The point here is that "local effects cancelling each outer out" sum up to what happens statistically on the whole. You need to account for all of these.