Follow by Email

Thursday, November 10, 2016

Don't Blame Nate Silver

After the Cubs dropped the 4th game of the World Series to the Cleveland Indians and fell behind 3 games to 1, Nate Silver's FiveThirtyEight published an article entitled, "The Cubs Have A Smaller Chance Of Winning Than Trump Does." At the time the Cubs had just a 15% chance of winning the World Series, which was less than Donald Trump's chance of winning the presidential race. Of course, the Cubs did come back to win the World Series, and Trump did win the race for the Presidency.

The victories by the Cubs and Trump highlight the nature of probabilistic models. When a model finds that such-and-such event has a 30% chance of happening (which was the probability FiveThirtyEight gave Donald Trump of winning on the morning of the election), there is a 1 in 3 chance that the event will happen. That is actually pretty high, a lot higher than most folks probably realize. To illustrate (as one of my colleagues put it), if you take a six-shooter, place bullets in two of the six chambers, and then spin it, how many of you would be willing to place the gun to your head and pull the trigger? Unless you have a death wish, I'm guessing not too many.

That is why people need to actually read Nate Silver and not just look at his models. It is in the discussion of the models that one can learn what probabilistic modeling is and isn't and how it can help us understand and make reasonable (but not perfect) predictions about future events. In fact, in the run up to the election Silver repeatedly cautioned about overconfidence, noting that Hillary Clinton's leads in the battleground states were slim and within the range of polling sampling error, which meant that if the polls were just slightly overestimating Clinton's support, the election could break in Trump's favor.

And that, of course, is what happened, and over the last few days much handwringing has been done about how the polls got the election "so" wrong. However, the polls weren't too far off. The average predicted margin of a Clinton popular vote victory was a little over 3.0%, and it looks like she will win the popular vote by a little over 1.0%. A 2.0% miss is not uncommon in presidential elections and is within the sampling error of most polls. It is also better than the polls did in 2012. It is just that in most years, the popular vote isn't as close as it was this year, so misses of 2.0% don't usually matter. However, to the dismay of some and the joy of others, this year the miss did matter ("What A Difference 2 Percentage Points Makes").

So, who's to blame? Well, you could blame the electoral college system or the Clinton campaign for making the race closer than it needed to be or the Democratic Party for nominating a highly unpopular candidate. But don't blame Nate Silver. It was't his fault.

No comments:

Post a Comment