Cluster Randomized Trial
Cluster Randomized Controlled Trial (CRT) · Also known as: CRT, cluster RCT, cluster trial, group randomization
A cluster randomized trial (CRT) randomizes intact groups—schools, clinics, villages, or hospital wards—rather than individuals. Developed by Campbell, Grimshaw, and colleagues in the late 1990s to address real-world settings where intervention delivery or contamination occurs at the group level, CRTs are now standard for evaluating population-level, community-based, and policy interventions.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use CRTs when: (1) the intervention is delivered at the cluster level (e.g., staff training, policy change), (2) contamination between groups is likely (shared workplaces, adjacent classrooms), (3) randomizing individuals within clusters is impractical or unethical (e.g., randomizing some teachers but not others within a school), (4) evaluating community health programs, educational initiatives, or organizational changes, (5) studying rare outcomes in large populations (clusters stratified by geography or demographic), (6) assessing public health or policy-level interventions.
Strengths & limitations
- Reflects real-world delivery: interventions often target groups, not individuals, so CRTs test feasibility and effectiveness in practice.
- Avoids contamination bias: control clusters are isolated from intervention, reducing the risk of spillover effects.
- Ethical and practical: many interventions (e.g., school curricula, community campaigns) cannot be randomized at the individual level.
- Policy relevance: results inform decisions about rolling out population-level changes across multiple sites.
- Facilitates implementation: organizations randomized at the cluster level may show higher engagement than individuals randomized separately.
- Large sample size: design effects inflate required N, often 2–10 times an individual RCT; costs rise substantially.
- Loss of statistical power: clustering reduces effective sample size; fewer independent observations (clusters) than total participants.
- ICC uncertainty: if ICC is misestimated at the design phase, sample size and power calculations become inaccurate. ICC varies by outcome, setting, and population.
- Fewer clusters = higher Type I error: when the number of clusters is small (<10–20 per arm), standard tests may not maintain nominal alpha. Use Satterthwaite or cluster bootstrap approaches.
- Generalization limits: results generalize to clusters similar to those in the trial; effects in new cluster types may differ.
Frequently asked
What is the intracluster correlation coefficient (ICC), and why does it matter?
The ICC is the proportion of total outcome variance that lies between clusters (as opposed to within clusters). A higher ICC means individuals within a cluster are more similar to each other, so fewer independent data points are truly available. ICC directly drives the design effect: larger ICC requires larger sample sizes. ICC is estimated from pilot data, prior literature, or conservative assumptions (typically 0.01–0.1 for health outcomes). Ignoring ICC in sample size calculations is a common source of underpowering.
Why can't I just randomize individuals and account for clustering in the analysis?
Individual randomization within clusters introduces structural imbalance: treatment and control individuals share the same cluster environment, causing contamination (spillover) and violating independence. Moreover, randomizing individuals but delivering cluster-level interventions (e.g., training all staff in a clinic) means every individual in the cluster gets the intervention regardless of randomization status. CRT avoids these by randomizing clusters as the unit, ensuring true independence and proper alignment between randomization unit and intervention unit.
How do I analyze a cluster randomized trial?
Use mixed-effects models (hierarchical linear models) with cluster as a random intercept, or Generalized Estimating Equations (GEE) with an exchangeable working correlation structure. Both account for within-cluster dependence. Cluster-level fixed effects can also be used if the number of clusters is large (>30). Standard t-tests, logistic regression, and ANOVA applied to cluster-level means are acceptable for simple comparisons but lack efficiency. Always report the ICC and design effect. For binary outcomes, logistic mixed models or log-binomial models apply.
How many clusters do I need?
The number of clusters is more important than the number of individuals per cluster. Aim for at least 10–20 clusters per arm if possible; fewer clusters increase Type I error (false positives). For small numbers of clusters (<10), use exact permutation tests, Satterthwaite-adjusted degrees of freedom, or cluster bootstrap confidence intervals. If you have only 3–5 clusters per arm, consider a cluster crossover or stepped-wedge design to increase power. The design effect formula includes cluster count implicitly: fewer clusters inflate the effective variance, requiring either more clusters or larger effect sizes to detect.
Sources
- Campbell, M. K., Grimshaw, J. M., & Elbourne, D. R. (2000). Intracluster correlation coefficients in cluster randomized trials: empirical insights into how should they be reported. BMC Medical Research Methodology, 4, 30. link ↗
- Eldridge, S. M., Ashmore, S., Frenkel, S., Cryer, C., & Underwood, M. (2006). Uncertainty in analyses of safety after cluster randomization. Clinical Trials, 3(2), 152–162. link ↗
- Campbell, M. K., Piaggio, G., Elbourne, D. R., & Altman, D. G. (2012). Consort 2010 statement: extension to cluster randomised trials. BMJ, 345, e5661. DOI: 10.1136/bmj.e5661 ↗
How to cite this page
ScholarGate. (2026, June 4). Cluster Randomized Controlled Trial (CRT). ScholarGate. https://scholargate.app/en/clinical-research/cluster-randomized-trial
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Adaptive Trial DesignClinical Research↔ compare
- Pragmatic Clinical TrialClinical Research↔ compare
- Randomized Controlled TrialExperimental design↔ compare