Tuesday was a brutal night for the Democratic Party – yet as Republicans celebrate their impending takeover of the Senate and an increased majority in the House of Representatives, the Dems aren’t the only ones licking their wounds.
The election returns were, to all appearances, a bit of a stunner to those in the business of predicting outcomes. Nowhere was that more obvious than in the state of Virginia, where in the weeks leading up to Nov. 4, incumbent Sen. Mark Warner was enjoying polls that gave him with a double-digit lead over challenger Ed Gillespie.
Fast forward to Wednesday morning. Warner, the former governor, is now facing a possible recount, since his lead over Gillespie, with almost all ballots counted, is less than one percent of the more than two million votes cast.
Other returns also took many prognosticators by surprise, including some of the most respected names in the business.
Proper respect, however, has to be shown to University of Virginia political scientist Larry Sabato and the staff at the school’s Center for Politics. More than any other major election forecaster, Sabato’s Crystal Ball team laid it on the line Tuesday. Doing away with the wishy-washy toss-up category, as is their wont, they took a stand on every House, Senate, and gubernatorial election held last night. Unfortunately, a lot of those stands turned out to be wrong.
Sabato’s crew had a particularly rough go of it in the governors’ races. The most leeway the Crystal Ball allowed itself in expressing doubt about an outcome was to identify a race as one that “leans” in a particular party’s direction.
They identified four races as leaning toward the GOP and all four went that way. But the vote tallies for some suggest that “leaning” was a little too generous toward the Democrats’ chances. In Georgia, incumbent Nathan Deal won handily, by an 8-point margin. Wisconsin Gov. Scott Walker was reelected with a 5-point margin, and Michigan Gov. Rick Snyder crossed the line with a 4-point lead. All three won more than 50 percent of the vote.
It was in the “leans Democrat” category, though, where the Crystal Ball ran into the most trouble. Sabato had nine races leaning toward Democrats, and of the seven that have so far been decided, Republicans took five. In some cases, they won going away. Maryland Republican Larry Hogan won his state’s governorship by a 9-point margin. In Illinois, businessman Bruce Rauner ousted Democratic incumbent Pat Quinn with 5 points to spare. Embattled right-wingers Sam Brownback of Kansas and Paul LePage of Maine both survived with a four-point cushion.
The Center for Politics performed a little better in the close Senate races. It identified only two races as leaning Democratic and split the pair. Democratic incumbent Jeanne Shaheen took New Hampshire by four points, while North Carolina Democratic incumbent Kay Hagan was sent packing by two points.
In the races identified as leaning toward Republicans, Sabato is batting 1000 among the races already called. But some of the victory margins, such as Pat Roberts’ 11-point win in Kansas and 8-point wins by both David Perdue in Georgia and Joni Ernst in Iowa, suggest that in reality these races were doing a lot more than “leaning R.”
Early on, Sabato and company made clear that forecasting is a tricky business. “While we’re proud of our overall record over the years, we always miss a few calls, sometimes more than a few,” they wrote on their website.
“The day after any election (or runoff or recount), when the actual winners are known, it all seems so obvious in retrospect. But of course it isn’t, pre-election. We apologize in advance — and we’ll do it again post-election — for all the races we will inevitably miscall. Our goal is perfection, and we’ll achieve it on the proverbial twelfth of never.”
“Finally,” they concluded, “let’s keep the enterprise of election prediction in perspective. We’re not exactly curing cancer. This is fun for us and, we hope, entertaining for you.”
Sabato’s colleague in the election forecasting business, Charlie Cook, allows himself and his team a little more wiggle room on making predictions by classifying close races as “toss-ups.” And the Cook Political Report made good use of that category Tuesday, classifying 13 governors’ races as toss-ups in their final analysis.
So far, 10 of those races have been called, and with 9 of them going to Republicans, it doesn’t take a statistician to tell you this series of “toss-ups” doesn’t appear to have used a fair coin.
Among the Cook Political Report’s gubernatorial toss-ups were Hogan’s 9-point win in Maryland, Deal’s 8-point victory in Georgia, and Rauner’s and Walker’s respective 5-point wins in Illinois and Wisconsin.
In the Senate races, Cook’s team also made liberal use of the “toss-up,” dumping nine races into that neither-here-nor-there limbo. Trouble was, a few of them clearly didn’t belong there.
In Arkansas, incumbent Sen. Mark Pryor took an absolute drubbing from GOP Rep. Tom Cotton, losing by 18 percentage points. In Kansas, as mentioned earlier, Sen. Pat Roberts won by 11 points, and Republicans in Iowa and Georgia sailed through with 8-point margins.
Over at FiveThirtyEight.com, the domain of statistician Nate Silver – whose 50-for-50 call of the 2012 president state-level results earned him wide acclaim – results were also a little surprising. Silver, while not expressly predicting results, offered percentage chances of victory for all the Senate races as well as a range of vote distributions in which the results has a predicted 90-percent chance of falling.
The FiveThirtyEight forecasts that predicted a better than 50 percent chance of victory for a candidate were only wrong in two cases so far: Kansas, where it gave Independent Greg Orman a whisker-thin advantage, and North Carolina, where it gave the nod just barely to Kay Hagan.
Still, as a measure of how surprising Tuesday night’s results were, it’s interesting to note how many vote tallies fell well outside Silver’s 90 percent certainty range. Republicans in Kansas, Iowa, Georgia, and Arkansas, for example, all won with margins outside Silver’s range. Some, like Cotton, won by a considerable distance. And going into last night’s vote count, Virginia’s Warner would have looked at Silver’s ranking and learned he had a better than 99 percent chance of winning. A margin of less than 1 percentage point wasn’t really in the cards, according to Silver.
If he weren’t already a millionaire, you’d bet Gillespie wishes he’d bought a lottery ticket.
To be fair to Silver, his system relies purely on aggregating poll results (though he applies some weighting to allow for historical reliability). So if polls show an overall bias, his findings will too. That’s exactly what he said on Wednesday.
“Based on results as reported through early Wednesday morning … the average Senate poll conducted in the final three weeks of this year’s campaign overestimated the Democrats’ performance by 4 percentage points,” he wrote. “The average gubernatorial poll was just as bad, also overestimating the Democrats’ performance by 4 points.”
Silver says evidence shows that systematically skewed polls are more common in elections than we might think.
“[P]olling bias has been largely unpredictable from election to election,” he wrote. “Beyond the shadow of a doubt, the polling was biased against Democrats in 1998, 2006 and 2012. However, just as certainly, it was biased against Republicans in 1994, 2002 and now 2014. It can be dangerous to apply the ‘lessons’ from one election cycle to the next one.”
Top Reads from The Fiscal Times: