Dynamic Structural Equation Models: Promising Yet Concerning (2023)

Undergraduate: Suryadyuti Baral

Faculty Advisor: Patrick Curran
Department: Psychology & Neuroscience

---

Dynamic Structural Equation Model (DSEM) is a powerful statistical modeling approach that has recently gained popularity among researchers studying intensive longitudinal data. Despite its exciting potential, the stability and replicability of DSEM are yet to be closely examined. The purpose of our study was to empirically investigate DSEM using recently published data to explore its strengths and potential limitations. Our results show that while some of its parameter estimates are stable, others are characterized by substantial variation as a function of seemingly innocuous initial model estimation conditions. Indeed, some parameters fluctuate between significance and non-significance for the same model estimated using the same data. The instability of DSEM estimates poses a serious threat to the internal and external validity of conclusions drawn from its analyses, challenging the reproducibility of findings from applied research. Given the recent focus on the replication crisis in psychology, it is critical to address these issues as the popularity of DSEM in psychological research continues to rise. We investigate several potential solutions to address this problem and offer recommendations of best practice to applied researchers who plan to use DSEM for causal inference in intensive longitudinal data analysis.

Link to Poster