Networks in overdrive as polls close
The mood is tense as election day comes to a close, especially at the TV networks as they compete to predict the outcome of the presidential vote.PT0M51S http://www.canberratimes.com.au/action/externalEmbeddedPlayer?id=d-28xld 620 349 November 7, 2012
AS AMERICANS head to the polls a parallel election battle at times as fierce is coming to a head – that between statistician Nate Silver and the rest of the political pundits.
Using "big data" techniques to crunch the numbers in a more detailed way than ever before, Silver correctly predicted the winner of 49 out of 50 states in the 2008 US presidential elections and all 35 Senate races that year.
The ability to analyse large amounts of data is starting to replace expert knowledge. In this case political journalism is being augmented (or arguably supplanted) by statistical techniques.
He made his name by creating a system for forecasting the performance and career development of Major League Baseball players but the US election shot him into superstardom. His long-term confidence in another Obama victory has made him an outlier and a Pariah among Republican supporters.
A political wizard or blinded by big data? Nate Silver has the political pundits in a lather. Photo: JD Lasica, Socialmedia.biz
His blog FiveThirtyEight, now part of The New York Times, accounted for 20 per cent of the traffic to the Times website on Monday. That is a huge number, considering the Times is the sixth-most-viewed news site in the US.
The Times public editor said Silver, a contractor, was probably its most high-profile writer at the moment.
Silver has been predicting a decisive Obama victory for the entire election cycle and his final prediction is that Obama has a 90.9 per cent chance of winning the election, with 313 Electoral College votes compared to Romney's 225.
Big data pioneer Anthony Goldbloom believes pundits are under threat. Photo: Hugh Hamilton
That's an even less bullish prediction than that made by others including the Princeton Election Consortium, but Silver has been unique in the scorn he's attracted from pundits - mainly those who support the Republicans.
To the Australian big-data pioneer Anthony Goldbloom, founder of Kaggle, the reason for the backlash is clear.
"The ability to analyse large amounts of data is starting to replace expert knowledge," he said. "In this case political journalism is being augmented (or arguably supplanted) by statistical techniques."
Goldbloom adds: "His approach involves taking public poll data from several sources, weighting it depending on things like recency and sample size (more recent polls get more weight as do polls with large sample sizes). He then makes some statistical adjustments, adds in extra data about the candidates (have they been senators or governors before, how much money did they raise...) and uses this data to simulate 100,000 fake elections to spit out each candidate's probability of victory."
Or as the comedian Stephen Colbert told Silver when he appeared on The Colbert Report on Monday night: "You're taking bread out of my kids' mouths because I'm a pundit. Those of us in the punditocracy make our bread and butter by telling people what the truth is as we see it from our gut."
Just last week Politico wrote that Silver could be a "one-term celebrity" as despite his success in predicting 2008 election results "this year's polls suggest a nail-biter".
MSNBC's Joe Scarborough was even more scathing, calling Silver "a joke". "Anybody that thinks that this race is anything but a toss-up right now is such an ideologue, they should be kept away from typewriters, computers, laptops and microphones for the next 10 days, because they're jokes," he said in the lead-up to the poll.
The Wall Street Journal's Peggy Noonan on Monday was still predicting a Romney victory. "Is it possible this whole thing is playing out before our eyes and we're not really noticing because we're too busy looking at data on paper instead of what's in front of us?," she wrote.
The national popular vote polls have been neck and neck for some time but this figure is less useful because the way the US system is set up means in reality only a few swing states end up deciding the election.
Even Australian political analysts such as the ABC's Antony Green are sceptical of Silver. Green said all Silver was doing was taking published polls state-by-state and turning them into probability predictions of an outcome.
"That's what everyone else does in words," he said. "If there weren't opinion polls he would just be a pundit because he wouldn't have the statistics."
Just a week or two ago Silver's probability of an Obama victory was closer to 80 per cent. "He's got one chance in five of being wrong which is the sort of odds you're only interested in if you're a gambler," said Green, adding the ABC's election model was probability-based but only awarded seats to candidates when the probability reached 99 per cent.
Green said nearly everyone else was predicting an Obama victory based on his perceived standing in key states and Silver was only different because he applied a probability number to his prediction.
"The problem with Nate Silver is that people don't understand what a probability measure is and that's what all this controversy's about."
But CSIRO big-data expert Alan Dormer said Silver was doing much more than take an average of the hundreds of past political polls. He said there was something "a bit deeper" going on and polls alone were often unreliable predictors of election results.
"He analyses a large amount of data that your average pollster probably cannot hope to understand by gut feel," he said.
Dormer said that Silver was using "Bayesian" techniques so that when new information came in – such as a bad news story or Hurricane Sandy – the new data could automatically be incorporated into his analysis.
He said that Silver, as in his baseball algorithms, has also analysed the game of politics carefully so that out of the large amounts of data feeding into his algorithm he can carefully choose what is relevant and what's not.
In baseball, for instance, Silver predicted which player would go on to become successful based on their earlier career and also matched characteristics of the younger players with characteristics of more successful players when they were young. Dormer said similar signals would be used in his political algorithm, such as which candidates in certain positions in past elections actually became president.
These extra signals were crucial because with polls the limited number of people surveyed wouldn't necessarily vote the way they said they would or their opinion could change over time.
"If [individual polls] is what political pundits feed on I think they've got to move on and I think they've got to start feeding on a more sophisticated analysis," said Dormer.
"Maybe in five years time there'll be several people in the same position as Silver and the political pundits will make their living by comparing the predictions of them rather than the polls."
As for today's election Silver admitted to Buzzfeed he had a lot riding on the outcome. "I'm also sure I'll get too much credit if the prediction is right and too much blame if it is wrong," he said.