There are two problems: statistical knowledge being easily faked or misinterpreted and life being a multiple-repetition game. It is hard to apply the knowledge of "many X are Y and it is bad" when X is easier to check than Y in such a way as to not diminish the return on investment of X who work hard to not be Y. The same with the positive case: if you think that MBA programs teach something useful and think "many MBAs have learnt the useful things from MBA program" then getting into the program and not learning starts making sense. And we have that effect! http://www.freakonomics.com/2011/10/12/why-do-only-top-mba-programs-practice-grade-non-disclosure/

The problem is that one of the only ways to prove someone is indeed using statistical knowledge, on the handful of cases that we have forbidden it, is to analyse their patterns of behaviour, basically look at the recorded statistics of their interactions. Both the records and the results of such an analysis which can be easily faked and misinterpreted. Which means that if the forbidden statistical knowledge is indeed useful and reliable enough to be economical to use it, and someone else is very very serious about preventing it from being used, the knowledge will both be employed in a clandestine way and most of the economic gains from it will be eaten up by the cost of avoiding detection. This leads to a net loss of wealth. Say a for-profit company that spends 90% of the gains from forbidden knowledge on avoidance of detection, the governments spends half or a third of that amount to monitor the company. The company would be indirectly paying for government monitoring regardless if it used the knowledge or not. It is therefore irrational for the company to not use the particular forbidden set statistical knowledge in such a situation.