Path and Direction Discovery in Individual Dynamic Factor Models: A Regularized Hybrid Unified Structural Equation...

Area of Science:

• Statistics
• Econometrics
• Psychometrics

Background:

• Multivariate time series data often contain measurement error, necessitating advanced modeling techniques.
• Dynamic Factor Models (DFM) combine latent factors with Vector Autoregressive (VAR) models to capture complex temporal dependencies.
• Existing DFMs have limitations in modeling both directed and undirected contemporaneous relations and in optimal model selection.

Purpose of the Study:

• To address limitations in current Dynamic Factor Models (DFM) for time series data with measurement error.
• To propose a novel DFM framework incorporating hybrid VAR representations and LASSO regularization.
• To provide guidance on model selection and estimation for person-centered dynamic assessments.

Main Methods:

• Development of a hybrid VAR representation for DFMs to capture diverse dynamic relationships.
• Utilization of LASSO regularization for selecting dynamic implied instrumental variables.
• Application of a two-stage least squares (MIIV-2SLS) estimation strategy for robust parameter estimation.

Main Results:

• The proposed method offers enhanced flexibility in modeling the directions of dynamic relations within multivariate time series.
• The LASSO regularization aids in selecting appropriate instrumental variables, improving model stability.
• The MIIV-2SLS estimation provides a robust approach for DFM with measurement error.

Conclusions:

• The advanced DFM framework provides a more comprehensive approach to modeling multivariate time series with measurement error.
• The proposed methodology enhances the ability to accurately capture complex dynamic interdependencies.
• This research offers valuable tools and guidance for researchers in dynamic assessment and related fields.