Crossover Factorial Experiment — Crossover Factorial Experimental Design

A repeated-measures ANOVA is an analytic technique applied to within-subject data; a crossover factorial experiment is a design strategy that structures how treatments are assigned and ordered across periods. The crossover factorial experiment uses randomized, balanced sequence assignment to separate treatment, period, and carryover effects — going well beyond the simple repeated-measures structure. Crossover factorial data are typically analyzed with mixed-effects models that explicitly model these components, which subsumes repeated-measures ANOVA as a special case.

Because each participant contributes data under every treatment combination, the variance in effect estimates is substantially reduced. As a rough guide, a crossover factorial design can require 50–75% fewer participants than an equivalent parallel-group factorial design to achieve the same power, depending on the within-subject correlation. Power calculations for crossover designs must account for the number of periods, the intra-class correlation, and the carryover variance.

If the test for differential carryover is statistically significant, the crossover analysis is compromised. The standard response is to restrict analysis to data from the first period only — which effectively converts the study into a parallel-group factorial design using only first-period observations. This greatly reduces power and sample size, which is why preventing carryover through adequate washout is far preferable to correcting for it after the fact.

Yes, but each additional factor multiplies the number of treatment combinations and thus the number of periods each participant must complete. A 2x2 factorial crossover requires four periods; a 2x2x2 requires eight. Participant burden and dropout risk grow accordingly. In practice, crossover factorial designs with more than two or three factors are uncommon because the required number of periods becomes logistically prohibitive. Fractional factorial crossover designs can reduce the number of conditions at the cost of confounding higher-order interactions.

In a full crossover factorial design, yes — the design assumes every participant experiences every combination, which is what enables the within-subject comparisons. Dropouts who do not complete all periods contribute incomplete blocks, requiring missing-data methods (typically mixed-effects models with maximum likelihood estimation) to use their available data without discarding them entirely.