Hillary Clinton and Donald Trump in Las Vegas, Nevada during a 2016 presidential election debate.Chip Somodevilla / Getty
"Even a stopped clock tells the time twice a day."
This saying reflects well the illusion we usually fall into when evaluating a prediction: if we only look at the coincidence between what was predicted and the final result, and not at the way in which the forecast was arrived at, on the quality of the mechanism, We can end up looking at a stopped clock waiting for it to give us the correct time at any moment.
Exactly the same thing happens with the polls, starting with Trafalgar Group, one of the few pollsters that expect a Trump victory in 2020: its director, Republican strategist Robert Cahaly, expects according to his latest wave of polls that the president will keep a handful of Key states: Arizona, North Carolina, Florida and Michigan.
Since you were right in 2016, we are tempted to pay special attention to what you have to say.
But both its methods and the transparency with which it handles its operation raise serious doubts.
Trafalgar justifies its results by conducting very short polls, which (as defined on its own website) "last one to two minutes and are designed to quickly collect opinions from those who would not typically participate in political polls."
In addition, they pride themselves on innovating in methods to reduce the “social desirability bias” according to which (and this is Cahaly's central thesis) no one reveals their preferences clearly unless special techniques are used (which he does not define in detail) to be able to recognize them.
It is true that the response rate in surveys (not just political ones) has fallen dramatically in recent years.
It is also true that there is a fear of the insincerity of certain voters.
But even pollsters who have searched hard and intent for the hidden Trumpian voter have found that the margin of victory hardly changes.
Morning Consult, for example, conducted an experiment in this campaign: 2,642 respondents completed a questionnaire
online
(where desirability bias is usually attenuated by not interacting directly with an interviewer) or via telephone.
The results: although the bias was activated with questions about discrimination or racism in margins that ranged from 7 to 20 percentage points, in the intention to vote for Trump it barely changed by one point.
With regard to the times and the non-response rate, the duration of the questionnaire does not usually affect it as long as it does not exceed a certain threshold, closer to five, ten or fifteen minutes than two: most people who declines to participate, it does so from the second one, and it is not clear how a shorter duration can modify any partisan bias that exists in the lack of response.
Also, it appears that Cahaly has not been completely transparent about who is paying for his polls.
On October 30, Nate Silver, the benchmark demographic analyst in the US (and the only one of the great forecasters who gave Trump a decent chance of victory in 2016: 30%), explained on Twitter that they had discovered that Trafalgar received partisan funding without clearly communicating it.
Silver and his team have become the benchmark polls aggregator, distinguishing between those paid by a match.
When they detect that a company does not clearly indicate the client of any of its publications, they classify all of them as partisan.
For the rest, Trafalgar appears with a grade of C- (a kind of approved scraping) in the constant evaluation that FiveThirtyEight maintains of all relevant pollsters.
Its mean error is 5.6 points, with a pro-Republican mean bias of almost +1.
That is to say: nothing particularly prominent;
not too good, not terribly bad.
Nothing is stopping Donald Trump from winning again this year.
In fact, EL PAIS's forecast gives it a 15% chance of victory: in one out of every six or seven worlds, the president revalidates his presidency.
If this happens, the polling companies should seriously review their data and the methods with which they arrive at them: the eventual error of 2020 would make that of 2016 pale. The margin of failure would be double that of then in Michigan, for example , or quadrupling that of Pennsylvania.
But even if this happens, once the mechanism behind the time dial in Trafalgar has been gutted, nothing indicates that his work is a role model.
Subscribe here to the
newsletter
about the elections in the United States