What Is Contamination Between Groups, and How Can Study Design Reduce It?

What If Your Control Group Starts Receiving the Intervention Too?

You randomly assign teachers to intervention and comparison conditions. The intervention teachers receive training in a new feedback strategy.

A few weeks later, they begin sharing the materials with colleagues in the comparison group. Those colleagues adopt some of the techniques in their own classrooms.

The groups are still labeled intervention and comparison in your dataset. In practice, however, the conditions are no longer as distinct as the design intended.

This is contamination: exposure to elements of another study condition crosses the boundary between groups. It is particularly important when interventions involve information, behavior, training, shared personnel, or environments in which participants interact.

Contamination Blurs the Contrast Your Study Was Designed to Create

Contamination Is a Problem With What Participants Actually Experience

Random assignment determines what condition participants are intended to receive. Contamination concerns what happens afterward.

Cochrane describes contamination as application of an intervention to participants intended to receive another intervention. NIH research-methods guidance similarly describes contamination as occurring when participants not selected to receive an intervention receive at least some part of it.

The problem is therefore not simply that participants communicate. Communication matters when it changes exposure to the intervention or comparison condition in a way that threatens the intended contrast.

Contamination can occur in several directions

The classic example involves control participants receiving elements of the experimental intervention.

But cross-condition exposure can be more complicated.

Participants assigned to the intervention may seek services characteristic of the comparison condition. Providers may combine elements of both protocols. Participants from different conditions may exchange materials or strategies. Organizational changes introduced for one arm may spread to everyone in the setting.

The useful question is: did participants receive something materially different from the condition to which they were assigned?

Why contamination can dilute an intervention contrast

Suppose the intervention would improve an outcome if delivered as intended. If comparison participants begin receiving part of it, their outcomes may improve too.

The difference between the intervention and comparison groups can then become smaller.

This can make an effective intervention appear less different from the comparator than it would have under clean separation of conditions.

Cochrane identifies contamination as one form of deviation from intended interventions that can bias estimates when it arises because of the trial context and is not appropriately addressed.

However, the direction and magnitude of bias should not be assumed mechanically. The consequences depend on what crosses between conditions, who is affected, the estimand, adherence patterns, and how contamination relates to outcomes.

Educational and behavioral interventions can be especially vulnerable

Consider an intervention that teaches faculty members how to use a new assessment strategy.

Once trained, faculty can discuss the method with colleagues. They can share rubrics, examples, presentations, or templates. If intervention and comparison participants work in the same department, preventing all information exchange may be unrealistic.

The same problem arises in behavioral and organizational research. Participants may discuss counseling techniques, clinicians may alter their general practice after receiving training, or managers may apply a new policy more broadly than intended.

Interventions involving knowledge and behavior do not stay inside a pill bottle. That is precisely why contamination should be considered during design rather than discovered in the discussion section.

Shared intervention providers can become a pathway for contamination

Suppose the same instructor teaches both intervention and comparison classes.

After being trained in the intervention, the instructor may unconsciously incorporate elements of it into the comparison condition. The participants do not need to interact with each other for contamination to occur.

Shared clinicians, teachers, counselors, therapists, facilitators, or other intervention agents can therefore create cross-condition exposure.

Possible responses include using separate providers, clearly differentiating protocols, training providers carefully, monitoring fidelity, or choosing a design in which providers or delivery units are allocated at a higher level.

Shared environments can make individual randomization difficult

Some interventions operate at the level of a classroom, clinic, school, workplace, or community rather than purely at the individual level.

Imagine randomizing individual students within the same classroom to whether their teacher uses a new classroom-management strategy. The intervention is delivered through the classroom environment. It may be impossible to expose one student to the strategy without exposing nearby students.

Individual randomization is poorly aligned with such an intervention.

NIH guidance notes that when contamination risk is substantial, group- or cluster-randomization may be preferable. Cochrane similarly identifies avoiding contamination as one reason cluster-randomized trials may be used.

Cluster randomization can reduce contamination by separating conditions

Instead of randomizing individual students, researchers might randomize entire classrooms. Instead of individual clinicians, they might randomize clinics. Instead of employees, they might randomize worksites.

Members of each cluster then receive the same assigned condition.

This can reduce opportunities for intervention exposure to cross treatment boundaries within the cluster because the boundary is moved to a higher organizational level.

Cluster randomization has costs and does not eliminate contamination completely

Moving randomization to the cluster level is not a free solution.

Participants within the same classroom, clinic, or community tend to have correlated outcomes. The statistical analysis and sample-size calculation must account for that clustering.

There may also be relatively few clusters available, creating challenges for balance and precision. Recruitment after cluster assignment can introduce selection concerns. Clusters themselves may interact, allowing contamination to occur across schools, clinics, or communities.

Cochrane emphasizes that cluster-randomized trials require analysis that respects the level at which randomization occurred.

The decision to cluster-randomize should therefore reflect the intervention and contamination risk, not a belief that cluster designs are inherently stronger.

Physical separation is not the only way to reduce contamination

Researchers can sometimes preserve individual randomization while reducing cross-condition exposure through implementation procedures.

Intervention sessions might be scheduled separately. Materials may be provided through condition-specific accounts. Different providers may deliver different conditions. Staff may receive clear protocol instructions. Access to intervention components can be tracked where ethically and practically appropriate.

These measures should not interfere unreasonably with participants' normal lives or ethical obligations. The goal is to preserve the intended contrast, not to create an artificial research environment at any cost.

Blinding may reduce some forms of contamination

When participants or intervention providers do not know the assigned condition, they may be less likely to seek or deliver the other intervention because of expectations about group membership.

Cochrane notes that successful blinding can reduce deviations from intended intervention, including contamination.

But blinding is not feasible for many behavioral, educational, surgical, or organizational interventions. Even when feasible, it does not prevent every pathway through which intervention components might spread.

Contamination should be monitored when it is plausible

If contamination is a credible threat, researchers should consider how they will know whether it happened.

This might involve documenting attendance, intervention use, access to materials, receipt of outside services, provider fidelity, or other exposure indicators relevant to the study.

Monitoring serves at least two purposes. It tells researchers whether the intervention contrast was delivered as planned, and it helps them interpret unexpectedly small differences between groups.

Measurement itself must be credible. Asking participants a vague question at the end of a long trial may not capture the extent or timing of cross-condition exposure accurately.

Contamination is related to adherence but is not exactly the same thing

Nonadherence means participants do not follow the intervention to which they were assigned as intended. Contamination is specifically concerned with exposure across study conditions.

A participant assigned to an exercise program who attends only half the sessions is nonadherent. A control participant who begins following the experimental exercise program after receiving materials from a friend is contaminated by the intervention.

The two can coexist and both can reduce separation between assigned conditions.

Contamination is not always simply a nuisance

Sometimes spillover is part of how an intervention works in the real world.

A public-health campaign may intentionally influence people beyond those directly targeted. A teacher-training intervention may improve practices among untrained colleagues because teachers share knowledge. A community intervention may diffuse through social networks.

In such cases, what looks like contamination under an individually randomized design may actually be a substantive spillover effect.

The study design should then reflect the question. If researchers care about both direct and indirect effects, preventing all spillover may not be the appropriate objective. They may need a design capable of estimating those effects rather than simply treating every cross-person influence as protocol failure.

The intended causal contrast should determine whether contamination matters

Contamination is consequential because causal effects are defined relative to conditions.

If the “control” condition gradually receives the intervention, the study is no longer comparing the originally specified alternatives as cleanly as intended.

This connects to the counterfactual logic of causal research. The comparator must represent a meaningful alternative if the contrast is to answer the intended causal question.

When a Teaching Intervention Spreads Into the Comparison Group

Cluster assignment is not automatically the correct answer. If departments routinely collaborate with one another, contamination could simply move from within departments to between departments. The researchers must understand how the intervention is likely to spread.

The design should follow the social and organizational pathways through which exposure actually occurs.

Common Misunderstandings About Contamination

Design Around How the Intervention Can Actually Spread

Before deciding how to randomize participants, map the pathways through which intervention components could move between conditions.

Who interacts with whom? Who delivers the intervention? Are materials shareable? Does the intervention alter a common environment? Can participants access it independently outside the study?

Those questions often reveal contamination risk before the first participant is recruited.

If you are still deciding what the comparator should represent, first clarify what makes the comparison group appropriate. Preventing contamination cannot rescue a comparator that answers the wrong scientific question.

Before Finalizing a Design With Separate Study Conditions

Frequently Asked Questions About Contamination Between Groups

Protect the Difference Between Conditions Before It Disappears

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.