Key research themes
1. How can identifiability and parameter estimation challenges in two-factor models, especially Bayesian and state-space formulations, be effectively addressed to improve inference and interpretability?
This research focuses on methodological challenges in two-factor models related to parameter identifiability, multimodality in the Bayesian posterior, and estimation of latent factors and parameters jointly. These challenges impede interpretability and reliable inference, particularly in Bayesian factor analytic models and state-space applications such as commodity pricing. Addressing these issues improves factor interpretation, convergence of MCMC algorithms, and accurate calibration of two-factor models.
2. How are time-varying factor loadings and dynamic inference methods advancing the modeling and testing of two-factor models in large high-dimensional data?
This theme investigates methodologies for extending traditional two-factor (and general factor) models to allow factor loadings and dynamics to change over time, and how simultaneous statistical inference techniques can be developed and applied to such dynamic factor models, especially in large panel and time series contexts. Addressing time-variation in loadings captures structural changes and nonstationarities commonly observed in economic and financial datasets, improving model fidelity and inferential robustness.
3. What are the distinct modeling approaches and challenges in bifactor frameworks for psychological constructs, and how do they relate to interpretability and factor validity compared to two-factor or multifactor models?
This research area examines bifactor models applied to psychological constructs such as personality and psychopathology, highlighting challenges such as factor meaning ambiguity, anomalous loading patterns, and interpretational issues of general versus specific factors. It contrasts bifactor approaches with two-factor or higher-order models and proposes methodological approaches to clarify factor structure validity, interpretability, and meaningful bifactor decomposition. Findings are relevant for measurement theory and psychometric applications of multifactor latent variable models.