If you are a Trump supporter and an avid follower of the polls you have reason to be discouraged. Since the Democratic National Convention ended last Thursday night the bounce Trump received at his Convention has vanished.
The most worrying poll for the party is a new Gallup survey that found only 36 percent of voters were more likely to back the GOP nominee after his Convention compared to 51 percent who would not. In comparison, after the Democratic National Convention, 45 percent said they are more likely to vote for Hillary Clinton, compared to 41 percent who say they are less likely to do so.
This is jarring to Republicans who thought Trump righted the ship after his Convention. Combined with the plethora of polls put since last week showing Clinton up by as little as 3 points (Economist) to 9 points (CNN/ORC), it has Republicans seriously rattled as they watch their candidate try to extricate himself from a battle he cannot win over a deceased Muslim soldier’s family attacking Trump and the candidate firing back.
Still, I am not worried. Right now it looks like doom and gloom for Trump but the news cycle changes quickly and in addition, let’s be honest, the polls have a shoddy track record of late.
In regards to the Gallup survey, it is not surprising Clinton’s message of inclusiveness and greatness appealed. But here is the rub. There is no evidence that the voters who said they were more likely to vote for Clinton after the Convention would. The main benefit Clinton can claim after the Convention is more voters say they are solidly behind her compared to Trump (not that it matters, a vote is a vote).
But, the polls have been pretty inaccurate for the last few years. All the way back in 2010, the polls gave Republican Senate candidates a combined +3 advantage. This made it seem like Republicans in Nevada and Colorado had a shot (they lost). In 2012, the same polls showed an incredibly tight national race between Obama and Romney. Obama won the popular vote by over 4 percent out of 125 million votes cast.
The polls were fairly accurate in 2013 with the New Jersey and Virginia gubernatorial races. But it has been primarily in the last few years the bottom has fallen out of polling not just here in the US but internationally.
In 2014, the vast majority of polls underestimated GOP support in statewide races. This was particularly true for Southern Senate races according to fivethirtyeight. Up to the day the election was held there were well-known detractors of this theory. But, in the end, the polls were wrong, very wrong. For example, the average Senate bias towards Democrats in the 9 seats Republicans gained was a whopping 5.84 points. Including every Senate contest it was 4 points. By comparison, when the polls were biased towards Republicans in 2010 it was by .9 points and 3.4 points in 2012.
Like I said, pollsters have not just struggled in the US. In March of last year, Israel held legislative elections. Not a single poll, several conducted by American based companies, accurately predicted the outcome. The right-wing Likud Party under Benjamin Netanhayu netted 30 seats and other right-wing parties (combined with Likud) netted over 60 seats to control Parliament (Knesset).
Hot off this polling debacle came the British Parliamentary elections. In the run-up to the election the majority of polls showed Labour slightly ahead or tied with the governing Conservative Party. Almost every single pundit anticipated that Parliament would be split (yet again). But, on election day, Conservatives outperformed every single election forecast by at least 40 seats. They over performed their anticipated vote share by 3 percent. Most significantly, and ominously for the polling industry, Conservatives outperformed the exit polls seat total by 14. Yikes!
You would think that such an experience might make pollsters change their habits. Not so much. In 2015, pollsters were off yet again in the Kentucky gubernatorial race. Polls showed the race neck and neck until the very end and then Republican candidate Matt Bevin vaulted to a 9-point win. The win was surprising because Kentucky is fairly blue at the state level and Republicans had nominated a controversial nominee (just like now).
Pollsters did recover somewhat in the 2015 Canadian federal election. Several caught the last-minute movement in favor of the Liberals and gave them a better than 50/50 shot of taking control of the House of Commons. They did.
All this brings us to the Presidential election. Pollsters were fairly accurate in the primaries. They caught Trump’s sharp rise and noted his strength in the later primaries. But they had some notable misses as well. They missed the late surge to Ted Cruz in Iowa. They also massively missed Bernie Sander’s win in Michigan (a whopping 20 point swing). But primaries are a different can of worms than general elections. Or so we are told.
Most pollsters predict their results will stand this November. They point to the fact that the vast majority of surveys have shown Clinton ahead. They also point to the fact they have taken numerous steps to rectify this from increasing the number of cell phones called to using hybrid surveys of online participants and phone respondents.
Yet, these steps might not be enough. They also cover over the fact that pollsters are anticipating a more traditional, more diverse electorate in a very nontraditional election cycle. The problem is not necessarily pollsters’ faults. Technology, voter angst and laws and regulations designed to limit robocalling (the No Call List) have limited their abilities to be accurate.
But pollsters bear some of the blame as well. They have a penchant to rely on old election results, use Census estimates and anticipate voter turnout rather than let their samples speak for themselves. In the most extreme cases, pollsters may even sit on their data (as in Virginia in 2014).
Further, pollsters tend to mimic other surveys results. Known as herding, this means the polls rarely diverge and a false narrative about a race can emerge (again, like we saw in 2014). To some degree we have seen this in Pennsylvania already with every poll but Quinnipiac showing Clinton ahead by 3-4 points.
Not trusting one’s sample can lead to pollsters manipulating their samples via weighting. While every pollster will weight their samples to match a particular anticipated electorate or demographic sample, some pollsters will do it to a significant degree. Considering how polarized America’s politics are along partisan, racial and ideological lines this can have a huge impact on a survey’s results.
Case in point returning to Pennsylvania for a second. The surveys with the largest Clinton leads, NBC and Suffolk, anticipate a more Democratic electorate than 2012 (yikes). PPP is somewhere in between 2012 and 2008 and Quinnipiac expects a very pro Republican (for a blue-state like Pennsylvania) turnout.
Much as polls have historically struggled to anticipate the turnout among minorities the same polls have struggled to account for the fact that millions of non-voters who fit the traditional view of a voter (low to middle-income, white, nuclear family) may turn out this November. As it was eloquently put by an author I recently read, it’s not when a group of people realize they have a newfound power that matters, it’s when they use it. If Trump supporters come out in force in November from places pollsters missed the results could be dramatically different than expected.
In a decisively abnormal election the past is certainly not prologue. But it probably means we should also consider throwing away the old assumptions and politics. It also means polls built on old methods, methods that have failed many times over the last 6 years, should be viewed as but one of many pieces of analysis about this election.