Effect of Crossover on the Statistical Power of Randomized Studies