How Can Contamination Between Study Groups Threaten Internal Validity?

What Happens When Your Control Group Receives Part of the Intervention?

A researcher evaluates a new AI-supported learning platform by randomly assigning students to an intervention group and a conventional instruction group. The intervention students receive access to personalized learning materials and automated feedback.

Several weeks later, the researcher discovers that intervention students have been sharing platform-generated explanations and practice materials with their classmates in the comparison group. Some comparison students have also started using the platform through shared accounts.

The two groups no longer differ in the way the research protocol intended. If their final examination scores are similar, does that mean the intervention was ineffective?

This is a problem of study group contamination. When participants receive intervention components outside their assigned condition, the actual exposure contrast may differ from the planned comparison. Depending on the research question and analysis, this can complicate or undermine causal interpretation.

How Does Contamination Between Study Groups Affect Internal Validity?

Why Intervention Spillover Can Compromise Research Comparisons

What Is Contamination in an Intervention Study?

Contamination occurs when participants or units assigned to one study condition receive elements of another condition, contrary to the intended separation of interventions.

In educational research, contamination may occur when students share instructional materials, teachers adopt practices intended for another group, or participants access restricted learning resources.

In healthcare research, clinicians trained in an intervention may apply their new practices to patients assigned to usual care.

In organizational studies, employees assigned to different conditions may exchange training materials or adopt similar work practices.

Contamination is particularly likely when participants interact regularly, intervention materials are easy to share, or the same personnel deliver different study conditions.

The methodological concern is not simply that participants communicate. It is whether their interactions alter the intervention exposure or causal comparison specified by the research design.

Why Does Contamination Matter for Internal Validity?

Intervention studies generally compare outcomes under defined treatment conditions.

Suppose the intended comparison is between receiving a digital learning program and receiving conventional instruction without that program.

If comparison students gain access to the digital program, their outcomes may no longer represent the intended untreated condition.

The observed difference between groups may consequently be smaller than the effect of receiving the program compared with receiving no program.

However, the interpretation depends on the causal estimand, meaning the precisely defined effect the study aims to estimate.

A randomized comparison analyzed according to original assignment may still provide a valid estimate of the effect of being assigned or offered the intervention under the actual study conditions, assuming randomization and other requirements remain intact.

That assignment effect may differ from the effect of receiving the intervention without contamination.

Contamination therefore does not automatically invalidate randomization. It may instead change which causal question the observed comparison can answer.

This distinction is essential when assessing threats to the credibility of a study's main comparison.

What Are the Common Forms of Study Group Contamination?

Contamination can arise through direct access to intervention materials, behavioral spillover, shared personnel, or institutional processes.

These mechanisms may overlap. For example, shared teachers can transmit intervention practices while students independently exchange materials.

Researchers should identify the actual route of contamination rather than treating all unintended exposure as a single undifferentiated problem.

How Is Contamination Different From Nonadherence?

Contamination and nonadherence are related, but they describe different aspects of intervention implementation.

A comparison student who independently obtains intervention materials may be described as receiving a contaminated control condition. This may also constitute nonadherence to the assigned condition.

An intervention student who never uses the assigned learning platform demonstrates nonadherence, but this does not necessarily involve contamination between groups.

The distinction matters because failure to follow an assigned intervention and spillover across participants can have different causal implications.

Does Contamination Always Make Treatment Effects Appear Smaller?

No. Contamination is often expected to reduce the contrast between intervention and comparison groups, particularly when the intervention is beneficial and comparison participants receive some of its active components.

Under those circumstances, the observed difference may be closer to zero than the effect under complete separation.

However, this direction is not guaranteed.

Contamination may involve harmful components, selective uptake, compensatory behavior, or interactions that change how an intervention operates.

For example, students who receive shared materials without accompanying instruction may use them differently from students receiving the complete program.

Alternatively, intervention students may benefit from teaching their peers, creating additional learning opportunities that would not occur under strict separation.

The observed effect may therefore be attenuated, exaggerated, or otherwise altered depending on the mechanism.

Researchers should avoid assuming the direction of contamination bias without examining the causal pathways involved.

What Is the Difference Between Contamination and Interference?

In causal inference, interference occurs when one participant's treatment assignment or exposure affects another participant's outcome.

For example, a student may improve because a classmate assigned to the intervention shares newly learned strategies.

Contamination is a common practical mechanism through which interference can occur, but the concepts are not identical.

Interference may occur even when no participant directly receives another group's intervention materials. A vaccination program, for example, may affect untreated individuals through changes in transmission within a community.

Traditional individual-level causal analyses often rely on an assumption that one participant's outcome does not depend on other participants' treatment assignments. This is sometimes called the no-interference component of the stable unit treatment value assumption.

When interference is present, researchers may need to distinguish direct effects from spillover effects and define the intervention at an appropriate individual, group, or population level.

Contamination is not necessarily an unwanted phenomenon in every research question. If the objective is to study how practices spread through a community, spillover may be part of the effect of substantive interest.

Why Is Contamination Common in Educational Research?

Educational interventions frequently operate within interconnected environments.

Students communicate outside formal lessons, instructors teach multiple classes, and learning materials are often available through shared digital systems.

These interactions make strict separation difficult, particularly when intervention and comparison participants belong to the same institution.

For example, a researcher may assign students individually within one class to receive different instructional approaches. Yet students continue collaborating on assignments and discussing the intervention activities.

The resulting comparison may not represent isolated exposure to two independent teaching conditions.

Researchers should therefore assess whether individual randomization is compatible with the way the intervention is delivered and the social relationships among participants.

Can Cluster Randomization Reduce Contamination?

Cluster randomization assigns intact groups, such as classrooms, schools, clinics, or workplaces, rather than individual participants to intervention conditions.

It may reduce contamination when participants within a cluster interact frequently but interactions between clusters are less common.

For example, assigning entire classrooms to instructional conditions may be more practical than assigning individual students within the same classroom.

However, cluster randomization does not guarantee that contamination disappears. Teachers may work across classes, students may interact across groups, and institutional policies may spread intervention practices beyond assigned clusters.

Cluster-randomized trials also introduce additional design and analytical considerations.

Outcomes within clusters may be correlated, reducing the amount of independent information available relative to a study with the same number of independently randomized individuals.

Researchers must account for clustering when determining sample size and conducting statistical analysis.

When only a small number of clusters is randomized, baseline imbalance and uncertainty about treatment effects may be particularly important.

The Cochrane Handbook identifies avoiding contamination as one reason researchers may choose cluster randomization, while emphasizing the distinct methodological requirements of cluster-randomized studies.

Can Physical or Administrative Separation Prevent Contamination?

Separating intervention and comparison activities may reduce opportunities for unintended exposure.

Possible measures include distinct instructional sessions, restricted access to intervention-specific digital resources, separate intervention delivery teams, and clear procedures for staff working across conditions.

However, restrictions must be proportionate, ethically acceptable, and consistent with participants' rights.

Researchers should not assume that students or patients can reasonably be prevented from ordinary communication or from accessing services to which they are otherwise entitled.

In educational settings, restricting essential learning resources merely to maintain an experimental contrast may raise ethical concerns.

When complete separation is impractical, the design should acknowledge expected spillover and consider whether a different unit of assignment or estimand would be more appropriate.

How Can Researchers Measure Contamination?

Contamination assessment begins by defining which intervention components are expected to remain exclusive to each condition.

Researchers can then document whether participants or providers receive, share, or deliver those components outside the assigned condition.

Depending on the setting, evidence may include platform access logs, implementation records, structured observations, participant reports, instructor logs, or intervention fidelity assessments.

Each source has limitations. Platform logs may show access without establishing meaningful engagement. Self-reports may be affected by recall or social desirability. Observations may capture only a subset of interactions.

Researchers should distinguish evidence that contamination occurred from evidence about its intensity or causal consequences.

Where feasible, monitoring procedures should be specified before the trial begins and applied consistently across groups.

Does Intention-to-Treat Analysis Solve Contamination?

Intention-to-treat analysis compares participants according to their original randomized assignments, regardless of subsequent adherence or crossover.

In a properly randomized trial without problematic interference between participants, this approach can preserve the causal interpretation of the effect of assignment under the actual implementation conditions, assuming outcomes are adequately observed and other relevant conditions hold.

However, it does not necessarily estimate the effect of receiving the intervention under complete separation.

When participants' outcomes depend on other participants' assignments or exposures, the assignment effect itself may depend on the allocation pattern and interaction structure.

Researchers should therefore avoid treating intention-to-treat analysis as a universal solution to contamination.

Excluding contaminated participants or analyzing only those who followed their assignments can introduce selection bias because postassignment behavior may be associated with prognosis or potential outcomes.

Alternative causal estimands and analytical strategies may be appropriate, but they require explicit assumptions and sufficient information.

Can Researchers Adjust Statistically for Contamination?

Researchers may consider methods that account for treatment receipt, adherence, crossover, or interference, depending on the design and research question.

However, simply adding observed contamination status as a covariate does not necessarily produce an unbiased estimate of the intervention's effect.

Contamination may occur after assignment and may be associated with motivation, access to resources, social networks, or other characteristics influencing outcomes.

Conditioning on such postassignment variables can introduce bias or change the causal effect being estimated.

Instrumental-variable approaches, principal stratification, and interference-aware causal models may be relevant in some settings, but their validity depends on assumptions that cannot be established merely by fitting a statistical model.

Researchers should prioritize a design that addresses likely contamination mechanisms before relying on post hoc adjustment.

How Should Contamination Be Reported?

Transparent reporting should describe the intervention and comparison conditions as planned and as actually implemented.

Researchers should identify known routes of contamination, the available evidence about their frequency or intensity, and their implications for the intended estimand.

The CONSORT 2025 reporting guideline calls for information about intervention and comparator delivery, adherence, and concomitant care where relevant. Such reporting can help readers evaluate whether the actual conditions corresponded to the intended comparison.

Researchers should also distinguish prespecified analyses from exploratory investigations prompted by contamination discovered during the study.

Acknowledging contamination does not necessarily invalidate the research. It allows readers to understand what the study actually compared and how confidently its findings can be interpreted.

How Shared Digital Learning Materials Can Alter an Experimental Comparison

The final examination results are summarized below.

The observed difference is three points. Further investigation finds that 30 comparison students accessed at least some intervention materials.

The researchers should not simply remove the 30 comparison students who accessed intervention materials and compare the remaining participants with the intervention group.

Those who obtained access may differ in motivation, academic preparation, peer networks, or other outcome-related characteristics. Excluding them could undermine the comparability created by random assignment.

Nor can the researchers conclude that the platform's true effect must exceed three points. The direction and magnitude of contamination's consequences depend on how shared materials influenced learning and participant interactions.

The example demonstrates why intervention delivery must be documented and why researchers should distinguish the effect of assignment from the effect of receiving a particular treatment under specified exposure conditions.

Common Misconceptions About Study Group Contamination

How to Prevent and Manage Contamination in Your Research Design

Before assigning participants, identify which intervention components could move between groups and how participants or providers might transmit them.

Consider the social and organizational structure of the research setting. Students in the same class, employees in the same department, or patients treated by the same clinicians may not be independent in their exposure to intervention practices.

Then determine whether the intended causal comparison requires separation of those exposures or whether spillover is itself part of the intervention's real-world effect.

For educational technology interventions, access logs and implementation records may help characterize exposure, but researchers should avoid equating access with meaningful engagement.

Where contamination cannot be prevented, report the actual intervention conditions and interpret results according to the causal comparison supported by the data.

As with other validity concerns, anticipating threats during research design is generally more useful than attempting to correct every implementation problem after data collection.

Contamination Checks Before and During an Intervention Study

Frequently Asked Questions About Study Group Contamination

Intervention Exposure Must Match the Causal Comparison You Intend to Make

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.