Longitudinal Data Analysis Using Generalized Estimating Equations

Longitudinal data arise whenever repeated measurements are taken on the same experimental units over time or under varying conditions. Generalized estimating equations (GEE) provide a popular marginal modelling framework for analysing such correlated outcomes without requiring full specification of their joint distribution. By focusing on the mean response as a function of covariates and employing a working correlation structure to approximate within-subject dependence, GEE yields consistent parameter estimates even if the correlation is misspecified, provided the mean model is correctly specified. The approach accommodates a wide range of response types—binary, count or continuous—via appropriate link and variance functions drawn from the exponential family. Robust “sandwich” variance estimators guard against model misspecification, enhancing inferential validity in practical settings. Applications span epidemiology, clinical trials and social science, where population-average effects are often of primary interest. Over the past two decades, accessible software implementations and methodological refinements have cemented GEE as a cornerstone of longitudinal analysis, facilitating flexible handling of time-varying covariates, missing data mechanisms and complex study designs.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

Recent methodological innovations have extended the flexibility and predictive power of GEE. A novel algorithm integrates gradient boosting within the GEE framework, randomly subsampling covariates at each step to reduce overfitting and better capture hierarchical dependence. Simulation studies demonstrate that this Generalized Estimating Equations Boosting (GEEB) machine outperforms standard GEE and popular machine-learning competitors in both predictive accuracy and error reduction across varied correlation structures. Building on this, a new “sandwich boosting” approach replaces fixed working-correlation weights with data-driven functions learned via modern gradient boosting. This method minimises a bespoke sandwich loss, yielding asymptotically efficient estimates under mild conditions and robust performance when parametric covariance forms are misspecified. Foundational software developments continue to underpin applied uptake: one of the earliest comprehensive implementations remains a gold-standard package offering user-defined link and variance functions, sparse-matrix algorithms and fast analytic inversion of working correlations, thereby enabling practitioners to fit large-scale longitudinal models entirely within a high-level programming environment.

Longitudinal Data Analysis Using Generalized Estimating Equations publication trend

The graph below shows the total number of articles in longitudinal data analysis using generalized estimating equations across all publications each year (not limited to Nature Index journals).

Technical terms

Longitudinal data: Observations collected repeatedly on the same units over time or conditions, inducing within-subject correlation.

Generalized estimating equations (GEE): A semi-parametric marginal modelling approach that estimates population-average effects while accounting for correlation via a working correlation matrix.

Working correlation structure: A user-specified matrix form (eg, independence, exchangeable, autoregressive) that approximates within-cluster dependence to improve efficiency.

Sandwich estimator: A robust variance estimator that remains consistent even if the working correlation or variance function is misspecified.

Quasi-likelihood: An extension of likelihood theory used in GEE to derive estimating equations in the absence of a fully specified joint distribution.

References

1. Generalized Estimating Equations Boosting (GEEB) machine for correlated data. Journal of Big Data (2024).
2. Sandwich boosting for accurate estimation in partially linear models for grouped data. Journal of the Royal Statistical Society Series B Statistical Methodology (2024).
3. The R Package geepack for Generalized Estimating Equations. Journal of Statistical Software (2006).