Crossover Trial Design
Also known as: within-subject crossover, cross-over design, AB/BA design, Çapraz Desen (Crossover Design)
A crossover design is an experimental design in which each participant receives all treatments under investigation, but in a different sequence and across separate time periods. Each subject thus acts as their own control, which substantially reduces between-subject variability and allows efficient treatment comparisons with smaller sample sizes. The approach has been central to clinical pharmacology and comparative research since the mid-20th century, with foundational methodology codified by Senn (2002) and Jones & Kenward (2014).
Key highlights
- Each participant serves as their own control, eliminating between-subject confounding and dramatically increasing statistical power relative to a parallel-group design of the same size.
- Requires substantially fewer participants to achieve the same power, which is critical when study populations are rare, recruitment is costly, or exposing large numbers to experimental treatments raises ethical concerns.
- Enables direct within-person comparison of treatments, which is the most informative comparison possible in individual-difference-heavy domains such as pharmacokinetics.
- Williams sequences extend the design to three or more treatments while maintaining balance, making crossover approaches competitive with factorial designs for multi-arm trials.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use a crossover design when each participant can receive all treatments sequentially, the condition under study is stable (chronic rather than curable), and a sufficient washout period can be introduced between treatment periods to eliminate carryover. It is most powerful for chronic-condition pharmacology, bioequivalence studies, and any setting where between-subject variability is large relative to within-subject variability. Core assumptions are: (1) no residual carryover from one period to the next — guaranteed by an adequate washout; (2) the period effect is included in the model to account for temporal trends; (3) the sequence assignment (AB vs BA) is randomized; (4) the treatment effect is the same regardless of which period it falls in (no treatment-by-period interaction beyond carryover). A minimum of about 12 participants per sequence is recommended for reliable estimation. If the condition can be cured by one of the treatments, a parallel-group design is more appropriate.
Strengths & limitations
- Each participant serves as their own control, eliminating between-subject confounding and dramatically increasing statistical power relative to a parallel-group design of the same size.
- Requires substantially fewer participants to achieve the same power, which is critical when study populations are rare, recruitment is costly, or exposing large numbers to experimental treatments raises ethical concerns.
- Enables direct within-person comparison of treatments, which is the most informative comparison possible in individual-difference-heavy domains such as pharmacokinetics.
- Williams sequences extend the design to three or more treatments while maintaining balance, making crossover approaches competitive with factorial designs for multi-arm trials.
- Carryover (residual) effects can confound treatment estimates if the washout period is inadequate, and detecting versus correcting for carryover is a persistent methodological challenge.
- Only applicable when the condition is stable enough to persist across multiple treatment periods; conditions cured by the first treatment or subject to spontaneous remission invalidate the design.
- Period effects — systematic changes in the outcome over time unrelated to treatment — must be modeled; if they interact with treatment effects the analysis becomes complex.
- Dropout between periods can bias results because participants who complete all periods may differ from those who do not.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
What is a washout period, and how long should it be?
A washout period is a deliberate gap between treatment periods during which any residual biological effect of the previous treatment is allowed to dissipate. Its required length depends on the half-life of the substance or intervention: a common rule of thumb is at least five half-lives of the drug or, for non-pharmacological interventions, a period long enough for the outcome to return to baseline. If washout is too short, carryover effects will confound the treatment comparison.
What is the Williams sequence, and when is it needed?
A Williams sequence is a balanced Latin-square arrangement used when there are three or more treatments. It guarantees that each treatment appears in each period the same number of times and that each treatment is preceded by every other treatment equally often across sequences, thereby eliminating both first-order carryover confounding and period effects from the treatment estimate. For two treatments the simple AB/BA design already achieves this balance.
What happens if the carryover test is significant?
A significant sequence-by-period interaction is interpreted as evidence of differential carryover. In the two-period AB/BA design, Senn recommends using only the first-period data for the treatment comparison, but this reduces power substantially and makes the study equivalent to a parallel-group design. Senn also argues that the carryover test itself is often underpowered, so adequate washout is a better preventive strategy than relying on the test.
How does a crossover design differ from a repeated-measures ANOVA?
Repeated-measures ANOVA is the analysis technique applied to data collected from the same subjects over time; a crossover design is the experimental structure that determines how treatments are assigned across those time points. In a crossover trial, the sequence in which treatments are administered is randomized and explicitly modeled, whereas in a general repeated-measures study there may be no treatment assignment or the order may be fixed.
Sources
- 1.Senn, S. (2002). Cross-over Trials in Clinical Research (2nd ed.). Wiley.ISBN 978-0471496533
- 2.Jones, B. & Kenward, M. G. (2014). Design and Analysis of Cross-Over Trials (3rd ed.). CRC Press.ISBN 978-1439861424
You have read it. What now?
Cite this page
ScholarGate. (2026, June 1). Crossover Design. ScholarGate. https://scholargate.app/experimental-design/crossover-design