Saturday, March 27, 2010

Autistic Teen Picks First Two NCAA Rounds Perfectly

NBC News report "Autistic Teen Picks First Two NCAA Rounds Perfectly" (March 25th, 2010) appears to report a non-event. It notes that an autistic individual picked the first two brackets perfectly. Corrections suggests that is expected to happen about 4 times every year.

In fact, he picked every game through the first two rounds correctly. The odds of anybody doing that? One in 13,460,000, according to It's easier to win the lottery. Twice.

According to a survey, about 58 million people will fill out a bracket (more brackets may be filled out, as the average per participant is presumably greater than or equal to one). If the tournament is based purely on chance, we expect a one-in-thirteen-million event to happen four times, on average, each trial of 58 million. Indeed, the NCAA tournament is not composed of random events, and we should likely expect this to happen more than four times, not less.

Finally, while it's easier to win the lottery, according to NBC News, Corrections notes that the winning of the lottery is an everyday event.

1 comment:

  1. Just one note of caution. The backdrop of independent draws is not at one end of the spectrum in terms of likelihood. It is tempting to say that, compared to random draws, people's informed guesses will fare considerably better, and this is true for the individual's odds of getting it all correct. But it doesn't automatically scale up to the group.

    The correlation of brackets with the true odds and across people are BOTH important. For instance, if everyone makes exactly the same bracket (perfect cross-correlation), then we should almost never see a perfect bracket, EVEN IF the brackets are all constructed from the true most likely outcomes.

    We are not in such an extreme case. A more realistic adaptation would be to consider that if there were a few upsets that almost nobody would ever have predicted, then it is quite unlikely that anyone would get the bracket exactly right. It is hard to say how realistic this is, without seeing actual data on people's actual prediction behavior.