Can Nate Silver and friends nail their presidential predictions?
FiveThirtyEight statistician and blogger Nate Silver has taken fire for his polling aggregation methods. But he and others may be able to hit the final numbers spot on. CNET tracks the data.
Anyone who's even remotely interested in this year's contesthas no doubt seen countless polls, many of which have shown the incumbent in the lead, while many others have given the nod to the challenger.
In recent weeks, many headlines have declared the election a tossup. A common narrative being spread in newspapers, on blogs, on social media, and on TV nationwide, is that no one will have any idea who will be elected president until all the counting is done because the race is simply too close to call.
But adherents of another school of thought are scoffing at that notion. To them, it's long been evident that President Obama is cruising comfortably to a second term, never mind the fact that the polls don't start closing on the East Coast for 13 more hours.
Welcome to the world of polling aggregators, a group of people and organizations who lay claim to confidence, based on math and science, that they can predict what is likely to happen.
Looking at eight national polls released in the last few days, it's easy to see why many would conclude that the race is a virtual tie. Those eight polls, from organizations as august as NBC News, the Wall Street Journal, ABC News, the Washington Post, Gallup, and others, showed, in no particular order: three ties, two Romney one-point leads, an Obama one-point lead, and two three-point edges for the president.
The poll aggregators, however, report no such ambivalence. For five of the most well-known aggregators, the probable winner of the election goes like this: Obama, Obama, Obama, Obama, and Obama.
Nate Silver and FiveThirtyEight
Perhaps the most well-known of these aggregators is Nate Silver, who runs the FiveThirtyEight blog -- a reference to the total number of electoral votes available -- for The New York Times and who first established a reputation for accurate election modeling during the 2008 Democratic primaries, and then bolstered it by nailing 49 out of 50 states in that year's general election. This time around, Silver has forecast Obama's re-election consistently since launching the 2012 version of his so-called Political Calculus in June, relying on a complex recipe that starts with endless polls, weights them by historical accuracy, throws in a dash of economic indicators, sprinkles on some demographic data, and stirs it all together in order to run thousands of simulated elections.
The goal? To predict the probable winners of each state -- without question a more relevant data point than any national poll, given that the winner of the presidential election is determined by who is first to 270 electoral vote, which are awarded on a state-by-state, mainly winner-take-all basis. The national popular vote is nothing more than a symbolic beauty contest.
But Silver has paid the price for bucking the ever-equivocating national polls. In recent weeks, Silver regularly pegged the odds of Obama winning at more than 70 percent (it's currently at 92.2 percent) -- even when Romney seemed to pull ahead for a time in national polling. Some began to mock Silver's methods, despite (or perhaps because of) his being thought of in Democratic circles as a genius, or a savant.
Silver, though, is a statistician, and to him, his numbers told everything he -- and his readers -- needed to know. "You may have noticed some pushback about our contention that Barack Obama is a favorite (and certainly not a lock) to be re-elected," he wrote on November 2. "I haven't come across too many analyses suggesting that Mitt Romney is the favorite. (There are exceptions.) But there are plenty of people who say that the race is a 'tossup.' What I find confounding about this is that the argument we're making is exceedingly simple. Here it is: Obama's ahead in Ohio."
Indeed, almost all serious political experts would agree that the likely winner tonight will be the man with the most votes in Ohio. Though both candidates have paths to victory that don't go through Ohio, winning the Buckeye State is pretty much the easiest way to capture the White House. Plus, no Republican has ever won the presidency without taking Ohio.
But what Silver's statement -- which some might see as a tad arrogant -- really attests to is the reality that in the United States, presidential elections are won state by state, not at the national level. And with remarkable unanimity, the leading aggregators have consistently concluded that polling in the swing states -- Colorado, Florida, Iowa, Michigan, Nevada, New Hampshire, North Carolina, Ohio, Pennsylvania, Virginia, and Wisconsin -- has favored Obama. And that in the vast majority of the 100,000 simulated elections Silver runs each day, the president has come out on top.
Still, each of the aggregators -- FiveThirtyEight, TPM PollTracker, HuffPost Pollster, RealClearPolitics Average, and the Princeton Election Consortium -- has its own methods, and its own results. While each of the five went into Election Day predicting a second Obama term, their numbers vary a bit.
With that in mind, and with many pundits seeming to question the entire concept of poll aggregation -- choosing instead to point to the contradictory national polls as evidence of a tossup election -- CNET is presenting each of the five aggregators' (near) final predictions for each candidates' electoral vote totals, their popular vote percentage, and the margin of victory in each swing state. And for good measure, we've thrown in their predictions for the balance of power in the U.S. Senate as well. The chart above shows those data as of noon Pacific time yesterday. Below, you can see data as of 10 p.m. PT last night. Note that among other changes in the numbers, two of the aggregators have moved Florida from Romney's column to Obama's since yesterday, based on last-minute polls.
CNET will attempt to do a final update today, if there is additional data. And tomorrow, or when there is enough data to proceed, we will collect the actual results, compare them to the five aggregators' predictions, and see which of them was, finally, the most accurate. If Romney emerges as president-elect, each of the aggregators will have come up well short of the mark.
Below are links to explanations (or brief explanations) of the methodologies of each of the five polling aggregators:
TPM Poll Tracker (Click "Methodology" in the upper right)
RealClearPolitics Average: To arrive at its polling Average, RealClearPolitics simply takes the average of a recent series of national polls. To get its prediction for the Electoral College, it uses the average of each state's most recent polling to determine which candidate would win that state's Electoral Votes.
Update (Tuesday, 10:43 a.m. PT): Below is an updated chart reflecting some last minute modifications to some of the polling aggregators' numbers. If there are further modifications during the day, CNET will update again, and the numbers we use to compare the performances of the five aggregators will be the final ones we post today.