QuantDev Brown Bag Seminar Series (In-Person)

Google map

Measurement Model Performances in Dynamic Structural Equation Models  

Abstract: 

Dynamic structural equation models (DSEMs), a special case of which is the multilevel dynamic factor models, are a powerful tool for analyzing intensive longitudinal data (ILD). Researchers can analyze the individual dynamics of latent factors identified with manifest indicators in DSEMs framework. Due to their versatility by integrating techniques from multilevel modeling, time series analyses, and structural equation modeling, Bayesian DSEMs are becoming increasingly popular. However, measures used in ILD studies are susceptible to measurement error due to their characteristics of data collection, underscoring the importance of psychometric soundness, such as reliability. On top of that, many applications of DSEMs utilize composite scores without accounting for measurement errors and differences in indicator quality. Therefore, the goal of this study is to provide a framework for model performances including power and sample size planning in three different DSEMs under different conditions: 1) single-indicator model without incorporating measurement error, 2) measurement model with single-indicator, and 3) measurement model with multiple indicators. In pursuit of this aim, we provide results from a Monte Carlo simulation study with four manipulated factors: sample size, the number of repeated measurement occasions, the extent of reliability, and the number of indicators in the model. Consequently, we will provide guidance on the conditions under which specific models would be suitable for use and discuss future directions.

Keywords: Dynamic Structural Equation Modeling, Measurement Error, Within-person Dynamic Process, Intensive Longitudinal Data