The impact of an extreme observation in a paired samples design