Statistician Nate Silver has been under criticism for partisan bias because his model evaluating the presidential election has heavily favored Obama since it launched this summer. This criticism comes from a misunderstanding of statistics, both how they are calculated and what they mean.
As most polls show, this is a very close election. The popular vote will likely be between 1-2 percentage points. The closest swing states also may be decided by only 1-2 percentage points. Nate Silver’s model understands this and seeks to explain this with more depth than any pollster or pundit.
A single poll, for example may show Obama and Romney tied with a margin of error of plus or minus three points. This means, based on the sample population polled, there is a 95 percent chance that the true poll result is within three percentage points. There is a 95 percent chance, known as the confidence interval, that the true result should have showed Romney up three points and Obama down 3 points. There is also a five percent chance the actual result is outside of that 3 percent margin of error.
This is where poll aggregators like Real Clear Politics and Cook Political Report offer more insight. By simply averaging polls, the margin of error is reduced and the confidence interval increases. More polls means a larger sample size means a lower chance the sample population isn’t reflective of the full population.
Nate Silver takes his analysis a step further, by incorporating economic data, fundraising, incumbent status, and other, in his opinion, indicators of electoral success. While his model is in many ways, an opinion, it’s a very transparent one that allows others to test and verify his analysis.
Silver’s success came, in part, from correctly predicating 48 out of 50 states in 2008. But this mistakes Silver’s predictions and even his claims for success. He is providing odds, similar to odds makers for betting. When he says a state has a 51 percent chance of going for Obama and 49 percent chance for Romney, that means 51 out of 100 times, Obama will win that state. When Silver claims Obama has an 80 percent chance of winning this election, he’s still saying Romney has a 20 percent chance of winning. If Romney did win the election, that wouldn’t disprove Silver’s model. Since we can’t hold the election 25,001 times (as Silver does in his simulations), we can’t know for certain how accurate his model is.
Upsets happen, in politics, sports, and any situations with odds; because we expect the unexpected to happen occasionally. Thankfully we have tons of data related to presidential politics, from prior elections and thousands of polls for the current one. That’s why a model like Silver has created is believable, because it is based strictly at looking at prior and current data to make the best possible prediction. When unexpected outcomes happen, that only serves to improve the model.
Many in the media would probably love to vilify Silver if Romney were to win. Even if Silver fails to predict every state properly, that’s not his goal. He’s simply showing what the data shows – what is the more likely outcome. Unexpected things do happen, but you likely won’t expect them.