Why Did the Polls Miss McAuliffe's Margin of Victory?

With public polling averages indicating Terry McAuliffe would win yesterday's gubernatorial election in Virginia by at least a touchdown, many were surprised to see his Republican challenger Ken Cuccinelli come within three points of the win.

So... what happened? Twitter is abuzz with theories on what went wrong, ranging from a late-breaking Cuccinelli surge fueled by the rocky rollout of Obamacare's exchanges, to complications stemming from polling with a third-party candidate, to the difficulty of nailing the makeup of an off-year electorate.

Anyone making a concrete claim is overstating what we can know about why McAuliffe did less well on election day than the polls suggested. But I do think there are some clues we can find by digging into one of the last polls taken before the election: the Quinnipiac University October 28-November 3rd survey of likely voters in Virginia that showed McAuliffe with a six-point lead heading into election day.

Quinnipiac is kind enough to release both crosstabs and a breakdown of the demographics of their sample. And sure enough, Quinnipiac's sample makeup was not too far off the mark. Some 71 percent of the Quinnipiac sample was white, while exit polls estimate 72 percent of voters on election day were white. Quinnipiac had some eight percent in the "other/dk/na" camp for party identification, making it hard to compare apples to apples with the exit polls (where it appears "other" was not an option), but Quinnipiac had Democrats with a +5 advantage over Republicans on party ID, the same as the exit polls.

Where Quinnipiac and the exit polls tell different stories is in how independent voters broke. Quinnipiac's final poll had independents split evenly, 40-40, between Cuccinelli and McAuliffe, with 14 percent going to Sarvis. The exit polls showed Cuccinelli enjoying a sizable lead over McAuliffe among independents.

The public polling was far better in the other two contests on the Virginia ballot that did not have a third-party candidate. The Huffington Post Pollster model showed -- I kid you not -- 43.7 percent for Obenshain, 43.7 percent for Herring. (The race was so close it is currently headed for a recount.) The model for the Lieutenant Governor's race was similarly accurate, predicting a Northam win by twelve points (as the voters were counted, the actual margin was eleven).

This alone seems to suggest that the inclusion of the third-party candidate complicated the public poll's ability to get it right in the governor's race. However, within the Quinnipiac survey is some additional evidence to support the case. Quinnipiac asked respondents how likely they were to stick with the choice they'd just made. Cuccinelli and McAuliffe supporters said they were quite certain they'd stick with their man; only 2 percent of McAuliffe and 3 percent of Cuccinelli voters said there was a good chance they'd change their minds. Meanwhile, nearly one out of four Sarvis voters said they might change their minds.

The volatility of Sarvis voters seems to have added some uncertainty to the ballot test. Consider the many scenarios: Quinnipiac underestimated Cuccinelli's support among independents, and so it is possible that some of those independent Sarvis voters in the poll "came home" to Cuccinelli on election day. Or independent voters who leaned McAuliffe (but didn't really like him) felt safe casting a Sarvis "protest vote," figuring it wouldn't actually spoil the election given his large lead in the public polls. (How meta!) One could create theory after theory about what happened here. All we know is that Sarvis voters said they were still more up for grabs, tended to be more independent, and that the poll missed how independents would break -- therefore missing the ballot test.

The polls basically got it right in two out of three statewide contests, and missed the mark in the one race where a third party candidate added a higher level of uncertainty. While we may not know for sure, there seems to be some evidence that the "Sarvis factor" had something to do with why the polls got the size of McAuliffe's win wrong.