11 November 2016

Statistical Mumbo-Jumbo

After Game 4 of the World Series, some guy at 538.com said "The Cubs Have A Smaller Chance Of Winning Than Trump Does".

15% chance, he says.

Well, they won, didn't they, and so did Trump, so everybody who believed this meaningless mumbo-jumbo got burned. But there is another reason to dismiss this "percent chance" business.

Assigning a 15% chance to a single event with a discrete outcome - either the Cubs win, or they don't - is quite meaningless.

It's one event with exactly two potential outcomes: win or lose. No other type of result is possible; you cannot end up with "15% of a win".

The concept is nonsensical. So what possible use is it to say a team has a 15% chance of winning one World Series?

Try placing a bet in Vegas on that.

Now it's possible to look back in history and feed all kinds of data into a computer and come out with a number like 15% because you have hundreds, or thousands, or even millions of historical events to draw from.

When looking back at such a series of discrete outcomes, using percentages is somewhat useful, to understand how often it has happened before.

They are not useful at all with a single discrete outcome in the near future.

I also think people make the mistake of thinking of everything as if it were just like flipping a coin ten times. Percentages may help (but not always) predict the result of a series of discrete outcomes: heads, tails, tails, heads, heads, heads, tails, tails, heads, tails. 10 flips, 5 heads, 5 tails. 50%, just like you'd expect. But even then, it's not always right. Flip a coin 1000 times, and see if you get exactly 500 heads. The percentage says you should. Maybe, maybe not. It's far from a certainty.

But predicting a winner of a single World Series is nothing like predicting 10 coin flips. It's one "flip", not ten. It can only result in this outcome or that, win or lose, nothing in between like "15% of a win". And it isn't a flip at all, it's a game contested by people who have both successes and failures, and big emotional ups and downs, impacted by impassioned leadership, from guys like David Ross (after that same Game 4) and Jason Heyward (in Game 7).

Historical data alone is never the whole picture.

If you want to rely on percentages for things like this, go ahead, but only to describe events in the past, a way to understand history, which is interesting but when you really think about it, has almost no bearing on these players, on these teams, playing these games, today.

via Ann Althouse "The Cubs Have a Smaller Chance of Winning Than Trump Does"