Dynamic Structural Equation Models
Dynamic Structural Equation Model (DSEM) is a statistical method used to analyze data collected over time. It helps researchers understand how different variables interact and influence each other in longitudinal studies. DSEM is valuable for studying complex relationships of psychological constructs. For example, say you are interested in studying rejection and aggression in children. You might hypothesize that rejection from family and friends might lead to a child developing aggressive tendencies. However, it is also possible that the child’s aggressive behavior lead to rejection. Leveraging on intensive longitudinal data, DSEM allows researchers to test hypotheses on such lead-lag relationships between constructs as they co-develop over time. Thus, DSEM provides insights into the underlying dynamics of a system. It is widely used in various fields, including psychology, sociology, and economics, to uncover meaningful patterns and relationships in longitudinal data.
Under the guidance of Dr. Patrick Curran, I conducted research focusing on Dynamic Structural Equation Models (DSEM) during the PSYC 395 course in Fall 2022. This research involved analyzing longitudinal data and investigating inconsistencies in the model estimates. In my written report, I found that DSEM provided unstable parameter estimates, which can lead to incorrect conclusions. I presented these findings to the lab, consisting of three PhD students and my mentor, and they found the results interesting. During this research, I learned about Structural Equation Models and gained experience in coding statistical models using MPlus. In Spring 2023, I extended the work I had done last semester. I used MPlusautomate in R, to run DSEM’s estimator with 1000 different initial values. As I had observed last semester, the parameter estimates turned out to be unstable. Moreover, one of the parameters even switched from being non-significant to being significant depending on the initial value of the estimator. I tried improving the estimation by using more iterations of the estimator and stricter convergence thresholds. I believe my recommendations will allow many applied researchers avoid potential pitfalls of DSEM’s unstable estimation.
To share my work, I presented a research poster at UNC's Celebration of Undergraduate Research, as well as at UCLA's Psychology Undergraduate Research Conference. Additionally, I had the opportunity to give a 15-minute research talk at UVA's L. Starling Reid Psychology Undergraduate Research Conference in Spring 2023.
