Crossover Pretest-Posttest Experimental Design
Also known as: within-subjects pretest-posttest design, repeated-measures crossover design, AB/BA pretest-posttest design, crossover repeated-measures design
A crossover pretest-posttest experimental design is a within-subjects experiment in which each participant receives two or more treatments in a randomized sequence, with outcome measurements taken both before and after each treatment period. By serving as their own control across conditions, participants allow direct intra-individual comparison, dramatically increasing statistical power while reducing the sample size required relative to a parallel-group design.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use a crossover pretest-posttest design when: (1) the outcome is expected to return to baseline after treatment ends (the condition is chronic but stable, not rapidly progressive); (2) you need high statistical power with a small sample; (3) individual variability in baseline is large relative to the expected treatment effect. Ideal domains include pharmacology, pain research, exercise physiology, and educational intervention studies with stable skills. Do NOT use it when: carry-over effects are likely to persist (e.g., treatments that permanently alter the participant, surgical interventions, or learning tasks where improvement cannot be unlearned); the underlying condition changes rapidly or fluctuates erratically over time; ethical issues prevent exposing participants to a control condition after they have already benefited from an active treatment.
Strengths & limitations
- Each participant serves as their own control, eliminating between-subject variability and substantially increasing statistical power.
- Requires far fewer participants than an equivalent parallel-group pretest-posttest design to achieve the same power.
- The addition of pretests in each period allows detection and correction for period effects and secular trends.
- Within-person change scores are more sensitive to treatment effects than between-group comparisons at a single time point.
- Random sequence allocation (AB/BA or Latin square for more treatments) controls for order and period biases.
- Carry-over effects — residual influence of the first treatment on the second period — can invalidate the cross-period comparison if the washout is insufficient.
- Requires a stable, chronic condition; rapidly changing or self-remitting conditions violate the design's assumptions.
- Longer study duration than a parallel-group design increases participant attrition and the risk of time-varying confounders.
- If carry-over is detected, analysis must fall back to period-1 data only, losing the efficiency advantage of the crossover.
- Period effects (e.g., seasonal variation, practice effects on the outcome measure) must be explicitly modeled.
Frequently asked
How long should the washout period be?
There is no universal answer, but the washout must be long enough for all treatment effects — biological, pharmacological, or psychological — to return to baseline. In pharmacology, a minimum of five drug half-lives is a common rule of thumb. For behavioral or educational interventions, the required washout depends on how durable the learning or behavioral change is. A second pretest at the start of period 2 provides empirical evidence of whether washout was successful.
What if carry-over is detected in the analysis?
If the treatment-by-period interaction is statistically significant, carry-over is likely. The recommended strategy is to discard period-2 data and analyze only the period-1 change scores — effectively treating the study as a parallel-group pretest-posttest design. This halves the effective sample size and forfeits most of the efficiency advantage, which is why adequate washout planning before the study is essential.
How is this different from a simple crossover design without pretests?
A standard crossover design measures outcomes only at the end of each treatment period (posttests only). Adding a pretest at the beginning of each period allows computation of within-period change scores, which control for any residual differences in baseline between the two periods. This improves precision and makes it possible to detect and adjust for incomplete washout or secular trends.
Can this design be used with more than two treatments?
Yes. With three or more treatments, a Latin square or Williams design assigns participants to balanced treatment sequences so that each treatment appears equally often in each period. The statistical model becomes more complex, and the study duration increases proportionally, but the within-subject efficiency gain is maintained.
What statistical method should I use to analyze the data?
A linear mixed model is the most flexible and recommended approach. It can handle unbalanced data, missing observations, and explicitly models period effects, sequence effects, and carry-over alongside the treatment effect. For simpler two-period, two-treatment designs, the classical period-and-sequence ANOVA (Grizzle, 1965) is also used, though the mixed-model approach is generally preferred in contemporary practice.
Sources
- Senn, S. (2002). Cross-over Trials in Clinical Research (2nd ed.). Wiley. ISBN: 978-0471496533
- Campbell, D. T., & Stanley, J. C. (1963). Experimental and Quasi-Experimental Designs for Research. Rand McNally. link ↗
How to cite this page
ScholarGate. (2026, June 3). Crossover Pretest-Posttest Experimental Design. ScholarGate. https://scholargate.app/en/experimental-design/crossover-pretest-posttest-experimental-design
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Crossover Randomized Controlled TrialExperimental design↔ compare
- Factorial Pretest-Posttest Experimental DesignExperimental design↔ compare
- Latin Square DesignExperimental design↔ compare
- Pretest-Posttest Experimental DesignExperimental design↔ compare
- Repeated-measures ANOVAStatistics↔ compare