A participant can be their own control only when the first period does not rewrite the second
Crossover trials compare treatments within the same participant, often with impressive precision. That advantage depends on stable conditions, reversible effects and periods separated well enough to prevent one treatment from contaminating the next.
Choose a reading route
- Start from first principles — define the unit and the threat.
- Follow the worked example — make the hidden arithmetic visible.
- Test transfer — change the setting and preserve the rule.
- Audit the evidence — inspect records, assumptions and failure points.
- Use the practical checklist — apply the method to new work.
Chapter contents
Foundations: Chapters 1–3
Method: Chapters 4–6
Transfer and limits: Chapters 7–9
Connections and use: Chapters 10–12
1. Crossover designs compare treatment sequences
In a two-period AB/BA design, participants are randomised to receive A then B or B then A. Each person contributes outcomes under both treatments, reducing variation caused by stable differences between people.
The design is efficient only when treatment effects can be separated from period, sequence and carryover effects.
First principle
Use crossover when the condition is stable and the treatment effect is reversible on a defensible timescale.
2. Carryover changes the next treatment period
Carryover occurs when a previous treatment continues to affect later outcomes. It may be pharmacological, physiological, behavioural, educational or psychological. A washout can remove drug concentration while habits or learning remain.
| Carryover route | Example | Can time alone solve it? |
|---|---|---|
| Drug exposure | Long half-life | Sometimes |
| Physiology | Persistent adaptation | Maybe |
| Learning | Skill acquired | Usually no |
| Belief or preference | Experience changes response | Uncertain |
Map every pathway from first-period treatment to second-period outcome before choosing the design.
← Previous · Contents · Next →
3. Period and sequence effects are different
A period effect is a general change between periods, such as disease progression, season, fatigue or practice. A sequence effect is a systematic difference between AB and BA groups that can reflect carryover, imbalance or interaction between treatment and period.
- Randomise sequence and conceal allocation.
- Measure baseline before each period where appropriate.
- Keep period lengths and procedures comparable.
- Record events occurring between periods.
A simple two-period design has limited ability to distinguish complex carryover from treatment-by-period interaction.
← Previous · Contents · Next →
4. Worked example: unequal within-person contrasts
In a fictional AB/BA study, mean outcomes are higher when better. Sequence AB averages 14 under A in period 1 and 13 under B in period 2. Sequence BA averages 10 under B in period 1 and 15 under A in period 2.
| Sequence | Period 1 | Period 2 | Within-sequence A−B |
|---|---|---|---|
| AB | A = 14 | B = 13 | +1 |
| BA | B = 10 | A = 15 | +5 |
| Naive pooled | A = 14.5 | B = 11.5 | +3 |
The pooled difference of 3 hides the disagreement between +1 and +5. The data alone do not identify which mechanism caused it; design knowledge and prespecified analysis are required.
← Previous · Contents · Next →
5. Justify washout from effect kinetics
A washout should address the duration of every relevant effect, not follow a generic number of days. For a first-order pharmacokinetic process, five half-lives leave about (1/2)⁵ = 3.125% of the original amount, but biological response can outlast concentration.
| Half-lives elapsed | Fraction remaining |
|---|---|
| 1 | 50.0% |
| 2 | 25.0% |
| 3 | 12.5% |
| 5 | 3.125% |
When complete reversibility is doubtful, choose a parallel design or a different estimand.
← Previous · Contents · Next →
6. Plan analysis before testing carryover
Do not rely on a low-powered preliminary carryover test to decide whether the main analysis is valid. Select the design because carryover is unlikely or prevented, then analyse treatment with period, sequence and participant structure as planned.
Report period-specific outcomes and sensitivity analyses. Missing second-period data can break the within-person comparison and introduce selection.
Practical check
Draw a timeline for both sequences showing treatment start, full effect, measurement, washout, residual effect and next-period baseline.
← Previous · Contents · Next →
7. Transfer test: learning interventions
Move the design to two teaching methods. Method A may teach a reusable strategy that improves performance during Method B. No weekend washout can erase that learning without harming the student.
Transfer succeeds when the investigator recognises irreversible learning as carryover and uses a parallel or counterbalanced task design instead of claiming a clean treatment switch.
Crossover logic fits temporary supports better than lasting skills, surgery, vaccination or cure.
← Previous · Contents · Next →
8. Misconceptions and limits
- “Everyone receives both treatments, so confounding disappears.” Time and carryover remain.
- “A long washout guarantees reset.” Behavioural and biological effects may persist.
- “Test carryover and ignore it if nonsignificant.” Such tests often have low power.
- “Period effects cancel automatically.” They can interact with treatment or missingness.
- “Crossover is always more efficient.” Attrition and long washouts can erase the advantage.
The within-person design is powerful because assumptions are strong, not because assumptions are unnecessary.
← Previous · Contents · Next →
9. Evidence audit
Audit eligibility, condition stability, reversibility, sequence generation, allocation concealment, period length, treatment onset, washout rationale, pre-period baselines, adherence, missing periods and analysis code. Verify that all randomised sequences are reported.
| Audit question | Evidence | Warning sign |
|---|---|---|
| Could effects reverse? | Mechanistic rationale | Permanent outcome |
| Was washout justified? | Kinetic and biological data | Customary duration only |
| Were periods comparable? | Protocol timeline | Season or method shift |
| Was attrition balanced? | Flow by sequence | Second-period exclusion |
← Previous · Contents · Next →
10. Questions readers often ask
How long should washout be?
Long enough for the clinically relevant effect to reverse, based on mechanism and evidence rather than a universal rule.
Can period 2 be discarded if carryover appears?
A first-period-only analysis loses the within-person advantage and may be underpowered; it should not be an improvised rescue.
When is crossover unsuitable?
When conditions progress rapidly, treatments cure or permanently change participants, outcomes are irreversible, or washout is unsafe.
← Previous · Contents · Next →
11. Reading routes and ecosystem connections
Begin with the Gold Standard of Education: learning that transfers into life, The Gold Standard and the How X Works hub. The subject foundation is the Gold Standard of Experimental Design.
Continue with treatment fidelity and estimands, attrition and missing outcomes, cluster randomisation, noncompliance and complier effects, and treatment-effect heterogeneity.
← Previous · Contents · Next →
12. Sources and practical checklist
Authoritative and primary references include the crossover-design review, the CONSORT extension for randomised crossover trials, and the International Council for Harmonisation E9 statistical principles.
- Confirm stability and reversibility.
- Randomise treatment sequence.
- Justify period and washout lengths.
- Measure period-specific baselines.
- Prespecify treatment, period and sequence terms.
- Track attrition by sequence.
- Use a parallel design when reset is implausible.