Ask Mallard

Mixed effects or GEE?

Match the interpretation of a correlated-outcome model to the question.

The decision

Repeated observations and participants within centers create dependence. Both mixed models and generalized estimating equations can represent that dependence, but the intended effect and the role of between-person or between-center variation should guide the choice.

What a mixed model emphasizes

A mixed model specifies random effects and their distribution. It can estimate between-cluster variability and support conditional comparisons or cluster-specific predictions. In a random-intercept logistic model, the fixed-effect odds ratio is conditional on that random intercept. Estimation, convergence, random-effect structure and distributional assumptions all need assessment.

What GEE emphasizes

GEE specifies a marginal mean model and a working correlation. With suitable regularity and enough independent clusters, a sandwich variance can protect uncertainty against a misspecified working correlation. It does not repair a wrong mean model or unmeasured confounding, and it does not estimate a random-effect distribution.

Interpretation before software

  • For nonlinear links, conditional and marginal coefficients generally differ. The binary logistic example makes this distinction especially visible.
  • For continuous outcomes with an identity link, fixed-effect interpretations can be closer under suitable models; do not generalize the logistic distinction to every mixed model.
  • Few clusters make ordinary sandwich inference unreliable. Name an appropriate small-sample strategy or obtain specialist advice; a large participant count does not supply more independent clusters.
  • Missing outcomes require their own assumptions. Standard GEE and likelihood-based mixed models do not automatically make the same missing-data assumptions; weighted GEE or imputation may be needed.

Cluster curves and the population curve

Several steep S-shaped curves, one per cluster, and a thicker population-average curve that is flatter, because averaging steep logistic curves spreads them out. This holds for a logistic link; with an identity link the two can coincide.Subject-specific curves and the population-average curveExposureProbability of the outcomePopulation average: flatterEach cluster: steepWith a logistic link, averaging steep cluster-specific curves gives a flatter population curve, so a mixedmodel's odds ratio and a GEE odds ratio answer different questions. With an identity link the two cancoincide.
Conceptual diagram. It shows the structure of the idea, not data from any study.

Example

For patients nested in clinics, an investigator may want a population-average association useful for population planning, or a conditional association plus clinic heterogeneity. Agree on that interpretation first, then choose a model and a defensible variance calculation.

Record in the workspace

Record the independent clustering unit, numbers and sizes of clusters, outcome family, link, conditional or marginal target, correlation assumptions, missingness approach and small-sample limits. Confirm the interpretation before requesting code.

What to watch out for

For a binary outcome, a cluster-conditional odds ratio from a mixed model and a population-average odds ratio from GEE are different estimands. One is not a drop-in replacement for the other.

Recent studies that used this
Searching PubMed…
Related Learn pages
Sources & further reading
  1. Models for longitudinal data: a generalized estimating equation approach (opens in a new tab) · Biometrics
  2. Real longitudinal data analysis for real people: building a good enough mixed model (opens in a new tab) · Stat Med · doi 10.1002/sim.3775
  3. Repeated Measures Designs and Analysis of Longitudinal Data: If at First You Do Not Succeed-Try, Try Again. (opens in a new tab) · Anesth Analg · doi 10.1213/ANE.0000000000003511
Prepared with AI assistance. Methodological and source checks are documented in this release; this is educational material, not independent professional approval. Reporting guidance supports transparent reporting and does not establish validity.
Last source review: 2026-09-09
AI-generated educational content
Mallard uses reputable methodological and reporting sources, but this material may contain errors or omit important context. Verify important methodological decisions with the cited sources and, when appropriate, a qualified research methodologist or statistician.