Ask Mallard
Analysis method

Cluster-level summary analysis

4 min read

A cluster-level summary analysis compares one outcome summary per randomized cluster, keeping the cluster as the independent unit.

Fit

When to use it, and when not to

Use it when

  • Treatment was randomized to independent clusters and a meaningful summary can be computed for each.
  • A simple cluster-level comparison fits the target effect and allocation scheme.

Not when

Before you start

What you need

Decide these first
  1. The randomized cluster and allocation schemeKeep matching and stratification in the analysis when used.
  2. A cluster outcome summaryChoose a mean, proportion or rate that answers the question.
  3. The target and weighting ruleState whether clusters or participants receive equal weight.
  4. A missing-participant and missing-cluster planState the missingness assumption before computing summaries.
It rests on these
  1. The independent units are clusters.How to check: Check spillovers and shared higher-level groups.
  2. The uncertainty method fits the number of clusters.How to check: For a t comparison, assess the summary distribution and use cluster-based degrees of freedom.
  3. The weights match the estimand.How to check: Assess unequal or outcome-related cluster sizes; more participants do not create more independent clusters.
  4. Missing data follow the stated assumption.How to check: Under missing at random, retain recorded predictors; assess departures and whole-cluster loss separately.

Plain language

How it works

Compute a prespecified outcome summary for every cluster. Compare those summaries between randomized groups, respecting the allocation scheme. Weighting and covariate adjustment should match the target effect.

The analysis structure
Compute one outcome summary per cluster; then Choose weights for the target effect; then Compare groups using independent clusters; then Use cluster-based uncertainty and degrees of freedom.Cluster-level summary analysisCompute one outcome summary per clusterChoose weights for the target effectCompare groups using independent clustersUse cluster-based uncertainty and degrees of freedom
Conceptual diagram. It shows the structure of the idea, not data from any study.
What the answer means

Kerry and Bland demonstrate cluster-level comparison; the eCRT trial applies it to prescribing proportions. The simple example uses equal cluster sizes and an unpaired t interval. It does not justify that interval for skewed summaries, matched clusters or very uneven sizes.

Worked example

A worked example

Invented for illustration
Setting
Eight clinics are randomized equally to a prescribing intervention or usual care, with 20 eligible consultations per clinic.
Question
What is the difference between average clinic prescribing proportions?
Approach
Intervention clinic counts are 9, 10, 11 and 12; control counts are 12, 13, 14 and 15. Weight clinics equally.
Result
The control mean is 67.5% and intervention mean is 52.5%, a difference of -15 percentage points. The 95% t interval is -26.17 to -3.83 points.
What it means
The interval uses 6 degrees of freedom under equal-variance, normal-summary assumptions. Eight clinics remain the independent units, despite having 160 consultations.

In practice

Recent studies PubMed search · not vetted

Searching PubMed…

Watch out

Common mistakes

The one to remember

Weighting clusters equally targets an average cluster; unequal cluster sizes, few clusters and matched allocation need deliberate choices about estimand and inference.

  • Analyzing each participant as independent.Instead: Base the comparison and degrees of freedom on clusters.
  • Weighting by cluster size without naming the target.Instead: Choose and justify weights before looking at effects.

Inside Mallard

How Mallard uses this

The method registry links this method to its published analysis article and code. Learn search and the Atlas offer it as a reference.

Code example in Mallard: Cluster-level summary analysis for few clusters (cluster-level-summary-analysis).
Atlas node analysis.cluster_summary, the id a plan cue records.

How to report it

Report the cluster counts, summaries, weighting target, allocation adjustments, missingness assumptions, cluster-based degrees of freedom and confidence interval. A reporting guideline is not a design manual: use it to make the study transparent, not as a substitute for careful design. Check the target journal's own requirements as well, which this page cannot know.

Code for this method
Loading…

Related Learn pages

Related designs and methods

Evidence

Sources & further reading

  1. Analysis of a trial randomised in clusters (opens in a new tab) · BMJ · doi 10.1136/bmj.316.7124.54
  2. Electronic health records for intervention research: a cluster randomized trial to reduce antibiotic prescribing in primary care (eCRT study) (opens in a new tab) · Ann Fam Med · doi 10.1370/afm.1659
Prepared with AI assistance. PubMed titles, authors, years and DOIs checked on 2026-11-07. Record verified, confirm it supports this use. No independent professional approval is claimed.
Last source review: 2026-11-07
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.