Data journalism strikes gold in US election
Barack Obama has four more years as president. Florida is still to come in but he already has 303 electoral college votes, comfortably over the 270 mark needed to keep the keys to the White House.
Let’s rewind 24 hours. The American media largely agreed that the race would be very close. Many different pundits called the election for Mitt Romney. As it turned out, Romney lost and in terms of Electoral College votes it wasn’t very close at all. The final score is likely to be 332 to 206.
But there were a few amid the noise and haste who were quietly saying that the race wouldn’t be close at all. Foremost among them was Nate Silver of the blog FiveThirtyEight. Yesterday morning he gave Obama a 90.9% chance of winning and correctly predicted the results in all fifty states, as long as Florida stays Democratic. For Silver, the result has never been in any serious doubt. Since June he has almost always given the President a 60% chance or more of winning.
Another guy who got it spot on was Drew Linzer at Votamatic, who predicted yesterday a score of, you guessed it, 332-206.
How did these guys get it so right? And how did others get it so wrong?
Silver and Linzer used data. Lots of it. They crunched the numbers from polls across the US, took other factors into account and continuously refined their models as November 6th approached. Their methodologies look complicated, well beyond the average journalist’s mathematical ability, but you can’t argue with their results.
They also kept in mind that the election is decided by a handful of swing states. American presidential election days are like jigsaws with three quarters of the pieces already in place. The rest of the pieces fall into place on the day in the swing states, giving us the face of the president for the next four years. Silver said simply a week ago that Obama was most likely going to win because he was in the lead in Ohio. And so it proved.
The ones who got it wrong often rejected or underestimated data journalism. In one embarrassing effort, the author preferred instead to rely on a “strong hunch” and ignore opinion polling. Have a guess how that turned out.
Others cherry-picked statistics, extrapolated them across the whole country and called it for Romney that way. This one speculated that 75% of the 3 to 4% of undecided voters would vote for Mitt Romney. That’s potentially millions of votes backed up with no solid evidence.
A lot of the inaccurate punditry can be explained by wishful thinking. As this brilliant comment says, conservative pundits can live in an echo chamber, where their views are reinforced by their competitors and certain assumptions go unchallenged. The result is many of them seemed unable to accept an Obama victory was likely and discounted evidence that supported that possibility.
“Today, after four years of abject failure on virtually every level, Democrats still want to vote for Obama, but the passion is gone.”
“Today, after four years of failure and a steady march towards socialism, Republicans in general and conservatives in particular are the ones frothing, this time to depose a socialist king.”
See how focusing on “enthusiasm” can lead to sweeping statements and ultimately the wrong verdict.
They also overplayed the significance of historical factors. Take the high unemployment rate statistic, which says no president has been re-elected since Roosevelt in 1936 with an unemployment rate of higher than 7.2%. As the Washington Post points out, it’s a statistic that appears to be far more important than it actually is. At every presidential election certain trends are broken. In 2008 Americans elected their first ever African-American president and Missouri backed the loser for the first time since 1956. Rules, as they say, are made to be broken.
So what does this mean for 2016? It means more of a role for data journalism. Nate Silver brought in huge traffic for the New York Times this election, as well as the credibility of having such an accurate forecaster on your roster.
It shows that Americans (and other interested readers) want more than hunches. They want journalists to analyse the reams of polling data, cut through the bluster and tell (or show) us what is likely to happen.