Crossover Field Experiment — Within-Subject Design in Real-World Settings

There is no universal rule; the washout period should be long enough for the treatment's effect to dissipate fully before the next period begins. For behavioral or attitudinal interventions this may be weeks to months; for physiological interventions it is often defined by the biological half-life of the treatment effect. If theory or prior evidence cannot justify a plausible washout, consider a parallel-group design instead.

Both follow the same units over time, but a crossover field experiment actively administers randomized treatments in each period and uses random sequence assignment to enable causal inference. A panel study observes naturally occurring variation without researcher-controlled treatment assignment, which limits causal claims. The crossover design is experimental; the panel study is observational.

Yes — this is called a cluster crossover design. The cluster (e.g., a school) receives one treatment in period 1 and the other in period 2, with random assignment of cluster-sequence. Analysis must account for clustering within both periods. This is especially useful in field settings where individual randomization is logistically or ethically impractical.

If statistical tests reveal significant carryover, the standard recommendation is to use only first-period data for inference, which collapses the analysis into a conventional parallel-group comparison. You lose the within-subject efficiency advantage, but the validity of estimates is preserved. This possibility should be anticipated in the power calculation at the design stage.

Crossover designs are a special case of repeated measures, but with the critical feature of randomized treatment sequence. Repeated-measures designs in general may not involve different treatments — they may simply measure the same outcome at multiple time points. In a crossover, distinct interventions are applied in each period, and the sequence is randomized to enable causal comparison.