Anti-Buzz: Wish Fulfillment

Andrew has been writing Anti Buzz for 4 years resulting in almost 200 articles. For the next several weeks we will revisit some of these just in case you missed it:

So I like to harp about statistics and science and how they get portrayed in common media. This stuff matters to to you, of course, because the care you provide is, ultimately, justified by studies and findings and conclusions drawn from medical science and statistics. I came upon this joke image a few days ago.


For some context, the Toronto Maple Leafs are a team in the NHL that has, for about 15 months now, consistently done better than stats geeks think they deserve to. That projected season record, if it were to come to fruition, would be the best regular season in the NHL by a mile. Of course, it’s all a joke, because any reasonable person understands that whether or not a team comes from a “C-name” city has nothing to do with the outcome of a hockey game. Besides, The Leafs beat the Calgary Flames 4-2 on Wednesday, so this silly C-curse is clearly hogwash.

The elephant in the corner is that, mathematically speaking, there is nothing incorrect or inappropriate about the above analysis. We “know” that letters in team names are completely disconnected from the cause-effect chains that govern hockey. And yet, there’s no rule that says what factors we can and cannot consider when performing an analysis – if there were, it would be a bane on innovation – and so we are left with the fact that, well and truly, the Maximum Likelihood Estimate for the probability that the Toronto Maple Leafs will defeat a team from a city that has a name that doesn’t begin with the letter ‘C’ is 100%.


Even correcting with a Bayesian prior, (In this case, assume you spot each category a free win and free loss), We still have the Leafs at roughly an 89% chance of beating any non-C team, and only a 17% chance of beating any C team. The projected season record in this case would be about 65 wins, still an all-time best for the NHL. Winning and playing against non-C teams is perfectly positively correlated, with a correlation coefficient of 1. The list of stupid statistics you can generate on this goes on, and none of them are invalid outside of the fact that our intuition rightly tells us they are invalid.

The criteria of C vs. non-C teams is patent nonsense, but this should tell you something about the statistical models we use to make important decisions with: they are still very much a product of human biases. I’m giving you a silly example to play with, but more legitimate analyses are still plagued by the problem that we choose what to examine.

newface-620x461To bring this back around to technology, the new ‘big data’ phenomenon is in part about just keeping all the information we can and letting the computer figure out what is and isn’t important. This is not as easy as it sounds, and there’s a bit of a chicken and egg problem in that generalized, ‘unbiased’ algorithms still need to have certain parameters to be set, and those parameters are set by a human.

To bring this in a more critical direction, my skin usually crawls when arrogant literati like Jean Baudrillard say stuff like: “Like dreams, statistics are a form of wish fulfillment.” (First of all, nobody would tolerate a mathematician being so curtly dismissive of an entire intellectual endeavor – they would be quickly branded as “emotionless” or “cold” or whatever else people say when they don’t want to listen mathematicians and scientists).

Yet, sometimes, like in the case of the Toronto Maple Leafs, the statistics really are a form of wish fulfillment.

Leave a Reply

Your email address will not be published.