Can Crossover Designs Help Researchers Study Small Populations?

Can Participants Serve as Their Own Controls When You Cannot Recruit More People?

Suppose you want to compare two instructional approaches in a specialized university program with only 24 eligible students. Dividing them into two independent groups would leave approximately 12 students per condition.

Instead, you consider allowing every student to experience both approaches. Half would receive approach A followed by approach B, while the other half would receive B followed by A.

Each student would contribute information under both conditions. This might reduce the influence of stable individual differences and improve the precision of the comparison.

However, what happens if students retain knowledge from the first instructional approach? Could the first intervention influence their response to the second? And would a period between interventions genuinely eliminate those effects?

A crossover design may be efficient when the population is small, but its validity depends on whether the intervention and outcome can support a meaningful within-person comparison.

Crossover Designs Can Improve Efficiency When Treatment Effects Are Suitably Reversible

How Crossover Designs Work and When They Are Appropriate for Small Samples

What is a crossover research design?

A crossover design is an experimental arrangement in which participants receive two or more intervention conditions in a planned sequence, with outcomes measured during the relevant treatment periods.

In a conventional two-treatment, two-period crossover trial, participants may be randomly assigned to sequence AB or BA.

Participants assigned to AB receive treatment A during the first period and treatment B during the second. Those assigned to BA receive the treatments in the reverse order.

Because each participant experiences both conditions, the analysis can compare outcomes within individuals rather than relying entirely on differences between independent groups.

The design is sometimes described as a within-subject experimental design, although that broader term also includes arrangements that are not conventional crossover trials.

Why might a crossover design be efficient with few participants?

Individuals often differ in characteristics that influence outcomes, such as baseline ability, symptom severity, motivation, or physiological responses.

In a parallel-group trial, these differences contribute to variability between treatment groups.

In a crossover trial, each participant contributes observations under both treatments. Stable participant characteristics may therefore be controlled through the within-person comparison.

For example, a student's general academic ability may influence performance under both instructional interfaces. Comparing the same student's performance under each interface can reduce the influence of that stable characteristic.

However, the efficiency gain depends on within-person variability, treatment effects, measurement reliability, and the absence of consequential carryover or other design violations.

A crossover trial does not automatically have greater power than every parallel-group alternative.

How does a crossover design differ from a parallel-group trial?

Neither design is universally superior. The appropriate choice depends on the treatment, outcome, and estimand.

What is an AB/BA crossover design?

An AB/BA crossover is a common two-sequence arrangement for comparing two treatments.

Participants are allocated to one of two sequences:

Randomizing sequence can help balance treatment order and protect against systematic assignment differences.

However, sequence randomization does not guarantee that treatment effects are unaffected by period or carryover.

Researchers should specify the treatment periods, assessment windows, and relevant assumptions before recruitment.

What are carryover effects?

Carryover occurs when the effect of a treatment persists into a subsequent period and influences outcomes measured under the next treatment.

For example, a medication may continue affecting the body after administration ends. Similarly, students may retain knowledge or strategies acquired during an instructional intervention.

If the first treatment changes the participant's response during the second period, the observed outcome may reflect both current and previous treatment exposure.

This can undermine the interpretation of the within-person treatment contrast.

Carryover is especially problematic when treatment effects are cumulative, irreversible, or difficult to distinguish from lasting learning.

Researchers should evaluate carryover through substantive knowledge about the intervention rather than assuming that statistical adjustment will resolve it.

Can a washout period prevent carryover?

A washout period is an interval between treatment periods intended to allow effects of the preceding intervention to diminish.

For some short-acting treatments, a scientifically justified washout may reduce residual effects.

However, washout effectiveness depends on the mechanism and duration of the intervention's influence.

For example, waiting one week after students learn a research technique does not necessarily erase the acquired knowledge.

A washout period may therefore be ineffective for interventions designed to produce durable learning or behavioral change.

Researchers should not select an arbitrary interval and assume that carryover has disappeared.

Why are crossover designs difficult for educational interventions?

Many educational interventions aim to produce lasting knowledge, skills, or changes in learning strategies.

These outcomes may persist when participants move from the first instructional condition to the second.

For example, students who learn a problem-solving strategy under approach A may continue using it during approach B.

The second-period performance would then reflect both the current approach and previously acquired skills.

However, not every educational crossover is inappropriate. Researchers may compare transient interface features, presentation formats, or task conditions when prior exposure is unlikely to affect subsequent performance materially.

Different but suitably equivalent tasks, counterbalanced sequences, and careful assessment of practice effects may help in selected settings.

The critical issue is whether the intended treatment effect can be separated from prior learning and period-related changes.

What are period effects?

Period effects are changes in outcomes associated with when a measurement occurs rather than which treatment is received.

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

Patients may experience changes in symptoms because of seasonal variation or disease progression.

Randomizing treatment sequence can help distinguish treatment from period effects under appropriate assumptions.

However, period effects may interact with treatment, and complex time trends can complicate estimation.

The analysis should account for period where appropriate and evaluate whether the design supports a stable treatment contrast.

What is the difference between a period effect and a carryover effect?

A period effect concerns changes associated with the measurement occasion, regardless of the current treatment.

A carryover effect concerns the continuing influence of an earlier treatment on a later outcome.

For example, general improvement from repeated testing may represent a period or practice effect. Continued use of a strategy specifically taught under treatment A may represent carryover from A.

The distinction matters because these processes have different implications for the treatment comparison.

Researchers should not assume that including a period indicator in a regression model eliminates carryover.

How should the treatment effect be estimated?

In a simple two-period crossover, researchers may compare participant outcomes under treatments A and B while accounting for treatment period and relevant sequence structure.

The appropriate model depends on the outcome distribution, treatment schedule, missingness, and assumptions about carryover.

Paired comparisons may be useful in simple settings, while mixed-effects models can represent participant-level dependence and period effects.

However, a model that ignores systematic carryover may produce a misleading treatment estimate.

Researchers should define whether the estimand concerns an immediate treatment effect, a period-specific response, or another scientifically relevant quantity.

Analytical choices should follow that estimand rather than being selected solely for statistical convenience.

Does a crossover design require fewer participants?

It may, but the required sample size depends on the variability of within-person treatment differences and the intended precision or power.

For a simple paired comparison, the standard error of the mean treatment difference depends on the standard deviation of individual differences and the number of independent participants.

When stable participant differences cancel in the within-person comparison, the variance of treatment differences may be smaller than the variance relevant to a parallel-group comparison.

However, highly variable within-person responses, measurement error, or carryover can reduce or eliminate the expected advantage.

What happens when participants withdraw before completing both periods?

Crossover trials may be particularly vulnerable to incomplete treatment sequences because each participant is expected to contribute observations under multiple conditions.

For example, a participant who withdraws after period one cannot provide the intended paired comparison.

Analyzing only complete participants may reduce precision and introduce selection bias if completion depends on treatment response or other outcome-related factors.

Some models can use partially observed data under suitable assumptions, but they cannot automatically correct informative dropout.

Researchers should plan procedures to reduce attrition and prespecify appropriate missing-data analyses.

They should also distinguish discontinuation of treatment from withdrawal of permission for further outcome collection.

When is a crossover design inappropriate?

A crossover is generally unsuitable when the intervention produces irreversible outcomes, lasting effects that cannot be adequately separated, or major changes in the underlying condition over time.

For example, a surgical procedure cannot ordinarily be reversed so that the same participant can later experience an untreated condition in a comparable way.

Likewise, a course designed to produce durable learning may not permit a credible return to the pre-intervention state.

Unstable diseases, long treatment effects, substantial period-by-treatment interactions, and high dropout risk may also undermine feasibility.

In these circumstances, a parallel-group trial or another design may provide a more interpretable comparison despite requiring more participants.

Can crossover designs be combined with repeated measurements?

Yes. Researchers may collect multiple observations within each treatment period to characterize response patterns and improve measurement reliability.

For example, performance might be assessed on several comparable tasks during each interface condition.

However, repeated observations remain correlated within participants, and the analysis must account for treatment periods and within-person dependence.

Repeated measurement may also introduce practice or fatigue effects.

Researchers should distinguish the benefits of repeated measurement in small populations from the specific experimental contrast created by a crossover sequence.

Collecting more measurements does not by itself resolve carryover or period effects.

How should researchers decide between crossover and single-case experimental designs?

Both can use repeated observations, but they are organized around different experimental structures and inferential goals.

A conventional crossover trial generally compares treatment conditions across participants assigned to sequences.

Single-case experimental designs often emphasize repeated measurement within individuals and systematic demonstration of intervention effects across phases or replications.

Some designs share features, and the terminology can vary across disciplines.

The appropriate choice depends on the intervention, outcome, feasibility of repeated treatment changes, and desired causal inference.

Researchers should not treat the terms as interchangeable merely because participants experience multiple conditions.

What should be established before choosing a crossover?

Researchers should first determine whether each participant can meaningfully experience every treatment condition.

They should evaluate treatment reversibility, carryover, period effects, assessment comparability, and the burden of multiple treatment periods.

The protocol should specify sequence randomization, timing, washout where justified, outcomes, and analysis.

Sample size should be based on plausible within-person variability rather than the assumption that crossover designs always require fewer participants.

When those conditions cannot be supported, the apparent efficiency advantage may not justify the design.

Comparing Two Digital Learning Interfaces with 24 Students

The crossover may provide a more efficient comparison than assigning only twelve students to each interface, particularly when stable differences in student speed contribute substantially to outcome variability.

However, if learning to use interface A substantially improves performance with interface B, the second-period outcome may no longer represent an uncontaminated treatment comparison.

The researchers should therefore evaluate whether the crossover's assumptions are credible before prioritizing its potential precision advantage.

Common Misconceptions About Crossover Designs

Choose a Crossover Only When the Within-Person Contrast Remains Meaningful

Before selecting a crossover design, determine whether the treatment effects are sufficiently transient or reversible for participants to experience multiple conditions without consequential contamination.

Then evaluate sequence randomization, measurement timing, period effects, and the expected precision of within-person comparisons.

When crossover assumptions are not credible, researchers may need to revisit the broader design options for small populations rather than forcing a within-person experiment.

The value of a crossover lies in a defensible treatment comparison, not simply in allowing each participant to contribute more observations.

Before Choosing a Crossover Design

Frequently Asked Questions About Crossover Designs

Crossover Designs Help Small Populations Only When Treatment Comparisons Remain Valid

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.