Comparing Mixed Models and Structural Equation Models for Analyzing Longitudinal Data
Author Information
Author(s): Emily A. Blood, Debbie M. Cheng
Primary Institution: Children's Hospital Boston and Harvard Medical School
Hypothesis
The study aims to evaluate the performance of linear mixed models (LMMs) relative to structural equation models (SEMs) in analyzing mediated longitudinal data with time-dependent predictors.
Conclusion
The study found that when appropriately constructed, LMMs can adequately model mediated exposure effects that change over time, similar to SEMs.
Supporting Evidence
- The study simulated longitudinal data to compare the performance of LMMs and SEMs.
- Results indicated that LMMs can effectively model mediated effects over time.
- Both modeling approaches yielded similar results in terms of bias and power when appropriately specified.
- The naive delayed effect mixed model showed significant bias, highlighting the importance of model choice.
Takeaway
This study looks at two ways to analyze data collected over time to see how one thing affects another, like how drinking alcohol might affect health. It found that both methods can work well if used correctly.
Methodology
The study used simulation studies to evaluate the performance of LMMs and SEMs under various conditions, including a real-data example from a cohort study on alcohol use and HIV disease progression.
Potential Biases
The naive delayed effect mixed model produced biased estimates, indicating the need for careful model selection.
Limitations
The results may not be generalizable to other settings with more complex pathways and relationships between variables.
Participant Demographics
The study analyzed data from a cohort of HIV-infected individuals, specifically focusing on those reporting any ART use during follow-up.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website