He's saying that if you consider the race results as a control of sorts, and you operate under the assumption that there is nothing inherent in race that would create the results shown, it is obvious that the variable being measured is not well controlled, and there is plenty of noise in the results.