Why do we assume p=0.5 in determining the necessary sample size?