Modern methods for longitudinal data analysis, capabilities, caveats and cautions

Lin Ge, Justin X. Tu, Hui Zhang, Hongyue Wang, Hua He, Douglas Gunzler*

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review


Summary: Longitudinal studies are used in mental health research and services studies. The dominant approaches for longitudinal data analysis are the generalized linear mixed-effects models (GLMM) and the weighted generalized estimating equations (WGEE). Although both classes of models have been extensively published and widely applied, differences between and limitations about these methods are not clearly delineated and well documented. Unfortunately, some of the differences and limitations carry significant implications for reporting, comparing and interpreting research findings. In this report, we review both major approaches for longitudinal data analysis and highlight their similarities and major differences. We focus on comparison of the two classes of models in terms of model assumptions, model parameter interpretation, applicability and limitations, using both real and simulated data. We discuss caveats and cautions when applying the two different approaches to real study data.

Original languageEnglish (US)
Pages (from-to)293-300
Number of pages8
JournalShanghai Archives of Psychiatry
Issue number5
StatePublished - Oct 1 2016


  • Binary variables
  • Correlated outcomes
  • Generalized linear mixed-effects models
  • Latent variable models
  • R
  • SAS
  • Weighted generalized estimating equations

ASJC Scopus subject areas

  • Neurology
  • Psychiatry and Mental health


Dive into the research topics of 'Modern methods for longitudinal data analysis, capabilities, caveats and cautions'. Together they form a unique fingerprint.

Cite this