In the old days, life had many hardships. Among these: the need to wait until Election Day to determine who had won.
But now Big Data has saved us from this struggle. Even close races can now be predicted with mathematical precision.
We know, for example, with 98 percent certainty that Sen. Kay Hagan, an embattled Democrat, will win reelection in North Carolina next month. We are even more certain — 99 percent — that Sen. Mitch McConnell, a vulnerable Republican, will keep his seat in Kentucky. And we are darn near sure — 91 percent, to be specific — that Sen. Mary Landrieu (D-La.) will lose.
Throw all of these into our election model, add eye of newt and toe of frog, stir counterclockwise and — voila! — we can project with 84 percent confidence that Republicans will control the Senate next year.
The above data are from The Post’s Election Lab, run by George Washington University professor John Sides, but his is just one of several election models that claim to predict results with finely tuned accuracy. As of Tuesday afternoon, Nate Silver’s FiveThirtyEight, which turned the academic discipline of computer models into a media game, gives Republicans a 57.6 percent chance of taking the Senate. (Decimal points are particularly compelling.) The New York Times’s model goes with 61 percent, DailyKos 66 percent, Huffington Post 54 percent and PredictWise 73 percent. The Princeton Election Consortium gives a 54 percent advantage to Democrats . Apparently they forgot to add the toe of frog.
Some models have good records, and the theory behind them is sound. “Models help guard against any natural partisan bias,” Sides told me. “You use data to make decisions instead of your instincts or folklore. Instincts are sometimes correct, but on average data will be better.”
Yet there can be too much of a good thing — as when media outlets put too much faith in the numbers the models churn out. “Nate Silver’s team at FiveThirtyEight gives the GOP a 59 percent chance of retaking the Senate, down one point from last week,” ABC News’s George Stephanopoulos told his roundtable Sunday on “This Week.” “Was Nate Silver right to move it just an inch away from the Republicans?”
Modelers like to think Big Data can revolutionize election coverage the way “Moneyball” changed baseball recruitment. And some editors see it as an alternative to poll-driven coverage of politics. But models rely heavily (in some cases, exclusively) on polls and are subject to the garbage-in, garbage-out rule. Models also rely on historical patterns, and in politics, past performance is no guarantee of future results.
Recently, modelers have taken to discrediting each other publicly. Silver on Sept. 26 tweeted a shot at the “overconfident model” of Princeton’s Sam Wang: “Yesterday, Sam Wang’s snapshot had [Alaska Democratic Sen. Mark] Begich as a 99% favorite in Alaska. Today it gives him a 23% chance.” Silver later let people know Wang gave Sharron Angle a 99.997 percent chance to win the Nevada Senate seat that she lost to Harry Reid in 2010.
Wang, in turn, said Silver blew both the Nevada and Colorado Senate races that year.
Silver’s model has put Republican chances of winning the Senate between 53 percent and 65 percent over the past several weeks. The Times’s model has bounced around over a longer period from about 40 percent to the high 60s. The Post’s Election Lab has gone from 86 percent odds for Republicans in mid-July to a 51 percent edge for Democrats in mid-September before bouncing back up to the current 84 percent GOP advantage.
Smooth out the fluctuations, average the models together, and you end up with a reasonable forecast: Republicans are slightly favored to get the six seats they need to take the Senate.
Now compare that to Stu Rothenberg, an old-school political forecaster supposed to have been made anachronistic by Big Data. In August 2013, he predicted Republicans would gain three to six Senate seats, gradually moving that up to his current prediction of a five-to-eight-seat Republican gain, which he has held since August. He was earlier and more consistent than the models.
Rothenberg told me the models “hype” and are “intellectually deceptive” because they “convey a sense of mathematical certainty that is simply misleading.” Rothenberg uses polling and other data, but he also interviews candidates and is “humble enough to know that every election cycle is different.”
Charlie Cook, another old-school forecaster, thinks the Big Data models have a place, but they put too much science in politics. “They err,” he told me, in assuming “a precision and confidence that doesn’t exist in human behavior.”
I’m 98.7 percent certain Cook is correct.