One of the first topics to come up after an election is the failure of polls to capture voter intentions. Indicatively, many were asked to answer why in 2015, in the elections of Great Britain, while the polls gave a clear lead in favor of the Labor Party, in the end the Conservative Party won with a margin of 7 points, just like in France, after the first round of the presidential elections many wondered how the numbers fell out.
History repeated itself recently in the first round of the Brazilian presidential election. With all the votes counted, Lula had secured a 48.3% percentage to Bolsonaro's 43.3%, a much narrower margin than in pre-election gallops. Even Brazil's most authoritative polling firms, such as Ipec and DataFolha, indicated that Lula would avoid a runoff, surpassing 50% of the vote, void and excluding whites. The far-right Bolsonaro would remain at 36-37% of the vote, giving Lula a clear margin of victory. But the real numbers belied them.
Almost every week in our country there is a gallop on behalf of the channels or the major newspapers. A typical example of the abject failure of polls were those conducted during the referendum period, where the result was diametrically opposed to their numbers. So the question is, what is so wrong with gallops?
It may be a question that has been asked more and more recently, but it is not a new phenomenon. In the 1936 US presidential election between Republican Alfred Landon and Democrat Franklin Roosevelt, one of the largest sample polls was conducted. Literary Digest magazine conducted the poll with a sample size of 2.4 cm. giving a result of 57% in favor of Landon. In the end Roosevelt would be the winner with 62%. When the poll results were analyzed afterwards, it was found that the main reason for failure was the way the sample was selected. The magazine had selected the sample mainly from telephone directories, club subscribers and magazine subscription lists, social classes that were therefore prosperous and not supporters of the Democrats. Of course the questionnaires had been sent to 10 million and only 24% had responded. But beyond the rhetorical question of whether the result would have been different if everyone had responded, the lesson from this poll was that the sample should be drawn from different socio-economic strata. In the 1948 election again, although Dewey seemed to come out ahead with 50%, against Truman with 44%, the result was the exact opposite. This time, the lesson was that the sample should be random in every sense of the word.
It is clear that pollsters have had opportunities and time to demonstrate that they understood the value of these lessons. But in both the 2016 and 2020 US elections, the poll results were wrong, raising the question of whether some basic statistical rules are being followed in practice. In the case of the Brazilian election, there were polling exceptions that were closer to the actual results, especially for Lula's ratings, but no one predicted the far-right Bolsonaro's rating. According to Antony Wells, head of YouGov in political and social European surveys, when there are big differences, it's all about the sample. Typically, according to Wells, all samples are easily controlled for demographics such as age, gender and social class. But in all samples there is also some distortion. In the case of Brexit, pollsters have failed the education level demographic. AtlasIntel pollster Andrei Roman said in Bloomberg, that most samples represented the lower social strata supporting Lula. Also, there were many Bolsonaro supporters who, like Trump supporters, avoided answering the questions correctly so as not to reveal their electoral identity. As a simple conclusion, the gallops in this piece work like fake news.
And then there is the methodology with which the poll is conducted, an uninterpretable parameter for the uninitiated. Along with the methodology, the questions are sometimes framed in a way that guides the answerer. A typical question is who is considered best for prime minister, but two possible questions are given. But beyond the technical issues of each gallop, behind each "electoral thriller" are hidden numbers that may ultimately not reflect the real intentions of the electorate.
So why should the public care about polls and how much they influence the electorate?
First, although politicians say that opinion polls are not important to them, the exact opposite is true. Polls are done for them, as much as for the media that "order" them. But the results can create in the public what is called the bandwagon effect, the psychological phenomenon that influences some, regardless of their own beliefs and intentions, to vote for the candidate chosen by the many. It's a ridiculous reaction to Sirens-like polls.
Especially when the results are repeated through the press and social media, this phenomenon is even greater. But even if one does not follow the polls, studies show that the effect of contagion can work, where the opinion of others acts as a contagion virus to turn one's intention toward a candidate. A research by Neil Malhotra, a professor of political science, and David Rothschild of Microsoft Research, showed that in fact some voters do choose which candidate to vote for through polls, with the intention of belonging to the public that represents the winning side.
But in reality, the exact opposite should happen in a representative democracy. The political leadership must obtain information about public opinion and its views, while the electorate must remain unaffected by opinion polls.
But in the end, what role do opinion polls play?
For Jerome Jaffre, political scientist and head of the Cecop think tank, analyzing opinion polls before the French presidential elections, is an important democratic tool for analyzing situations, and its absence would "blind" the media. For Alain Garrigou, professor of political science at the University of Nanterre – Paris and author of the book l'ivresses des sondages (the drunkenness of polls), the polls they undermine public trust in politics and reinforce the view that they only work for a political elite.
But in the end, whatever the opinion of the experts, the power behind every poll is the electorate.