Generalized estimating equations
What question does this help answer?
“What is the average relationship between predictors and an outcome across the population when observations within people or clusters are correlated?”
When to use it
Use GEE for repeated-measures or clustered outcomes when the primary estimand is population-average and the number of independent clusters is adequate.
When not to use it
If subject-specific effects or the distribution of random effects are central, a mixed-effects model may better match the scientific objective. Small numbers of clusters can make standard robust variance estimates unreliable without corrections.
How it works in plain language
GEE extends generalized linear models by adding a working correlation structure for observations within the same cluster. Robust standard errors can provide valid large-sample inference even when the working correlation is not perfectly specified under standard conditions.
Why Mallard may recommend it
Mallard may recommend GEE for repeated binary, continuous, or count outcomes when the research question is about an average population effect.
Key assumptions
- Clusters are independent of one another under the analysis structure.
- The mean model and link function are appropriate.
- The number of clusters is sufficient for the inferential approach.
- The working correlation and variance estimator are chosen thoughtfully.
- Missingness and dropout are considered.
What to watch out for
GEE and mixed-effects models answer related but not identical questions. Their coefficients can have different interpretations, especially for nonlinear links such as logistic regression.
Clinical example
Estimate the population-average association between a clinic intervention and quarterly vaccination status while accounting for repeated observations within patients.
- Repeated Measures Designs and Analysis of Longitudinal Data (opens in a new tab) · Anesth Analg · doi 10.1213/ANE.0000000000003511
- Tutorial on Biostatistics: Longitudinal Analysis of Correlated Continuous Eye Data (opens in a new tab) · Ophthalmic Epidemiol · doi 10.1080/09286586.2020.1786590