent relationships. In this talk, I present two studies that address these limitations by incorporating asymmetric copulas to model tail dependence between latent variables in IRT and SEM. The first study introduces a copula-based joint model for speed and accuracy within the IRT framework. This approach improves the estimation accuracy of latent ability by jointly modeling item responses and response times using asymmetric copulas. The second study extends the copula-based approach to SEM with a focus on binary risk outcome prediction. While standard Gaussian models may underestimate joint risk probabilities, this study demonstrates that modeling tail dependence via asymmetric copulas can improve the prediction of rare risk outcomes such as suicide attempts. Overall, these studies advance latent variable modeling by incorporating asymmetric copulas to capture tail dependence. Specifically, the proposed approaches improve (1) the measurement accuracy of latent traits, (2) the overall model fit to empirical data, and (3) the prediction of rare risk outcomes. Meeting details can be found at this link.
Sunbeom Kwon is a Postdoctoral Scholar in the College of Education at the University of Washington. He earned his Ph.D. in Quantitative Psychology and M.S. in Applied Statistics from the University of Illinois Urbana-Champaign.
Date: 10/14/2026
Time: 12:30-1:30 PM
Deadline: 10/14/2026
Location: Savery 409