Cluster Randomized Fractional Factorial Experiment
Cluster-Randomized Fractional Factorial Experiment · Also known as: CR-FFE, cluster-randomized fractional factorial design, group-randomized fractional factorial trial, CRFFD
A cluster-randomized fractional factorial experiment combines two design principles: randomization is applied to intact groups (clusters such as schools, clinics, or communities) rather than individuals, and only a carefully chosen fraction of all possible factor-level combinations is tested. This pairing makes it practical to screen or evaluate multiple intervention components simultaneously in settings where individual randomization is infeasible, while keeping the number of required clusters manageable.
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 this design when you need to evaluate several intervention components simultaneously in a setting where individual randomization is impossible or undesirable — such as school-based, clinic-based, or community programs — and the total number of available clusters is too limited to support a full factorial arrangement. It is especially well-suited to intervention component screening and optimization phases (as in the MOST framework). Do not use it when the number of clusters is very small (fewer than about 6 per condition), as statistical power will be insufficient. Avoid it when two-way interactions among factors are of primary scientific interest but cannot be estimated by the chosen fraction's aliasing structure — upgrade to a higher-resolution fraction or a full factorial design in that case.
Strengths & limitations
- Enables simultaneous evaluation of multiple intervention components without requiring a cluster for every possible factor combination.
- Preserves the ecological validity of group-based interventions by randomizing the natural unit (the cluster) rather than the individual.
- Fractional factorial structure provides substantial savings in the number of clusters needed relative to a full factorial design.
- Produces efficient main-effect estimates when interactions are assumed negligible or are of lower scientific priority.
- Compatible with the Multiphase Optimization Strategy (MOST) for building and refining multicomponent interventions.
- Main effects are aliased with higher-order interactions; if those interactions are non-negligible, estimates are biased.
- Requires more clusters than an individually randomized trial of equivalent power because the ICC inflates the design effect.
- Planning demands specialized expertise in both fractional factorial aliasing theory and cluster-randomized trial methodology.
- Contamination between clusters assigned to different conditions can bias treatment effect estimates if clusters are geographically or organizationally close.
Frequently asked
How is this different from an ordinary cluster-randomized trial?
A standard cluster-randomized trial compares a single intervention condition against a control. A cluster-randomized fractional factorial experiment tests multiple factors simultaneously by assigning clusters to different combinations of factor levels chosen from a fractional factorial plan, enabling main effects of each factor to be estimated from the same pool of clusters.
What is the ICC and why does it matter here?
The intraclass correlation coefficient (ICC) quantifies how similar outcomes are within the same cluster relative to outcomes across clusters. The effective sample size is reduced by a design effect DE = 1 + (m-1) x ICC, where m is average cluster size. A higher ICC or larger clusters shrink effective sample size more. Estimate the ICC from prior data before calculating the number of clusters needed.
What does aliasing mean and why does it matter?
Aliasing means certain effects cannot be distinguished from each other because they are confounded by the fractional factorial structure. In a Resolution III design the main effect of Factor A may be aliased with the BC interaction. If that interaction is truly non-zero, the estimate of A's main effect will be biased. Choose your resolution level so that confounded effects are scientifically inconsequential.
How many clusters do I need?
The number depends on the ICC, cluster size, effect size, number of treatment combinations, and desired power. As a rough lower bound, most methodologists recommend at least 6 clusters per treatment condition. With fewer clusters, permutation-based tests may be more reliable than normal-theory tests.
Can I use this design if my clusters are very different sizes?
Yes, but variable cluster size reduces efficiency and complicates analysis. Mixed-effects models and GEE handle unbalanced data, but power calculations must account for size variability. Stratifying randomization by cluster size or including size as a covariate will improve precision.
Sources
- Box, G. E. P., Hunter, J. S., & Hunter, W. G. (2005). Statistics for Experimenters: Design, Innovation, and Discovery (2nd ed.). Wiley. ISBN: 978-0471718130
- Murray, D. M. (1998). Design and Analysis of Group-Randomized Trials. Oxford University Press. link ↗
How to cite this page
ScholarGate. (2026, June 3). Cluster-Randomized Fractional Factorial Experiment. ScholarGate. https://scholargate.app/en/experimental-design/cluster-randomized-fractional-factorial-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.
- Adaptive ExperimentExperimental design↔ compare
- Cluster Randomized Controlled TrialExperimental design↔ compare
- Factorial Randomized Controlled TrialExperimental design↔ compare
- Fractional Factorial ExperimentExperimental design↔ compare
- Multi-arm experimentExperimental design↔ compare