Cluster Randomized Laboratory Experiment
Also known as: cluster-randomized lab experiment, group-randomized laboratory study, cluster RCT laboratory variant, clustered lab trial
A cluster randomized laboratory experiment assigns intact groups — such as lab sections, cohorts, or naturally formed teams — rather than individual participants, to experimental conditions. All participants within a cluster receive the same treatment. The design is used when individual randomization would cause contamination between conditions, while retaining the controlled environment of a laboratory setting.
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 a cluster randomized laboratory experiment when treatment cannot be administered independently to individuals without contamination risk, when the intervention targets a group-level process such as team decision-making or classroom interaction, or when logistical constraints make individual randomization infeasible in a lab setting. It is especially valuable in educational psychology, organizational behavior, and behavioral economics experiments run in cohort-based lab formats. Do NOT use it when individual randomization is feasible and contamination is not a concern — unnecessary clustering reduces statistical power substantially. Do not use it if you have fewer than about 6 to 8 clusters per condition, as cluster-level estimates will be unreliable regardless of how many individuals are measured.
Strengths & limitations
- Prevents contamination between experimental conditions when participants share a social or physical environment.
- Mirrors real-world group-level interventions, enhancing external validity for educational, organizational, or team-based contexts.
- Retains the controlled conditions of a laboratory setting while accommodating group-structured recruitment.
- Compatible with stratified randomization to balance clusters on key baseline variables, improving precision.
- Appropriate when the unit of intervention is inherently collective such as a classroom exercise or team task.
- Statistical power is determined primarily by the number of clusters, not the total number of participants; many individuals per cluster add relatively little when the ICC is moderate or high.
- Requires more clusters — and therefore more logistical resources — than an individually randomized design of equivalent power.
- Intraclass correlation must be estimated to plan sample size, yet ICCs are often unknown in advance and pilot estimates are imprecise.
- Cluster-level imbalance on baseline characteristics can introduce confounding that is difficult to correct post-hoc.
Frequently asked
How is this different from a standard laboratory experiment?
In a standard lab experiment, individual participants are randomly assigned to conditions. In a cluster randomized lab experiment, entire groups such as lab sessions or teams are the unit of randomization, and all individuals within a group receive the same condition. This matters when participants within a group can influence each other, making individual randomization susceptible to contamination.
How many clusters do I need?
A common rule of thumb is at least 6 to 8 clusters per condition. The exact number depends on the expected intraclass correlation, the number of individuals per cluster, and the target statistical power. Use the design effect formula DE = 1 + (m-1) x ICC, where m is mean cluster size, to compute the required number of clusters. Because power is driven mainly by cluster count, adding more individuals per cluster yields diminishing returns when ICC is moderate or high.
What is the intraclass correlation coefficient and why does it matter?
The ICC measures the proportion of total outcome variance attributable to cluster membership. An ICC of 0 means clustering is irrelevant; an ICC of 1 means all within-cluster participants give identical outcomes. Higher ICCs require more clusters to achieve the same power and inflate the design effect, which is the factor by which required sample size increases relative to an individually randomized design.
Which statistical method should I use to analyze clustered lab data?
Mixed-effects multilevel models are the most flexible choice — they correctly partition variance between the cluster and individual levels and handle unequal cluster sizes. Generalized estimating equations provide a population-average alternative. At minimum, cluster-robust standard errors should be applied if multilevel modeling is not used. Never analyze cluster randomized data with ordinary regression or t-tests that assume independent observations.
Can I use this design if my clusters differ in size?
Yes, unequal cluster sizes are common and can be handled by multilevel models, which naturally accommodate imbalance. However, large variation in cluster size reduces efficiency and complicates power calculations. Pre-registration of the analysis plan specifying how size imbalance will be handled is strongly recommended.
Sources
- Murray, D. M. (1998). Design and Analysis of Group-Randomized Trials. Oxford University Press. ISBN: 978-0195120363
- Donner, A., & Klar, N. (2000). Design and Analysis of Cluster Randomization Trials in Health Research. Arnold. ISBN: 978-0340691533
How to cite this page
ScholarGate. (2026, June 3). Cluster Randomized Laboratory Experiment. ScholarGate. https://scholargate.app/en/experimental-design/cluster-randomized-laboratory-experiment
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.
- Blocked Laboratory ExperimentExperimental design↔ compare
- Cluster Randomized Controlled TrialExperimental design↔ compare
- Factorial Laboratory ExperimentExperimental design↔ compare
- Laboratory ExperimentExperimental design↔ compare
- Multilevel ModelingResearch Statistics↔ compare
- Randomized Controlled TrialExperimental design↔ compare