Polls, Predictions and the 2008 Election

On Thursday ACS was pleased to have Nate Silver of FiveThirtyEight.com speak on his work during the 2008 election cycle, the methodology of poll analysis, and his own plans for the future. Silver, who recently signed a two-book deal with Penguin Group USA, outlined the necessary considerations in political forecasting and how to account for the natural bias exhibited by polls.

FiveThirtyEight.com was created out of frustration with the unsophisticated treatment of polls in the national media, and as an attempt to offer a context that would distinguish different polls and hold polsters accountable for their results. Citing the bad apple effect, where outlier polls garner attention due to their sensationalist results, Silver showed how a flawed polling methodology, such as out-of-date likely voter models or narrow data collection practices, could lead to skewed reporting.

Using various statistical models, including trend adjustment, simulations, and tipping point states, Silver accurately predicted the winner of 49 of 50 states in the 2008 presidential election. But in keeping with his mission is to uncover the hidden agenda of numbers, Silver did not gloat in these results. Instead, he showed that most polls did a good job in this election cycle, and there was only a point of difference between the major poll aggregators. However, while there might not be a significant difference in forecasting, it is in the details that Silver believes there is the most to learn about electoral predictions.

Silver frankly acknowledged that there may be no necessary social purpose for polling, and that he does it for the fun of the numbers. He is turning his attention to new forecasting exploits, and has two upcoming books, treating the art of prediction, including weather forecasting and the science of fashion trends, and the nitty-gritty mechanics of electoral politics. We eagerly look forward to both.