Donald Trump and the war on polls
A campaign in which Donald Trump's favourite poll turns out to be wrong and he claims others are 'rigged'
One of the few remaining polls showing Donald Trump in a competitive race with Hillary Clinton turns out to have been overly influenced by one young African-American voter from Illinois who, unlike virtually everyone else who fits his demographic profile, is a committed Trump supporter.
It's another knock against the credibility of polls at a time when Donald Trump is talking about the election being "rigged" and questioning the legitimacy of many reliable election surveys.
- Pence says he and Trump would accept results of 'rigged election'
- Presidential Poll Tracker: The latest U.S. election projections
Trump has complained that a hostile media and widespread voter fraud, or a combination of both, could steal the election from him.
Voters appear to be listening. A recent poll showed 28 per cent of Americans said they would be unlikely to accept the election's results as legitimate if their candidate loses.
In this context, undermining the credibility of polling is particularly dangerous. One of the reasons for having public opinion polls during election campaigns is that they can help serve to validate a result — this is one of the reasons polls are banned in many dictatorships.
But it does not help matters when one poll proves to be of questionable value, and Trump had nothing to do with it.
Donald Trump's favourite poll
The troublesome poll that has been shown to give undue influence to a single African-American voter is the USC/Los Angeles Times daily tracking poll, which has stood out throughout this campaign for its much rosier estimations of Trump's chances. The poll has shown the Republican candidate ahead by wide margins at times when Hillary Clinton was seen as holding the lead in the vast majority of other polls.
It currently shows a tight race, while other polls are giving Clinton a comfortable edge.
It's one of Donald Trump's favourite polls to boast about at rallies. But as the New York Times' Nate Cohn reported last week, the poll has some serious methodological issues.
From the start, it was obvious that the USC/LAT poll was different. Rather than surveying a random sample of the population, as telephone polls do, it has surveyed people from the same pool of some 3,000 voters repeatedly.
For this reason, the USC/LAT poll has never been included in the CBC's Presidential Poll Tracker poll averages.
But the oddities did not stop there. The poll also weighted its sample to ensure it had the same amount of voters and non-voters as the general population in 2012, as well as the same proportion of people who cast a ballot for Barack Obama or Mitt Romney.
But poll respondents routinely over-report their past voting participation and are more likely to report having voted for the winner. This means that the sample probably has a disproportionate number of Republicans who have forgotten, or who won't admit, that they voted for Romney in 2012. The Times found that removing this weighting brought the poll more into line with other surveys.
When your vote really does matter
Which brings us to the disproportionate weighting of a single 19-year-old African-American participant.
Weights in polls are usually applied for factors like gender, age, and race. This is to ensure that every group is represented in a poll to the same proportions as in the general population. If, for example, a survey had only 25 per cent women, their responses would receive double the weight in order to get them to 50 per cent.
The USC/LAT poll took that weighting to the next level, giving individual respondents who fit specific demographic profiles enormous weights, making their responses worth many times those of other respondents.
The result was that a single African-American young male Trump supporter in Illinois was enough to move Trump's support by about a point, and swing his support among African-Americans by some 10 to 20 points, depending on whether this young man took part in the poll or not.
Encountering such an unrepresentative voter in a random poll would not normally be a problem. But the USC/LAT poll encountered him again and again because he was one member of their small panel who boasted a demographic profile that checked a lot of rare boxes.
Trump, by trumpeting this poll, was misleading his supporters. But he could not have known the methodological issues with it. He has no excuse for some of his other statements about polls.
Trump says polls showed he won the debates (they don't)
In the wake of the first and second presidential debates, legitimate polls surveying a random sample of debate-watchers showed, across the board, victories for Hillary Clinton over Donald Trump. Some showed wide margins, saying Clinton had performed better, while some showed narrower margins. But they all put her ahead.
If you only went by Trump's word, however, you would think that polls showed he had won by a landslide — with 80 or 90 per cent.
Veteran pollster Kellyanne Conway, Trump's campaign manager, certainly knows better. But that did not prevent her from tweeting out the results of these unrepresentative surveys herself (only to delete the tweets shortly afterwards).
It is part of the Trump campaign's assault on reliable information. The media report on legitimate polls showing Trump lost the debates. Trump points to his own numbers that show he won. His supporters know who they believe.
It's all rigged, and the pollsters are in on it
During a campaign rally in Florida last week, Trump pointed to women in the crowd holding up "Women for Trump" signs.
"I don't get these polls," he said. "Every place I go I see hundreds of 'Women for Trump' and then I see a poll, 'He's not doing well with women.'"
The Republican candidate shrugged in disbelief. It apparently had not occurred to him that the women at his rallies may not be representative of the wider, infinitely larger population of voters not attending his rallies on any given day.
It is a sad indication of the state of this election campaign that it is necessary to point out the ridiculousness of Trump's claim that most polls are deliberately rigged. For pollsters to be part of a conspiracy to prevent Trump from becoming president, this would require the thousands of people who work for pollsters to be colluding in putting out false results without a single one of these people breaking ranks.
It would also require the complicit lying of media outlets like Fox News and Breitbart (run by Trump's campaign CEO), that have commissioned polls that show Trump behind nationwide and in key swing states. It beggars basic reason.
Nevertheless, it is the increasingly common view of Donald Trump's most dedicated supporters.
And it suggests that if Hillary Clinton wins this election — as the polls indicate without a doubt that she is on track to do — the ugliness of this campaign will be far from over.