What is Wrong with the Polls?

Both camps have now complained about the polls. For months Donald Trump has suggested that the polls have been rigged and last week the Clinton camp suggested the apparent collapse in support was because of “bad polling” (Louis Nelson, Politico, November 1st). Hillary Clinton had been proclaimed the winner well ahead of election day. In mid October the Washington Post highlighted that CIinton was going after firm Red States such was her lead (Washington Post, Oct 16). Bookmaker Paddy Power paid out on Clinton bets around the same time and the Clinton camp jubilantly gave details of the ‘giant’ victory party planned for election night, complete with fireworks, under a glass ceiling (Washington Post, Oct 31). The zenith of this presumptive Democratic landslide was an ABC WaPo poll showing Clinton 12 points ahead with little over two weeks remaining. On the back of the poll, CNN reported that “Hillary Clinton is steadily moving her focus beyond Donald Trump, increasingly planning for what she believes will be her transition to the presidency” (Jeff Zeleny CNN Oct 24). Clinton herself revelled that “I debated him for four and a half hours…I don’t even think about responding to him anymore.” (BBC News US Election 23 Oct).

Just a few days later on November 1st Clinton tweeted 9 times in an hour, personally attacking Trump, “most of us learned by elementary school that its not OK to insult people’s looks. Donald calls women fat, ugly and disgusting”, and began to raise the volume of hysteria about Trump being a Russian plant, “It’s time for Trump to answer serious questions about his ties to Russia.” All composed, triumphant focus on policy and ‘going high’ vanished as the Democratic campaign came to resemble that of Trump’s. This of course was provoked by the FBI announcement that it was reopening the case into Hillary Clinton’s private email server (Business Insider, Natasha Bertrand, 28 Oct.). That precipitated a doom laden news cycle for the Democrats culminating in the same ABC WaPo poll showing Trump now one point ahead, just a week after seeming irreparably behind. The polls have remained close ever since.

The immediately obvious answer for such a turnaround was the FBI announcement, but did it really have such an effect? By the time of the announcement 20% of all votes had already been cast and the undecided portion was hovering around a paltry 5%, and the release itself was woolly at best. It wasn’t really the bomb shell Republicans were hoping was coming and were crowing that it was. It was just a public acknowledgement that her server was again being looked at. In terms of media hoping for an ‘October surprise’ so close to an election this was a big talking point and for the Republicans, seemingly out of the running just a day before, it was a ‘yuge’ revelation. But for the public in earnest? It didn’t say she had even done anything wrong – surely it was not enough in itself to produce such a swing in the polls, after all it was, as was pointed out, so close to the election. Opinions had been formed and solidified over the previous months, heels dug in in a bitterly divided contest. So why has there been such movement in the polls? It can’t have been a ‘bad poll’ as the race has stayed tight. It leaves one asking, have the polls been rigged, or has there been something fundamentally wrong with them?

The answer to the first question is yes and no. The Wikileaks release of the Podesta emails exposed that instructions were given to oversample certain demographics by the Clinton campaign. In Arizona it was urged to oversample Hispanics and Native Americans. In Florida older voters were cut down on while African American and Hispanic voters were singled out for oversampling. The tinkering even applied to independents of which the pool was directed to be drawn from Tampa and Orlando rather than north or south Florida in order to heighten the chances of an independent Democrat lean. Meanwhile, the same set of emails exposed that national polls routinely oversample “key districts/regions” and “ethnic groups” “as needed.” A further NBC poll in early October which gave Clinton an 11-point lead was conducted by a pro-Democrat SuperPac. The Podesta emails have also explicitly exposed collusion between the Clinton campaign and CNN; going as far as giving the Clinton camp questions in order to prepare answers before the first debate (Amber Jamieson, The Guardian, 31st Oct).

However, this does not suggest, as the more extreme supporters of the Tump campaign might, that there is an enormous hidden hand pulling the strings of all polls, attached to a Hillary Clinton shaped wrist. But it does suggest that polls should have been taken with a sizeable pinch of salt. Quite the opposite happened in reality and polls giving an ‘unbelievably’ large lead for Clinton were excitedly promoted by almost all of the media continually. However, even if some polls were influenced to a greater or lesser extent by the Clinton campaign, surely not all of them were. Yet the phenomena of an ostensibly insurmountable Clinton lead until the past few days had been depicted in almost all polls, even at a state level. The Real Clear Politics tracker in what ought to be solid blue Pennsylvania had Clinton and average 10 points ahead in early October, now she is only 2.6 point up. Polls usually tighten as we get closer to election day but such violent swings are unusual. One might still maintain that the FBI statement did manage to produce such a turn but if we take a look at the polls themselves then a different explanation emerges.

When one looks at the methodology of the polls a number of problems can be seen. It seems that it was routine to oversample democrats, usually by about 9% but sometimes as much as 12% or 15%. A CNN poll in Arizona in mid October gave Clinton a 5-point lead. But the poll of 713 people polled 413 Democrats, 168 Republicans and 132 Independents. A Reuters poll a few days before gave Clinton a 6-point lead, this time with a Democrat +12, women +22 sample. The retort is that polls have to be weighted to correspond with reality; more women and Democrats tend to vote - at least in the prior two elections – if a poll was random then it would not account for this reality. This is true, women and Democrats have tended to turn out in higher numbers on election day but not by such margins. In 2012 for example, more Democrats and Women voted but only by 5% and 6% respectively (ropercenter.cornell.edu).

The problems with general poll weighting multiply the more we look at the demographics which constitute the polls, chiefly because of the use of ‘likely voters’ and past trends to plot this year’s projections. Broadly, a “likely voter’ is aggregated based on their voting patters; did they vote in the last presidential election, did they vote in 2008, did they vote in the midterm elections, how often have they voted before etc. Those who have never voted or stopped many elections ago are deemed very unlikely and do not make up the samples. In addition, demographic trends and turnout primarily from the last two elections are used to project who will be president. The problem is that this is not 2012 or 2008. The Democrats had a uniquely popular candidate in 2008 who came with believable messages of hope and positive change, bringing about a landslide win of more than 10 million votes. He ran again in 2012 (though Democrat turnout dropped by almost 4 million). This time it is not Barak Obama running and if anything Hillary Clinton could be called uniquely unpopular when compared to ’08 Obama. This year, though both score highly for unfavorability, it is the Republicans who have the popular message, particularly for their target audience. Thus far this has been born out in the numbers. African American turnout has been down this year in comparison with the last two elections. In North Carolina it is down 16%, white turnout is up 15%, African American turnout is down more than 10% in Florida and this trend is born out across the country (New York Times, Nov 1). Some are arguing that this might be more than counterbalanced by increased Latino turnout, especially in battleground states such as Florida and North Carolina (Daniela Diaz, CNN, 5th November).

However, though Latino turnout is up in key swing states, the rise in white voter turnout has outstripped it (Fraga and Schaffner, Washington Post, 4th November). Latino voting has increased particularly in Arizona, Florida and Texas compared to 2008 and 2012 but the increase in white vote has outstripped it in all battleground states except Florida. Thus far there is only marginal evidence that Latino voting is outsizing the Latino electorate, it is only in Florida where Latino early voting is outpacing both the population vote and early vote for non Latinos. Yet, even in Florida the growth in participation thus far still “relatively small. ” The Washington Post goes on to highlight that in Texas and Nevada the Latino voters are underperforming relative to population vote and the steady/increasing early vote amongst white voters. Perhaps more importantly it is pointed out that so far white voters constitute a larger share of the vote than in the prior two elections.

Further issues arise in the projected unity of each party. This was displayed most glaringly in the `Democratic primary in Michigan. Clinton was projected to win by an average of 21 points (Real Clear Politics) and the much touted fivethirtyeight.com put her chance of winning at 99%. Sanders won the primary. After such an ideologically divisive battle and the Wikileaks revelations which plagued the Democratic National Committee, it is anyone’s guess where many of the Sanders block will cast their vote. The Republicans though divided at the top have not had anything like such a divide in their core support. This provides a further issue in polling predictions. Super Tuesday in 2008 saw 8,228,763 Democrats cast ballots to only 5,025,685 Republicans. This year there was almost a complete reversal with 5,557,243 Democrats votes to 8,307,884 Republicans. It is easy to retort that this was just one day, during primaries. However, the trend was repeated across all primary voting. In 2008 Democratic primary voting totalled 37,415,320 to 20,828,435 for the Republicans. This year Democratic turnout fell to 30,895,719 while the Republicans rose to 29,646,417.

Significantly, this trend is has played out across key battleground states where Republican turnout was up 74% this year compared to just four years ago. When compared with 2008 Republican primary turnout this year was up 122% in North Carolina, Democratic turnout was down 27%. In Ohio Republican turnout was up 87% on 2008, Democratic turnout down 47%. In Pennsylvania Republicans cast 91% more ballots this year than they did in 2008, Democrats 28% fewer (uselectionatlas.org). The same trend is born out in every battleground state. The trend from the primaries has been carried to the early voting. Republican turnout is up on 2012 in North Carolina, Iowa, Florida and Ohio by more than Democrat turnout is (Ryan Struyk, ABC News, November 3rd).

Nevertheless, one must remember this does not necessarily suggest the election is going to go one way or another. As has been mentioned this presidential election has been like no other and with such discord between and within the two parties one could argue registered Republicans might vote Democrat and vice-versa. Likewise, the boom in white voter turnout mentioned above does not mean that that is a growth in support of the Republicans, the rise could be to counter the fear of a popular Donald Trump. However, this article is not meant to predict a winner. It instead highlights that because of inherent problems with methodologies (weighting and projection based on past trends) the polls this year are almost uniformly unreliable.

Simply, Donald Trump has made this election a black swan event, he is not a normal candidate and this is not a normal election. Prior trends must therefore be discounted or at least only used as a rough guide. This has been apparent to many for months, but even though they have consistently acknowledged it, pollsters have consistently refused to account for it in their methodologies and projections leaving one to wonder whether they have been wilfully ignorant or deceptive?

A final point worth remembering, something seemingly overlooked by all polls, is changes to the largest voting block – white voters. This has occurred because this year especially such changes are far more difficult to predict. Contrary to popular belief, the assertion that “there are not enough white voters in America for Donald Trump to win while getting routed amongst minorities” as MSNBC Joe Scarborough stated is not true. Alienating minority voters is politically unwise but it is still leaves a road to victory, a small road but a road. There are more white voters than has been commonly thought. The electorate is in fact older and whiter than exit polls in 2008 and 2012 had pollsters and pundits believe. The evidence from this comes from the Current Population Survey (C.P.S) and Catalist, a Democratic data firm. Catalist suggests that in 2012 the white voter bracket over 45 years old was almost 10% larger than exit polls suggested, while C.P.S suggests that the white bracket of all ages with no bachelor’s degree was 10% larger than exit polls suggested. The overall implications of Catalist are that there could be at least 10 million more white voters over 45 without a degree than was implied by the exit polls at the last election (Nate Cohen, The New York Times, June 9th). Had they all voted Republican in 2008 John McCain would have won, even against uniquely popular Obama.

Moreover, the larger numbers of white working class suggest that the Democrats are more dependent on winning white working class voters than was thought, a forgotten point. 34% of Obama’s voters were white and without a degree compared to 25% in the exit polls. This forgotten group might provide a polling upset if their loyalties have changed in the face of new populism. But it is those who did not vote who provide the biggest threat to the polls and the Democrats, more so than white working class who might be flipped. The potential of a ‘monster vote’ is impossible to account for under current polling methodology. Those who have never voted or stopped decades ago, alienated and disillusioned can still vote. Just because they are not polled does not mean they do not exist. It is this demographic that was largely credited with the surprise of the recent Brexit vote and the exceeding of expectations in German elections by the AfD. It could be this undetected group, proved to be the biggest weakness in recent polling, that help Donald Trump create the black swan.