When Is a Crossover Trial Inappropriate?

When Does Giving Every Participant Multiple Treatments Undermine the Comparison?

A crossover trial allows participants to receive multiple interventions in a planned sequence. Because each participant contributes observations under different treatments, the design may reduce the influence of stable individual differences and improve statistical precision.

But consider an experiment comparing two instructional programs. Students receive the first program, acquire new knowledge, and then proceed to the second. Can the second program still be evaluated as though those students had not already learned from the first?

Similar difficulties arise when interventions have persistent biological effects, participants' conditions change substantially over time, or completing multiple treatment periods becomes impractical.

The central question is not whether participants can physically receive both interventions. It is whether their outcomes under successive treatments can support the intended treatment comparison.

Crossover Trials Are Unsuitable When Sequential Treatment Effects Cannot Be Separated Reliably

Conditions That Make Crossover Trials Methodologically Unsuitable

What Makes a Crossover Trial Different From Other Repeated-Measures Studies?

In a conventional crossover trial, participants receive two or more interventions in successive treatment periods. A common two-treatment, two-period design randomly allocates participants to AB or BA.

Each participant contributes outcomes under both treatments. The analysis uses these related observations to estimate a treatment contrast while considering the study's sequence and period structure.

This arrangement is most defensible when the intervention effects are sufficiently temporary, the underlying condition is reasonably stable, and earlier treatment does not materially distort the response to later treatment.

These requirements distinguish crossover treatment comparisons from longitudinal studies that intentionally investigate cumulative learning, disease progression, or other changes over time.

Understanding the circumstances that favor crossover trials is therefore a prerequisite to identifying when the design should be rejected.

When Do Irreversible Treatment Effects Rule Out a Conventional Crossover?

Some interventions produce changes that cannot reasonably be reversed before the next treatment period.

Examples include permanent surgical modifications, interventions that produce durable immunity, and educational programs that teach knowledge or skills participants retain.

Suppose students receive a program that teaches them to identify fabricated research citations. When they later receive a different verification program, their performance may reflect knowledge acquired from the first intervention.

The second treatment is no longer being evaluated under the same initial conditions as the first.

Counterbalancing can distribute treatment sequences, but it cannot restore participants to a state of not knowing what they previously learned.

A conventional crossover design may therefore be unsuitable when the intended estimand is the effect of each intervention administered independently to participants without prior exposure.

This does not prohibit research on intervention sequences. It means that the sequence itself may need to become the object of investigation rather than treating each period as an interchangeable treatment comparison.

Why Are Persistent Treatment Effects a Serious Problem?

Carryover occurs when an earlier treatment continues to influence outcomes during a later treatment period.

Suppose treatment A produces an effect that persists after participants begin treatment B. The observed outcome under B may then depend on both interventions.

The resulting difference cannot necessarily be interpreted as the independent effect of B compared with A.

Carryover may arise through residual pharmacological activity, biological adaptation, behavioral changes, or acquired strategies.

In a two-period crossover trial, treatment-specific carryover can be difficult to distinguish from other sequence-related differences. The available data may not support separate estimation of every relevant effect without additional assumptions.

Researchers should therefore evaluate whether residual treatment effects are plausible before choosing a crossover arrangement.

When Is Washout Insufficient or Impossible?

A washout period separates treatment periods to allow residual intervention effects to diminish.

Its adequacy depends on the mechanism, duration, and outcome of interest. A pharmacological intervention may require consideration of drug elimination and persistent biological activity. A behavioral intervention may produce learning or habits that remain after the formal treatment ends.

There is no universal washout duration that makes every crossover trial valid.

For example, waiting several weeks after an instructional program does not necessarily eliminate the knowledge students acquired.

Even when a treatment is no longer being administered, its consequences may persist.

Researchers should establish whether a defensible washout interval can be determined rather than choosing an arbitrary gap between treatments.

Why Are Progressive or Rapidly Changing Conditions Problematic?

Crossover trials compare treatments administered at different times. Their interpretation becomes difficult when participants' underlying conditions change substantially across periods.

For example, a rapidly progressive disease may alter treatment responsiveness between the first and second periods. Differences in outcomes may then reflect both treatment and disease progression.

Similar concerns arise in developmental or educational research when participants acquire substantial knowledge, mature, or experience important contextual changes between interventions.

A condition need not remain perfectly constant for crossover analysis to be possible. Natural fluctuations and common period effects can sometimes be accommodated.

The difficulty becomes consequential when time-related changes are substantial, unpredictable, or interact with the treatments in ways the design cannot adequately distinguish.

When Do Treatment-by-Period Interactions Undermine Interpretation?

A period effect describes an outcome difference associated with the measurement period rather than the treatment alone.

For example, participants may perform better during the second period because they have become familiar with the assessment procedure.

Balanced treatment sequences can help separate a common period effect from the average treatment contrast under suitable assumptions.

However, treatment effects may themselves depend on the period.

Suppose an intervention is effective for inexperienced participants but provides little additional benefit after participants have completed several related tasks. Its effect may differ between early and later periods.

In such circumstances, a single average treatment effect may obscure meaningful treatment-by-period variation.

Researchers should distinguish an additive period effect from an interaction in which the treatment contrast changes across periods.

What Happens When Participants Cannot Complete Every Treatment Period?

Crossover trials often require longer participation and repeated assessments. Participants may withdraw, experience adverse events, discontinue treatment, or become unavailable before completing the assigned sequence.

Incomplete sequences reduce the information available for within-participant comparisons.

More importantly, missingness may be related to treatment response. Participants experiencing poor tolerability or inadequate benefit may be more likely to discontinue, creating potentially informative missing data.

Some statistical models can incorporate incomplete repeated observations under appropriate assumptions. Nevertheless, analytical flexibility does not guarantee that missing outcomes can be recovered without bias.

If substantial withdrawal is expected, the anticipated efficiency advantage may be overstated and the interpretation of treatment effects may become uncertain.

When Are Crossover Trials Ethically Inappropriate?

Ethical suitability must be evaluated independently of statistical efficiency.

A crossover design may be unacceptable if participants must discontinue an effective treatment, experience unnecessary risk during switching, or undergo repeated procedures whose cumulative burden is disproportionate to the anticipated scientific value.

For example, a treatment interruption that exposes participants to serious deterioration cannot be justified merely because a crossover design would require fewer participants.

Ethical assessment should consider clinical or substantive equipoise, available alternatives, participant safety, informed consent, and the consequences of delaying or switching interventions.

Requirements vary by discipline and jurisdiction, so researchers should consult the applicable ethics review framework and professional standards.

Can Blinding and Randomization Solve These Problems?

Randomization helps protect against systematic differences in who receives each treatment sequence. Blinding may reduce selected performance, reporting, or assessment biases when feasible.

Neither technique eliminates persistent treatment effects or reverses changes caused by an earlier intervention.

Likewise, counterbalancing treatment sequences can distribute some order-related influences but cannot guarantee that the intended treatment contrast remains identifiable.

These safeguards are important, but their effectiveness depends on the particular threat they are designed to address.

Why Is Testing for Carryover Afterward Not a Reliable Rescue Strategy?

Researchers sometimes propose testing for carryover and then deciding whether to analyze the full crossover data or only the first period.

This approach can be problematic. Carryover tests may have limited power, particularly in modest samples, and a nonsignificant result does not demonstrate that residual effects are absent.

Choosing the primary analysis based on a preliminary significance test can also affect the statistical properties of subsequent inference.

The possibility of carryover should instead be addressed primarily during design development through substantive evidence, justified treatment intervals, and an appropriate prespecified analysis.

If the assumptions needed for the crossover comparison are not credible, researchers should consider a different design before collecting data.

When Is a Parallel-Group Trial Preferable?

A parallel-group trial assigns participants to different intervention groups rather than requiring each participant to receive every treatment.

This arrangement may be preferable when interventions have persistent effects, participants' conditions change substantially over time, or switching treatments creates ethical or operational problems.

The choice should reflect the intended treatment effect and feasibility rather than a general preference for one design.

Why a Crossover Trial May Fail When Comparing Instructional Programs

The proposed crossover trial is inappropriate for estimating independent initial learning effects because the first intervention changes participants' knowledge before the second begins.

The methodological problem is not that learning occurs. It is that learning changes the meaning of the subsequent treatment comparison.

Common Misconceptions About Unsuitable Crossover Trials

How to Decide When to Reject a Crossover Design

Begin with the treatment mechanism. Ask what changes after participants receive the first intervention and whether those changes could affect their response to the second.

Next, examine the natural course of the condition or outcome. A stable baseline process supports crossover comparisons more readily than one characterized by rapid progression, developmental change, or unpredictable shifts.

Researchers should document why the selected design can estimate the intended effect, which assumptions remain uncertain, and how those uncertainties influence interpretation.

A defensible design choice is not necessarily the one with the smallest sample size. It is the one that produces interpretable evidence under assumptions appropriate to the intervention and research question.

Before Rejecting or Approving a Crossover Trial

Frequently Asked Questions About Inappropriate Crossover Trials

Reject Crossover Designs When Their Sequential Comparisons Cannot Be Defended

What Happens When the First Treatment Still Affects the Second Treatment Period?

Imagine a crossover trial comparing two interventions, A and B. Participants receive A during the first period, stop treatment, and then begin B.

But what if A continues to influence their responses after B has started?

The researcher may observe an outcome during the B period that reflects not only B but also the lingering consequences of A. The result could make B appear more effective, less effective, or more similar to A than it would otherwise appear.

This problem is called carryover. It is one of the most consequential threats to the interpretation of crossover trials because it challenges the assumption that outcomes under successive treatments can be compared meaningfully.

Carryover Occurs When an Earlier Treatment Influences a Later Treatment Period

How Carryover Arises and Why It Matters for Causal Interpretation

What Exactly Carries Over Between Treatment Periods?

Carryover concerns the continuing influence of a previous intervention after the participant has moved into another treatment period.

The earlier intervention itself may remain active, or its consequences may persist even after the intervention has ended.

For example, a drug may remain biologically active after administration stops. Alternatively, an intervention may produce a physiological adaptation that continues after the drug has been eliminated.

In behavioral or educational research, participants may retain strategies, knowledge, habits, or expectations acquired during the earlier condition.

The defining feature is that exposure history affects the outcome observed under the subsequent treatment.

How Does Carryover Appear in an AB/BA Crossover Trial?

Consider a randomized two-treatment, two-period crossover trial. Participants are assigned to either AB or BA.

In the first period, participants have not yet received the other study treatment. In the second period, their responses may depend on both the current intervention and the previous one.

If the residual effects of A and B differ, the observed treatment contrast may vary according to the assigned sequence.

This is why randomized sequence allocation is necessary but not sufficient. Randomization determines who receives each order; it does not remove the biological or behavioral consequences of earlier exposure.

What Is the Difference Between Carryover and a Period Effect?

A period effect is an outcome difference associated with the measurement period rather than the current treatment.

Suppose participants perform better during the second period because they are more familiar with the testing environment. If this improvement occurs similarly regardless of the treatment sequence, it may be represented as a common period effect.

Carryover is different because the later response depends specifically on the treatment administered previously.

Both may occur in the same trial. Researchers must also consider treatment-by-period interactions, where the current treatment effect changes across periods for reasons that may not be reducible to a simple additive carryover term.

The broader problem of order-related influences in experiments includes these distinct mechanisms.

Can Carryover Be Positive or Negative?

Yes. Carryover may increase or decrease the outcome measured during a later period.

Suppose treatment A improves a participant's performance and some benefit persists during the B period. B may then appear more effective than it would have appeared without prior A exposure.

Alternatively, if A causes lingering fatigue or another unfavorable effect, performance under B may be worse.

The direction of bias depends on the treatment effects, residual effects, sequence allocation, and outcome scale.

Carryover is sometimes described as tending to reduce observed treatment differences in conventional crossover trials. That may occur under particular assumptions, but it is not a universal mathematical consequence. Differential carryover can bias an estimated contrast in either direction.

What Is First-Order Carryover?

First-order carryover refers to a model in which the outcome during a treatment period may depend on the treatment administered immediately before it.

For example, in a three-period sequence ABC, a first-order carryover model considers the influence of A on the B period and B on the C period.

It does not ordinarily include a direct residual influence of A on C after the intervening B period.

That restriction is an assumption about the persistence of earlier effects, not a guarantee that more distant exposure histories are irrelevant.

If an intervention produces effects that last across several subsequent periods, a first-order model may be inadequate.

Structured counterbalancing arrangements can balance selected immediate predecessor relationships, but they do not automatically remove the carryover effects themselves.

How Can Carryover Bias the Estimated Treatment Effect?

Consider a simplified two-period crossover model with additive treatment effects, a common period effect, and treatment-specific carryover affecting only the second period.

Let the direct treatment effects be represented by τ_A and τ_B, while λ_A and λ_B represent the residual effects of A and B.

For the AB sequence, the within-participant difference between the A and B periods includes the direct treatment contrast, the period effect, and residual A carryover.

For the BA sequence, the corresponding contrast includes the same direct treatment contrast, an oppositely signed period effect, and residual B carryover.

With equal sequence weighting, the common additive period effect cancels when the appropriately oriented contrasts are averaged.

The calculation illustrates why the direction and magnitude of bias depend on differential residual effects. It does not imply that carryover parameters can be reliably estimated from an ordinary two-period crossover dataset.

Indeed, the difficulty of separating direct treatment effects from treatment-specific carryover is a central reason for evaluating carryover plausibility before the trial.

Does Equal Carryover From Both Treatments Cancel?

Under the simplified balanced additive model, equal carryover parameters produce no differential carryover term in the estimated A − B contrast.

However, this result depends on the model and estimand. Equal residual effects do not establish that participants have returned to their original states, and carryover may interact with current treatment or measurement period.

Researchers should therefore avoid treating equal carryover as a general justification for ignoring exposure history.

Can a Washout Period Prevent Carryover?

A washout period is an interval between treatment periods intended to allow residual effects to diminish.

It can be useful when intervention effects are temporary and their duration can be reasonably established.

For example, a pharmacological crossover trial may consider elimination kinetics and the persistence of biological effects when determining the interval between treatments.

However, drug elimination and disappearance of treatment effects are not always equivalent. A biological response may persist after the administered substance is no longer present at an appreciable concentration.

In educational research, a waiting period cannot reliably eliminate knowledge acquired during an earlier intervention.

The appropriate question is therefore not simply how long participants should wait, but whether the residual influence of the first treatment is sufficiently negligible when the subsequent outcome is measured.

Determining an appropriate washout duration requires evidence about the intervention and outcome rather than an arbitrary interval.

Can Researchers Detect Carryover With a Statistical Test?

Statistical tests for carryover have been proposed, but their interpretation is difficult, particularly in conventional two-period crossover trials.

A nonsignificant test does not establish that carryover is absent. Limited statistical power may prevent detection of effects that are substantively important.

Furthermore, carryover parameters may be difficult to distinguish from sequence-related differences without additional assumptions.

Using a preliminary carryover test to decide whether to analyze both periods or only the first can introduce additional inferential problems.

The Cochrane Handbook identifies carryover as a principal concern when assessing bias in crossover trials and cautions against selecting first-period analyses on the basis of a preliminary carryover test.

How Can Carryover Be Investigated Without Relying on One Significance Test?

Researchers can examine several sources of evidence when evaluating possible carryover.

Prior studies may provide information about how long an intervention remains active. Pharmacokinetic or pharmacodynamic evidence may be relevant for medicines, while learning-retention evidence may be more appropriate for educational interventions.

Repeated measurements during washout may help characterize the return toward baseline when the outcome can be measured safely and meaningfully.

Descriptive outcomes by treatment, sequence, and period may also reveal patterns requiring explanation.

However, none of these observations automatically identifies the causal contribution of carryover. Their interpretation depends on measurement reliability, the underlying time course, and the assumptions of the design.

When residual effects remain plausible, researchers should examine how the conclusions depend on those assumptions and report the uncertainty transparently.

When Does Carryover Make a Crossover Trial Inappropriate?

Carryover becomes a fundamental design problem when earlier interventions produce persistent effects that cannot be sufficiently reduced before later outcome measurement.

This is particularly relevant to irreversible interventions, durable learning, permanent physiological changes, and outcomes that cannot meaningfully recur under different treatment conditions.

If the underlying condition also changes substantially across periods, distinguishing current treatment effects from exposure history may become even more difficult.

In such circumstances, the assumptions required for a conventional crossover trial may not be defensible.

A parallel-group trial may then provide a clearer comparison because participants are not required to receive both interventions sequentially.

How Residual Learning Can Distort a Crossover Comparison

The example does not establish that carryover occurred. The observed sequence pattern is compatible with several explanations, and the hypothetical means alone cannot isolate the residual effect of Tool A.

Its methodological lesson is that treatment responses may depend on prior exposure. Researchers should establish whether this dependency is part of the research question or a threat to the intended comparison.

Common Misconceptions About Carryover Effects

How to Address Carryover Before and During a Trial

The strongest response to carryover begins before data collection. Researchers should identify plausible mechanisms through which the first intervention might influence later outcomes.

Next, they should assess how long those effects may persist and whether the study can reasonably allow them to diminish before subsequent measurements.

For randomized crossover trials, the protocol should prespecify treatment sequences, washout rationale, outcome measurement timing, and the primary analysis.

The final report should present sufficient information about treatment periods and sequences for readers to evaluate the plausibility of carryover and the robustness of the interpretation.

Researchers should not treat the absence of a statistically significant carryover test as a substitute for substantive justification.

Before Assuming Carryover Is Negligible

Frequently Asked Questions About Carryover Effects

Carryover Makes Treatment History Part of the Outcome

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.