The overwhelming defeat of pollsters

"A lot of polls are like black pudding: fine until you see how they do it," says a widely heard saying in political circles. The truth is that yesterday's election results showed that the forecasts had again failed, but this time in a catastrophic way. Of the 19 polls published shortly before the ban that banned its public release (August 3), 18 gave Alberto Fernández a winner, but the majority made a small difference to Mauricio Macri, who averaged 3 points when today's gap was finally 15 points. The three points are the so-called margin of error, with which most media Buenos Aires spoke a few days before the polling day of a virtual technical link, from a fight to vote to vote . All, but all these pollsters yesterday have played a real role.

There was even one, that of the Brazilian Ideaia Big Data, which measured for the Bloomberg agency (related to the financial world), which placed the President of the Nation at the top of the preferences of the voters with three points more than the candidate Front of all. And as if that were not enough, last Friday, in violation of the ban, some media of Federal Capital published the data of the latest study of the company Elypsis, the former undersecretary of Macrista Economic Programming, Luciano Cohan , who had informally disclosed Casa Rosada and spread among the operators. The city of Buenos Aires will try to raise the prices of Argentine shares before the Paso: Macri won 37.8% to 37.6%.

The difference between the predictions of this election and what happened yesterday was overwhelming. But there were three consultants whose forecasts were a little closer to the outcome of the elections. The latest study by Celag (the only one to have conducted face-to-face surveys) allowed Fernandez to win a 42.5% victory, compared with 33.5% Macri. And the consultancy firm Federico González & Asoc awarded Fernández 42% against 34.6% of Macri. Finally, the company Tres Punto Cero predicted a Fernandez victory of 43.6% against 36% of Macri. The numbers of the rest of the polls compared to the result of yesterday are really shameful.

When the results of the polls do not match what is finally happening, the consultants try to justify it by saying that there was a "hidden vote", a "shameful vote", a "useful vote" and / or a "clandestine or silent vote" could not be reflected in the studies. They intend to relieve themselves of their responsibilities by pointing out that "people define the vote on the same day of elections" and that "one thing is what people say and another is what they do ". And so, they are quickly looking to move from the role of having a rude glance at the next election and to say without blushing that their numbers are reliable and their forecasts infallible.

They are calm with their future work. They know that none of them will be touched by the fate of Javier Otaegui, this interviewer of the TV channel Mariano Grondona who, in 1993, was surprised to predict a success: the defeat of the libertarian and favorite Alberto Pierri before the radical Federico Storani in the province of Buenos Aires. Otaegui said his prediction came from a new polling method, using phone calls and not face-to-face interviews. Pierri ended up winning 22 points and Grondona fired Otaegui. This was the end of his sounder career. At that time, no size error was forgiven. Today, collective amnesia erases everything: the same consultants who have crudely failed in yesterday's forecasts are those who will be presented in the coming days with the results of the new investigations to which the generals of the 27 October will face. And no one will doubt them.

Why do election polls fail so much? Do people lie when asked? Do consultants draw their results at the request of those who hire them? Are they contaminated from the beginning by methodological errors?

The polls are trying to find out what the total number of voters will do by asking a party of them. To do this, they are looking for a representative sample. That in a small group, the whole society is reflected in terms of age, social clbad, gender, education and others. But what happens if the sample on or under-represents certain segments of society? Are the results in these cases reliable?

The debate over the uses and limitations of surveys is taking place around the world, after the studies had enough difficulty making predictions on several occasions. The consultants chain a chain of errors on a global scale. They announced the results of the referendum on the Brexit in the United Kingdom, the US presidential elections, the Colombian plebiscite for the peace agreement with the Farc and the Brazilian elections.

At the beginning of last July, the American channel CNN set new standards for the publication of the polls of this news network. Thus, it decided not to disseminate more surveys with IVR methodologies (interactive voice response, the machine that asks by phone), online (with an online response) or with samples of dubious representation. In Argentina, the vast majority of national election surveys use IVR for its low cost (out of the 19 studied, only Celag's was conducted face-to-face), but they face serious problems of representativeness. Only 50% of Argentinean households have fixed lines, a figure that falls to 26% of 20% of the poorest population. On the other hand, it is not common for young people living alone to have fixed lines at home, since they exclusively use smartphones. Thus, an important part of the universe that we are trying to represent would not be included in this probability sampling.

Celia Kleiman, CK's consultant manager, says it's best to ring, that is to say conduct face-to-face interviews. Be wary of telephone surveys and online surveys, as the more modest sectors are under-represented because they usually do not have landlines and mobile phones have prepaid systems (they do not have the ability to spend data to answer a survey).

Consultant Federico González described the fact that the IVR system may have a bias. The day after the Santa Fe Governor's elections on June 16, González's consultant called the people of that province via IVR and asked them who they voted for the day before. The percentage of votes for Cambiemos candidate, José Corral, was significantly higher in the IVR system than what had actually occurred in the previous day's elections, when he was third at 18%.

The polling stations would have three investigations: one for the candidate and his small table, another for the increased militancy and the third – the most optimistic – to spread in the press in order to influence the voters with their air of objectivity De Obviously, the media mainly published these latest polls, which eventually became a real fiction.