The Spoiler Effect: Polls vs. Reality

More often than not, third-party candidates have little effect on the ultimate result of an election. The eventual performance of most third-party candidates is lower than the margin of error of most polls (2%-4%), and so a poll that includes every single candidate running in the election (yes, even the Rent Is Too Damn High Party) provides no useful information about the likely results of the election. But when a third party candidate begins polling well (above the margin of error), pollsters would do well to include that candidate in their survey questions, so as to more closely reflect the range of choices voters have. A poll that gives a choice between Al Gore, George W. Bush, and Ralph Nader, for example, could give a different result from a poll that gives a choice between just Gore and Bush.

So in an election like the Senate race in North Carolina, where Libertarian Sean Haugh has been supported by about 10% of respondents on average in the polls that include his name, it'd probably be a good idea to continue putting his name in polls, since even considering sampling error Haugh's performance is probably fairly strong. Certainly it's strong enough to possibly tip the scales in favor of either candidate in a close race like North Carolina.

The more pertinent question, though, is this: will Sean Haugh actually receive somewhere in the neighborhood of 10% of the vote on November 4? I highly doubt it, simply because of the historical fact that third-party candidates rarely do as well or better than their poll numbers predict, at least at the statewide level. In presidential elections where third-party candidates polled strongly enough to have a chance at tipping the election, polls came fairly close to getting the final result correct: 

• 1968, George Wallace (American Independent-AL): The final Gallup polls in October and November both put Gov. Wallace at 15% nationally; he eventually received 13.5%.
• 1980, John B. Anderson (I-IL): Rep. Anderson received 8-9% support in polls conducted in October and November 1980; he received 6.6% of the vote nationwide.
• 1992, Ross Perot (I-TX): With 18.9% of the vote on November 3, Perot actually overperformed most polling, which gave him around 13-16% of the vote.
• 1996, Ross Perot (Reform-TX): Polls conducted in the last two weeks of October 1996 predicted about 6-7% votes would go to Perot, not too far off from his eventual 8.4%.
• 2000, Ralph Nader (Green-CT): All polls in the two weeks before the election gave Nader either 3% of 4%. It turned out to be 3%.

However, polling for statewide races has been rather more hit-and-miss than polls at the presidential level. Here's a sample of the most recent statewide elections for which 1) multiple polls listing three candidates as options were conducted in the month before the election, and 2) the eventual margin of victory was lower than the vote share the third-party received:

• 2008, Senate (Minnesota): In the month before the second-closest election in U.S. Senate history, Libertarian Dean Barkley polled at an average of about 17%; he received 15.2%.
• 2009, governor (New Jersey): Every one of 31 polls conducted in the month before Election Day overestimated--often by double digits--the 5.8% of the vote that independent Chris Daggett would receive.
• 2012, Senate (Indiana): Polls taken in the week leading up to Election Day gave 6% to Libertarian Andy Horning, who ended up receiving 5.7% of the vote.
• 2013, governor (Virginia): Last year's gubernatorial election ended up being closer than polls predicted, largely because Libertarian Robert Sarvis, instead of receiving the 9% average predicted by October polling, won the support of only 6.5% of Virginians.
• 2014, Senate GOP primary (Mississippi): Thomas Carey was undoubtedly a spoiler candidate here in that his 1.5% of the vote was enough to throw the election to a runoff; however, he did not match the 3-6% performance polling would have predicted.

Sometimes statewide polls do as well as you could expect them to: polls in Indiana in the lead-up to the Senate election in 2012 got it exactly right. Most times they overestimate by a bit. But in almost no cases do third-party candidates overperform their polling performances. In cases like Montana's Senate election from 2012, where Libertarian Dan Cox received 6.6% of the vote, there was no third-party candidate listed in polls; it was either a head-to-head matchup between Sen. Jon Tester and Rep. Denny Rehberg or it included a generic "other" candidate. 

The important takeaway is that third-party candidates rarely do as well as their poll numbers predict and almost never overperform, barring an eleventh-hour surge on Election Day itself. Responding to a poll over the phone is entirely separate from actually casting a ballot, and voters who are comfortable answering that they intend to vote Libertarian are not necessarily comfortable with actually voting Libertarian in the booth. So in North Carolina, don't expect Sean Haugh to break double-digits in November. He's doing well, but probably not that well. And it's only because of the closeness of the race that either candidate has to worry about Haugh: if the race weren't this close, Thom Tillis would not have much to fear from Haugh siphoning of libertarian Republicans.