I have a/b tested this approach and have come to the conclusion that it is a little larger. I usually increase my estimates between 3/4 to double.