How Do You Measure Contamination Between Study Groups?

How Can You Tell Whether Participants Received Parts of an Intervention They Were Not Assigned to Receive?

You are evaluating a digital learning intervention. Students assigned to the experimental condition receive interactive tutorials, guided practice, and personalized feedback. Control students continue with conventional learning resources.

During implementation, you learn that some control students have obtained intervention worksheets. Others may have watched demonstrations or received explanations from experimental participants. Teachers might also have incorporated intervention-specific techniques into control lessons.

How should you measure these exposures? Is the percentage of control participants who accessed a worksheet sufficient, or do you need to know how much of the intervention they actually received?

Contamination is difficult to evaluate when researchers have not specified what counts as unintended exposure or established how that exposure will be recorded. The measurement strategy therefore needs to begin before the first incident is discovered.

Define Contamination Operationally, Then Measure Its Extent and Relevance

Developing a Meaningful Contamination Measurement Strategy

What Exactly Are You Trying to Measure?

Contamination concerns unintended exposure to intervention components across the boundaries separating assigned study conditions.

In a two-arm trial, control participants might receive components of the experimental intervention. Intervention participants might also receive components intended for the comparison condition, although the interpretation depends on the protocol.

Researchers should distinguish contamination from ordinary co-interventions, nonadherence, treatment switching, and permitted background services.

For example, a control student attending ordinary institutional tutoring is not necessarily contaminated. Receiving a tutorial that was specifically restricted to the experimental condition may constitute contamination.

The measurement system should reflect the intended treatment contrast rather than classify every additional activity as a protocol violation.

Start by Defining Contamination Before Data Collection

A useful operational definition identifies the intervention components that count as unintended exposure, the participants or units at risk, and the period during which exposure matters.

Consider an intervention consisting of personalized feedback, guided exercises, and weekly coaching.

The protocol might define contamination as control participants receiving any of these components outside permitted comparison procedures. Researchers could additionally distinguish minor exposure from substantial receipt.

Definitions should be sufficiently precise for different staff members to classify the same incident consistently.

They should also identify permitted activities, such as ordinary feedback available to both groups, to avoid overstating contamination.

The TIDieR framework can help researchers describe intervention components and delivery procedures with enough detail to support this distinction.

Measure More Than the Number of Affected Participants

A contamination rate is useful, but it describes only one dimension of unintended exposure.

The appropriate indicators depend on the intervention's expected mechanism of change. Ten minutes of exposure may be inconsequential for one intervention but sufficient to transmit the essential content of another.

How Do You Calculate a Contamination Rate?

A simple participant-level contamination rate is the proportion of participants in a specified group who experienced the operationally defined exposure during a specified period.

This denominator is appropriate only when contamination status is available for all 100 participants. If exposure information is missing, researchers should report the number assessed, the missing information, and how missing contamination status was handled.

The proportion among assessed participants may not represent the true prevalence among all randomized control participants if exposure ascertainment is incomplete.

Researchers should also specify whether the numerator includes any exposure or only exposure exceeding a defined threshold.

Should You Distinguish Any Exposure From Substantial Exposure?

Often, yes. Binary indicators are convenient but may obscure meaningful differences in intervention receipt.

A control participant who opens a tutorial once should not necessarily be treated as equivalent to someone who completes every activity and receives personalized feedback over several weeks.

Researchers may therefore define categories such as no documented exposure, limited exposure, and substantial exposure.

These categories should reflect the intervention's content and causal theory rather than arbitrary percentages selected after examining results.

When possible, continuous measures such as completed sessions or cumulative exposure time may preserve more information.

However, exposure duration is not always a valid measure of dose. A participant may spend considerable time with a resource without engaging with its essential learning activities.

Which Data Sources Can Measure Contamination?

Different measurement methods capture different aspects of exposure.

Combining complementary sources may strengthen interpretation, but multiple imperfect measures do not automatically establish the true level of contamination.

Can Digital Logs Reliably Identify Contamination?

Digital logs can provide useful evidence when the intervention is delivered through identifiable accounts or recorded activities.

For example, an access log may show that a control participant used an intervention account or completed an experimental tutorial.

However, logs may fail to capture screenshots, downloaded files, shared devices, peer explanations, or activities completed outside the platform.

Conversely, opening a file does not establish that its contents were read, understood, or applied.

Researchers should therefore distinguish recorded interaction from meaningful exposure.

The measurement approach should also respect privacy requirements and avoid collecting unnecessary personal communication data.

How Can Participant Questionnaires Be Designed?

Questions should describe specific intervention components rather than ask only whether participants were contaminated.

Participants may not understand the methodological term or recognize which activities count as experimental components.

A questionnaire might ask whether participants accessed a particular tutorial, received specific feedback, attended a named session, or used materials shared by another participant.

Researchers can also ask about timing, frequency, and approximate duration.

Questions should use neutral language. Participants may underreport sharing if they believe disclosure will lead to punishment or exclusion.

Where feasible, exposure questions should be tested for clarity before the main study.

How Can Staff Observations Detect Contamination?

Independent observers may identify intervention-specific practices being delivered to the wrong condition.

For example, an observer may record that an instructor used a feedback technique restricted to experimental classrooms while teaching a control class.

Structured observation checklists can improve consistency by defining the behaviors to be recorded.

However, observation schedules should cover relevant delivery periods. A small number of observed sessions may miss intermittent contamination.

Staff members may also change their behavior when observed.

Researchers should distinguish documented incidents from conclusions about the overall frequency of contamination.

Should Contamination Be Measured Repeatedly?

Repeated measurement may be appropriate when exposure accumulates or changes during the intervention.

For example, control participants may gradually obtain more experimental materials as students collaborate throughout a semester.

Measuring contamination only at the end may obscure when exposure began and whether it preceded the primary outcome.

Short periodic assessments or continuous administrative records may provide more informative timing data.

However, frequent questionnaires may increase burden or draw attention to intervention differences.

The assessment schedule should therefore reflect expected sharing pathways and the information needed for interpretation.

What If Contamination Occurs at the Classroom or Site Level?

Contamination may affect an entire cluster rather than isolated participants.

For example, a control teacher may begin using an experimental instructional practice with every student in a classroom.

Researchers might record both the proportion of control classrooms exposed and the proportion of students within those classrooms affected.

These measures answer different questions.

In cluster-randomized trials, the relevant denominator and analytical unit should be stated explicitly. A classroom-level contamination rate cannot automatically be interpreted as an individual-level rate.

Researchers should also consider whether exposure differs across clusters and whether the measurement system captures that variation.

How Do You Measure Spillover That Does Not Involve Direct Intervention Receipt?

Some contamination pathways operate through social influence rather than direct access to intervention components.

Students may learn strategies from intervention participants, while teachers may adopt practices introduced through colleagues.

Researchers may investigate these pathways through network information, participant reports, interviews, or observations.

However, social contact alone does not prove that an intervention effect was transmitted.

In causal inference, interference concerns situations where one participant's outcome depends on another participant's treatment assignment or exposure.

Measuring contact patterns can support investigation of interference, but estimating causal spillover effects generally requires additional design features and assumptions.

Researchers should avoid treating every interaction as confirmed contamination.

How Does Contamination Measurement Relate to Intervention Fidelity?

Fidelity concerns whether the intended intervention was delivered as specified. Contamination concerns unintended exposure across treatment conditions.

These are distinct dimensions of implementation.

A study may deliver the experimental intervention with high fidelity while allowing its components to reach control participants.

Alternatively, the experimental intervention may be delivered poorly without any contamination of the control condition.

Researchers should therefore assess intervention fidelity separately from contamination while examining how the two jointly affect the actual treatment contrast.

Can Contamination Measures Be Used to Correct the Treatment Effect?

Not automatically. Contamination occurs after assignment and may be influenced by participant motivation, social relationships, prior outcomes, or other factors.

Participants who obtain intervention materials may differ systematically from those who do not.

Simply including a contamination indicator as a regression covariate may change the estimand or introduce bias.

For a treatment-policy estimand, the primary randomized comparison may appropriately retain participants in their assigned groups.

Estimating what would have occurred without contamination requires additional causal assumptions and potentially specialized methods.

Contamination measurement improves understanding of implementation but does not, by itself, identify the causal effect of unintended exposure.

How Much Contamination Is Acceptable?

There is no universal acceptable contamination percentage.

The consequences depend on which intervention components were received, the intensity and timing of exposure, and the intended treatment effect.

For example, a small proportion of control participants receiving the entire intervention may be more consequential than a larger proportion briefly viewing peripheral materials.

Researchers should therefore interpret measured contamination in relation to the amount of contamination a study can tolerate, rather than apply an arbitrary cutoff.

Prespecified monitoring thresholds can trigger implementation review, but they should not be confused with universal criteria for study validity.

How Should Missing Contamination Data Be Handled?

Missing exposure information should not automatically be interpreted as absence of contamination.

Participants who do not respond to exposure questionnaires may differ from respondents in their sharing behavior.

Similarly, incomplete platform records may underestimate exposure occurring through shared accounts or offline resources.

Researchers should report how much contamination information is missing and consider whether the missingness could distort prevalence estimates.

Where appropriate, sensitivity analyses may examine plausible alternative exposure classifications.

The assumptions underlying such analyses should be stated rather than presented as confirmed facts.

What Should Be Prespecified in the Protocol?

The protocol should define contamination, identify relevant intervention components, and specify the measurement methods and assessment periods.

It should also establish the denominators used for reporting rates, procedures for resolving inconsistent exposure records, and rules for documenting important incidents.

Where contamination is expected, researchers should describe how it will be considered in sample size planning, implementation monitoring, and interpretation.

The primary analytical strategy should remain aligned with the estimand.

Transparent reporting should distinguish observed contamination, suspected exposure, and uncertainty arising from incomplete measurement.

Measuring Contamination in a Digital Research Skills Trial

The results provide useful information about the treatment contrast but do not establish how much the intervention effect was changed by contamination.

That causal question would require additional assumptions and evidence about the effects of the transferred components.

The example also illustrates why an exposure definition should be established before outcome results are examined.

Common Mistakes When Measuring Contamination

Measure Exposure in a Way That Reflects How the Intervention Works

Begin with the intervention's core components and identify how those components could reach participants assigned to another condition.

Choose indicators that capture meaningful exposure without creating unnecessary participant burden or privacy intrusion.

For studies where substantial cross-group exposure is foreseeable, consider designing around anticipated contamination before implementation rather than relying on measurement alone.

The purpose of contamination measurement is to establish what exposure occurred and how it may affect interpretation, not merely to produce a percentage for the limitations section.

Before Measuring Contamination Between Study Groups

Frequently Asked Questions About Measuring Contamination

Measure What Crossed the Group Boundary, Not Just How Many People Were Exposed

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.