07/23/2014 06:08 pm ET Updated Jul 24, 2014

HUFFPOLLSTER: Why Did Polling Miss The Georgia Runoff?


Another Republican runoff brings another miss for the polls. The Wisconsin gubernatorial race remains deadlocked, while Florida looks increasingly close. And polling confirms that Americans still like beer. This is HuffPollster for Wednesday, July 23, 2014.

POLLS MISS IN GEORGIA - Polling suffered another black eye in Tuesday's Republican runoff election in Georgia, as businessman David Perdue narrowly defeated Rep. Jack Kingston, 50.9 to 49.1 percent. Nine of the 10 publicly released pre-election polls gave Kingston a nominal lead over Perdue (the exception, a live-caller WPA Research internal for Perdue, correctly predicted the margin, giving him a 1-point lead). Two final surveys, conducted the week before the runoff, showed Kingston leading by 5 and 7 percentage points. The Pollster model's final estimate gave him a 5.6 percentage point lead, 47.3 to 41.7 percent. [Georgia election results, Pollster chart]

Poor polls? - The first explanation for the miscue involves an entirely valid critique about the low quality and partisan affiliation of much of the Georgia polling. Eight of the 10 polls (including the one completely non-partisan poll conducted by SurveyUSA) were conducted using automated, recorded voice calls to landlines supplemented with interviews conducted online drawn and mostly from non-random, 'opt-in' panels. Nine of the ten (including both polls using live interviewers) had either partisan sponsorship or was conducted by a firm with partisan ties. None of the surveys combined live interviewers, calls to both landlines and cell phones and non-partisan sponsorship. Three of the polls were produced for Atlanta's Fox5 by InsiderAdvantage, a company whose founder is a former Republican officeholder and political consultant. As HuffPollster reported in May, InsiderAdvantage uses a combination of automated landline calls and online interviews culled from opt-in panels and solicitations from Facebook advertising, a method that can be described, most charitably, as unusual. [HuffPollster]

-Nate Cohn: "If you're surprised that the polls are wrong... well I don't know why you're surprised" [@Nate_Cohn]

-Mark Mellman (D) "GA polls all wrong. Don't say I didn't warn sponsors." [@MarkMellman]

'So many bad polls out there' - The most pointed critique came from Chuck Todd and his colleagues at First Read: "Folks, we’re living in a political age where so much of the data is coming from surveys with either questionable methodology (like not reaching those with cell phones) or with partisan manipulation. And so it’s a reminder to take them with a grain of salt, especially in a low-turnout runoff. And it also means: Be very careful of these aggregation polling sites. They don’t always make the data better… there are actually polling firms out there who seem to have been created for the sole goal of influencing or balancing out these aggregation sites."

Our take - In heavily-polled general election contests that include surveys using higher quality methods, we have the ability to calibrate our polling models based on relatively trustworthy data. But yes, polling aggregation has its limits in a scenario like the Georgia runoff where virtually all of the data are questionable. HuffPost Pollster's approach will continue to evolve during the 2014 election cycle -- watch this space in the coming weeks for more details -- in part because of the deluge of partisan data. For now, however, we'll repeat the words of political scientist Jonathan Bernstein: "Even if we had a single, gold-standard pollster, and even if we all could agree on who that was, we would still be subject to the kinds of standard random errors that are inherent in even the best polls. Given those constraints, we’re far better off averaging what we have." [Bloomberg]

Hard to poll primaries? - In addition to questionable data collection, a second set of issues involves those facing all polling for primary elections: Turnout is both low, relative to general elections and hard to predict. Voters choices are not constrained by partisan affiliation as they are in a general election, and primary elections are often prone to late last minute trends (something we know from when when polls track the election through the final weekend). As Nate Silver found, the accuracy of presidential primary polling was "pretty awful" in 2012 when conducted a week or more before the election: "Polls conducted just three days before the primary have missed by an average of about 7 points, and those conducted a week out have missed by about 10." That finding promoted Mark Mellman to remind us, a month ago, at the "clear lesson" of primary polling is "to never assume a lead will hold, especially in primaries." [NYTimes/538, The Hill]

Georgia polling showed a trend to Perdue - The final surveys in Georgia, conducted roughly a week before the election, gave Kingston healthy single-digit leads, but they also indicated a gradual trend to Perdue. Pollster's poll tracking model showed a roughly six-point gain in Perdue's support from late May through July 16 (from 36 to 41.7 percent). Republican pollster McLaughlin Associates, whose poll in late May had Kingston leading by a 14 percentage point margin, shared more internal data with Politico's Campaign Pro: "Their polls found the race closing in the final weeks. By June 30-July 1, Kingston's lead had shrunk to 6 points. On July 10, it was just 2 points. And in the final track, on July 13, Perdue was 1 point ahead, 42.7-42.3." [Pollster chart, Politico Pro]

Was low turnout a factor? - Harry Enten says no: "It would be easy to give the pollsters a pass and assign the error to low turnout. Overall, turnout was down 25 percent Tuesday from the first primary in May. But that probably wasn’t the cause of the pollsters’ error….On Tuesday, the drop-off in turnout was pretty uniform across counties. That is, it’s unlikely that pollsters overestimated turnout in strong Kingston areas and underestimated turnout in Perdue areas. It’s possible that there was an unexpected quirk in who turned out that doesn’t show up in county-by-county comparisons, but a change in turnout that would have caused this polling error should be at least somewhat evident in such an examination. For example, there wasn’t a significant connection between how Kingston did in a county and the turnout difference from Round 1. Nor was there any relationship between the change in Kingston’s margin over Perdue from Round 1 and the change in turnout. In fact, had turnout been the same as in May, Perdue still would have won." [538]

DO DEMOCRATS HAVE A SHOT IN GEORGIA? - Nate Cohn thinks so: "Georgia might well be moving down the same road as Virginia. No other plausibly competitive state — not Nevada or Virginia, not Colorado or North Carolina — has had a change in the racial composition of the electorate that’s as favorable for Democrats. That’s giving Georgia Democrats hope that they might win a race that they almost certainly would have viewed as a lost cause only a few years ago….Even under fairly optimistic assumptions, Ms. Nunn would need around 28 percent of the white vote. More realistically, Ms. Nunn will need to approach or even exceed 30 percent. The Democratic Senate incumbent Max Cleland broke 30 percent of the white vote in 2002, when he lost to Saxby Chambliss. But the Democratic Senate candidate Jim Martin won only 26 percent of the white vote in 2008, under extremely favorable national conditions….What’s telling, though, is that we’re even talking about this possibility in 2014, a year when nonwhite turnout should be lower and when the president’s party shouldn’t have much of a chance in a state with such a strong Republican advantage. " [NYT]

FLORIDA GOVERNOR'S RACE NARROWS - Rachel Lienesch: A Quinnipiac poll released Wednesday finds Charlie Crist (D) narrowly leading Gov. Rick Scott (R), 45 percent to 40 percent. Nine percent of respondents were undecided. The survey finds both Crist’s and Scott’s favorability ratings underwater, though Scott is slightly worse off. Forty percent of respondents had a favorable opinion of Scott, while 45 percent had an unfavorable opinion. Forty percent also said they had a favorable opinion of Crist, compared to 42 percent who had an unfavorable opinion. The results are in line with a SurveyUSA poll released Tuesday that found Crist leading Scott by a 6-point margin. Most polls throughout late May and June, however, showed Scott narrowly ahead, including a SurveyUSA poll taken three weeks ago. HuffPost Pollster's average, which includes all publicly available polling, shows a close race, with Crist at 45 percent and Scott at 43.4 percent. [Quinnipiac, Pollster chart]

SCOTT WALKER DEADLOCKED AGAINST CHALLENGER - Lienesch: "Wisconsin Gov. Scott Walker (R) is virtually tied with his Democratic opponent, Mary Burke, in the Wisconsin gubernatorial race, according to poll results released Wednesday by the Marquette Law School Poll. Among registered voters, Walker took 46 percent to Burke's 45 percent. Among voters likely to turn out in November, however, Burke took 47 percent to Walker's 46 percent....The poll also finds that Burke is still unknown to many voters. Forty-nine percent of respondents said they haven't heard enough about or do not have an opinion on Burke, while 26 percent have a favorable opinion of her and 24 percent have an unfavorable opinion. In comparison, 45 percent of respondents said they have a favorable opinion of Walker and 47 percent expressed an unfavorable opinion. Just 8 percent said they had not heard enough about or did not have an opinion on Walker." [HuffPost]

HUFFPOLLSTER VIA EMAIL! - You can receive this daily update every weekday via email! Just click here, enter your email address, and and click "sign up." That's all there is to it (and you can unsubscribe anytime).

WEDNESDAY'S 'OUTLIERS' - Links to the best of news at the intersection of polling, politics and political data:

-Just over half of Americans says the ACA has helped their family or others. [CNN]

-Most Americans think the MH17 crash is a threat to world peace, and most put the blame on Vladimir Putin. [YouGov]

-Jonathan Bernstein, Harry Enten and David Karol take issue with Chuck Schumer's call for a national top-two voting system. [Bloomberg, 538, WashPost]

-Domenico Montanaro takes a historical look at presidential approval in election years. [PBS]

-Alex Lundry (R) talks to Taegan Goddard about data mining in the midterm and presidential elections. [Political Wire]

-Jar Jar Binks is more popular than Congress, and Darth Vader outpolls the 2016 presidential field. [HuffPost, WashPost]

-Beer remains America's favorite alcoholic drink. [Gallup]

CORRECTION:An earlier version of this story mistakenly reported Jack Perdue's share of the Georgia vote at 50.1 percent. The actual percentage was 50.9