If i have two groups that are statistically different can I use a switching replications design to increase validity?