Nate Silver: Avoids errors by paying close attention to a poll's 'margin of error'
As Nate Silver reflects on the 2012 polling analysis that has made him a superstar among fellow statistical junkies, he offers one piece of advice above all else:
Pay attention to the “margin of error.”
Silver – author of the New York Times’ uber-popular blog, "FiveThirtyEight" -- addressed an overflow crowd Monday night at Washington University’s Graham Chapel (and a spillover room elsewhere on campus).
Silver is promoting his latest book, “The Signal and the Noise: Why so many predictions fail, but some don’t.”
Silver, who specializes in statistical analysis, started the FiveThirtyEight blog (named after the 538 electoral votes) as an independent website before the 2008 presidential election. (Earlier, he had become known for his statistical work in sports.)
The Times acquired him and his blog after Silver successfully predicted in 2008 how 49 of the 50 states would line up in the presidential contest.
In 2012, his record was 50 for 50.
Silver emphasized that there was no “secret sauce” that guaranteed his success. Rather, he methodically tracks the polling conducted in each state and aggregates the tallies of all of the respected polls, often tossing out the “outliers” that have widely different results from the majority of surveys in a given state.
He also plays close attention to the “margin of error’’ in each poll, which refers to the statistician’s estimate of how much higher or lower a particular number in a poll might be.
(For example: a poll showing Candidate A ahead by 5 points over Candidate B, but with a 5-point margin of error, means that each candidate’s number could be five points higher or lower. So Candidate A could be as much as 15 points ahead, or Candidate B could be ahead by 5 points.)
Silver took a lot of criticism from Republicans who, before the Nov. 6 election, accused him of unfairly skewing his predictions in favor of President Barack Obama. Those complaints disappeared after the election results showed Obama defeating Republican Mitt Romney by a larger margin than many pundits – but not Silver – had predicted.
Silver emphasized that his predictions were based solely on the numbers.
Silver blamed the partisan scrutiny directed at him on the increasingly partisan viewership habits of the public – especially those who care about politics. His visual display include surveys showing that Republicans and Democrats increasingly watch like-minded cable channels – particularly Fox (R) and MSNBC (D).
As a result, some political activists “had trouble reconciling themselves to the reality” of which candidates were really ahead, he said, and focused solely on polls – accurate or not – that best reflected the person’s political leanings.
Amateur poll junkies need to pay close attention to all the political polls conducted in their targeted area, he said, and be willing to acknowledge the likely inaccuracy of “outliers” – polls that have widely different results of others tracking the same contest.
However, he observed that the heavy press coverage given “outliers” can at times actually sway voters to shift into that direction – temporarily. Silver said that phenomenon occurred early last year when former Sen. Rick Santorum, R-Pa., briefly had momentum because conservatives began flocking to his camp after learning of a few pro-Santorum polls elsewhere.
Silver said that political polarization has spread beyond elections into other parts of life. Until 30-40 years ago, he said, gun ownership had little to do with political preferences. But now, Republicans are 2.5 times more likely to own guns than Democrats, he said.
Silver emphasized that he isn't out to affect events -- just accurately predict them.