Many have heard of Nate Silver’s prediction of the 2012 presidential election. For those of you who haven’t seen it yet, here was his prediction for last Tuesday, which may seem uncannily familiar:
It seems familiar because it bears a striking resemblance the actual results:
In fact, that’s all 50 states correctly predicted.
Given two equally likely options for each states, the chance to predict all 50 states correctly is one in , or one in 1.126 quadrillion. Granted, we already knew which direction states like Texas or Vermont would vote for, so for the sake of simplicity let’s consider only the 9 “swing” states: Colorado, Florida, Iowa, Nevada, New Hampshire, North Carolina, Ohio, Virginia and Wisconsin. That’s still a one in 512 chance to guess all 9 states correctly from sheer luck.
How did he do it? The answer is high-caliber aggregate statistics. When you conduct a poll, you are going to have very high uncertainty if you poll only a few people. But if you poll a lot of people, your prediction gets more accurate. And one way to poll many people at once is to aggregate the data from many, many polls.
In 2008, Silver’s prediction was accurate in 49 of the 50 states, missing only Indiana.