Longitudinal Models · GEE-2G-M

Sample Size for Two-Group GEE: Difference in Mean Response

Plan a population-averaged Wald test of an equally weighted comparison-minus-reference mean difference across follow-up visits.

1. Planning setup

Include a baseline measurement

Baseline receives zero weight in the average follow-up contrast.

Working correlation structure

Follow-up measurements per participant

Time from baseline to final follow-up

Follow-ups are equally spaced.

2. Information obtained from previous studies or pilot data

Marginal outcome standard deviation

Assumed common across groups and visits.

3. Inputs decided by the researcher for the new study

Smallest clinically important average mean difference across follow-ups

Equally weighted population-average comparison-group minus reference-group mean. Negative values are accepted.

Comparison-to-reference allocation ratio

Type I error, alpha (%)

Alternative hypothesis

Power (%)

Anticipated participant dropout/non-response (%)

4. Results

Enter the marginal design assumptions and select Calculate.

Formula, assumptions, limitations, and methodological note

Example clinical scenario

A randomized chronic-pain trial records a continuous symptom score at baseline and monthly for four months. Investigators define the primary estimand as the equally weighted population-average treatment difference across all four follow-up visits rather than at one particular month.

Formula and methodological note

For marginal covariance V = A½R(ρ)A½, categorical-visit model-based GEE information is D′V⁻¹D. The target contrast gives each follow-up comparison-minus-reference coefficient weight 1/m and baseline weight zero. Its allocation-unit variance is c′M⁻¹c. Exact allocation units are searched until normal-theory Wald power reaches the target.

This version assumes correctly specified model-based planning covariance, a continuous outcome, common marginal SD, equally spaced visits, independent participants, and randomized groups for common-baseline adjustment. Robust analysis does not remove planning sensitivity. Dropout is inflation only. Custom weights, binary outcomes, irregular schedules, visit-specific variances, informative missingness, extra clustering, and small-sample sandwich corrections are outside scope. This tool supports research planning and does not replace statistical consultation.

Technical results

Method references

Rochon J. Application of GEE procedures for sample size calculations in repeated measures experiments. Stat Med. 1998;17:1643–1658.
Liang KY, Zeger SL. Longitudinal data analysis using generalized linear models. Biometrika. 1986;73:13–22.
Statsmodels developers. Generalized Estimating Equations and NormalIndPower documentation.

Suggested citation

Yaacob, Najib Majdi. “Sample Size for Two-Group GEE: Difference in Mean Response (GEE-2G-M).” Web calculator. NajibMY, 2026. Accessed August 1, 2026. https://najibmy.netlify.app/tools/sample-size/clinical-complex-study-designs/gee-difference-in-mean-response/.