Blog Post

Data doesn’t play politics — and most of it suggests Obama will win

Stay on Top of Enterprise Technology Trends

Get updates impacting your industry from our GigaOm Research Community
Join the Community!

Updated: Data doesn’t really care who will be elected as the next U.S. president. And of all the data points that political scientists and others trying to predict the election care about, most of them point toward Barack Obama being reelected on Tuesday.

Statistical models have been hot topic of conversation (maybe even argument) over the past couple weeks thanks to New York Times’ FiveThirtyEight blog author and statistician Nate Silver. He has become a lightning rod for controversy because of his whirlwind media tour promoting his new book and his model predicting that Barack Obama has a greater than 80 percent chance of winning Tuesday’s election. However, as Princeton University Center for Information Technology Policy Fellow Zeynep Tufekci astutely explained last week, Silver isn’t guaranteeing Obama will win on Nov. 6 — just that there’s a high probability he will — and whatever outcome his model comes to is very likely not influenced by partisan politics.

And believe it or not, Silver isn’t the only guy around who spends his time building statistical models to predict the outcomes of politcal contests — he’s just the most famous. There are plenty of academicians, predictive markets, hobbyists and others who also do this, all of whom use different data with different methods for assessing the importance of any given piece of it. With a few notable exceptions, most of them also foresee an Obama victory.

Here’s how they (as well as some less-scientific sources, such as Twitter) see the contest playing out.

Who has Obama winning

FiveThirtyEight: Silver’s final (I believe) model is in, giving President Obama an 86.3 percent chance of being reelected. A late-campaign-season bump came as national polls finally aligned with state polls in giving the edge to Obama.

The New York Times: Two of Silver’s colleagues and fellow data junkies at the New York Times, Mike Bostock and Shan Carter, published an interactive version of their own model on Friday. Based on analysis of the states still considered competitive, they see 431 paths to victory for Obama versus 76 for Mitt Romney.

InTrade: Arguably the world’s most popular prediction market, InTrade gives Obama a 67.2 percent chance of victory as of 10:54 a.m. Pacific Time on Monday. The percentages change in real time, but Obama hasn’t relinquished his role as favorite since the campaign season began early this year.

PredictWise: Prediction market PredictWise (headed by Yahoo’s (s yhoo) The Signal blogger David Rotshchild and without a real-money investment model like InTrade) gives Obama a 72 percent chance of winning as of 9:48 a.m. Pacific Time on Monday. The president’s chances have risen steadily over the past week.

Twitter: It’s not really a predictive model, but the Twitter Political Index does provide a point for gauging how the social media platform’s user view the candidates. As of Nov. 4, Obama has a positive sentiment rating of 59 versus Romney’s 53, although Romney has closed the gap by 9 points since the index launched in July.

Who has Romney winning

At least six political scientists and/or economists: An October symposium from the peer-reviewed journal PS: Political Science & Politics includes the results of 13 statistical models generated by noted political scientists, of which eight give Obama the edge and five give Romney the edge. One of the those favoring Romney, an historically accurate and economy-centric model from University of Colorado professors Kenneth Bickers and Michael Berry, generated quite a bit of buzz when released in August and giving Romney a 67 77 percent chance of victory. Bickers and Berry have since ran the model again and it showed an even higher likelihood of a Romney victory.

Another model with a track record of success — that of Yale economics professor Ray Fair — also gave a slight edge to Romney (although well within the margin of error) as of Nov. 2.


PoliticIt (tied): The Provo, Utah-based startup that aims to measure politicians’ online footprint and popularity actually has the candidates tied with It scores of 50. The scores appear to be old (I’m checking with PoliticIt for an answer on how recent they are) and don’t claim to be a predictor of campaign success, but its founders did claim a high correlation between higher It scores and victories in primary campaigns last spring.

Updated: PoliticIt responded with more-recent It scores of 49 for Obama and 48 for Romney. They are now projecting a win for Obama, as they detail in this blog post released Monday.

The NFL: If mere correlations (and superstition) are any indicator, Romney’s chances of victory are high after this weekend’s slate of professional football games. The oft-cited Redskins Rule regards the Washington Redskins’ performance in their last home game before the election — they lost, an outcome that suggests a Romney victory — but Chris Wilson at The Signal has concocted a series of correlations for the other 31 NFL teams, as well. All told, this year’s 19 of this year’s results foretell a Romney victory, 12 foretell an Obama victory and 1 won’t be decided until Nov. 18.


9 Responses to “Data doesn’t play politics — and most of it suggests Obama will win”

  1. Two additional prominent poll aggregation models are provided by (1) Pollster / Huffington Post and (2) Princeton Election Consortium. The former’s models are built by Simon Jackman (Stanford) and rely on both national & state polls, the latter’s by Sam Wang (Princeton) rely solely on state polls. These models’ probabilities of an Obama victory are 88% and > 95% respectively. Generally it looks like poll aggregation-based models like these and Nate Silver’s are assigning a high probability to an Obama victory, while models that heavily rely on economic fundamentals favor Romney.

    • Derrick Harris

      Thanks for noting the additional models. You’re definitely right about the polls vs. economy distinction. I haven’t done an analysis, but it would be interesting to see which ones perform better historically.

  2. Getting those data points is absolutely political in nature. Because the question asked, and the format of the response, elicits a particular pattern that is wanted from the individual(s) asking the question.

    Nice try, though. It’s like saying that Global Warming is “Settled Science”.

    • Taylen Peterson

      Assuming you’re not trolling* let me help you out. These data models come from hundreds (and some) of polls across both national and state samples. They have right-leaning and left-leaning polls. They take into account MoE. They recognize our Electoral College system and factor that into their estimates (which explains how Romney could be leading the national polls but still losing in a landslide in EVs). They ask questions in an assortment of ways. They even each other out by using so much data that outliers are smoothed out by the rest of the data (and outliers in the opposite direction). They use cold hard numbers, not emotions or feelings, to predict the likelihood of each winning. I get your comment that the numbers come from people with opinions/feelings on the matter, but that’s why pollsters don’t ask only 5 people, pollsters ask thousands of people, and those thousands of people across thousands of polls adds up to a very large sample size.

      Let’s say your local weatherman says there’s a 30% chance of raining. If it rained, you wouldn’t be entirely surprised, but the data says that it most likely won’t. If a basketball player is shooting 90% from the free-throw line and he misses, it’d be out of the ordinary, but not completely surprising. So just because the polls don’t look the way you want them to, doesn’t mean that they’re biased.

      (Note: not a single one of these is saying they are 100% going to be right. It’s pundits who are paid to have ridiculous partisan opinions who say that.)

      And I guess you could say GW isn’t settled…but when 97% of the experts in a field agree on it, I’m going to trust their scientific research, not Bill O’Reilly’s opinion.

      *As Fry would say: Not sure if trolling…or just stupid.

      • But the issue is that 97% of the experts do not agree on GW, but rather 97% of the experts that you read. Hence, why polling and date are skewed. If I put out a biased poll, it makes no difference the size of the sampling it is simply a defective data set. And to borrow your meteorological analogy, if I have only 1 place that I “trust” to give me weather information but they are consistently wrong, but I still pay attention to what they say, then what conclusion do you draw? The rest of the logical world blames both parties, the one who consistently gives the incorrect information and the person who continues to believe that information. Polling is all about an intended answer. The polls that have asked the more “neutral” questions tend to get panned by the media for doing so, and consequently don’t get reported or get tossed aside as a “high/low” in the average assessment of polls. So not only are you dealing with the majority of polling being heavily determinant, but then you have to add the other factor of biased reporting.

        And no…not a troll. Just merely a student of logical reasoning. That anyone thinks the majority of political polling is not skewed is living in some bizarre reality that simply is not the case.