Analysis of the crossover design in the presence of residual effects

When a residual effect is suspected in a two-period crossover trial, an analysis of the first-period data is often chosen instead of the potentially biased crossover analysis. This paper indicates how the usual crossover test has to be interpreted correctly and that its bias has two different consequences, namely a conservative or a liberal test decision if a positive (carryover) or a negative (withdrawal) residual effect exists. A multiple testing procedure is presented allowing for simultaneous crossover and first-period analysis controlling the experimental error rate. This procedure together with the correct interpretation of the crossover test enables many useful applications of crossover designs.