— There is a problem of capturing extreme right-wing voters in polls, mainly through polls (polls), which is not a small challenge – says Professor Riccardo Seneviva of IESP/UERJ. – Social science has always had problems reaching specific populations, and this is happening now with the Bolsonaristas and elsewhere.
The two major milestones of these errors are the 2016 and 2020 US presidential elections, when the firms’ margin of error differed from poll results by an average of three to five points. In the first election, the result was a shock, as expectation models predicted Hillary Clinton’s chance of victory to be between 71% and 99%. Second, Joe Biden’s lead was large enough to secure Trump’s victory even though Trump received more votes than expected.
The U.S. will hold elections again in early November, this time for the legislature, and polls favor the president’s allies. This again raises doubts as it goes against the historical trend: almost invariably, opposition parties perform well in the legislature. This may lead companies to calibrate their model, which can produce effects in the opposite direction:
– It is possible that (US) institutions that over-represent right-wing voters want to compensate and lose their hand. That is, the error can be reversed,” said Clark McCauley, professor emeritus of Bryn Mawr University in Pennsylvania.
In the UK, the phenomenon of “shy conservative” voters has long existed – as The Economist put it, “It’s never fancy to vote Conservative, so people keep quiet about it.” It produced some historic surprises, such as the 1992 and 2015 general elections, when the Conservatives won unexpected majorities. However, the discrepancy peaked in the last general election in 2019 when opinion polls predicted a margin of between 29 and 52 seats, far short of the 80-seat margin given by opinion polls.
In Chile, last month’s referendum on a constitutional referendum — which was won by Augusto Pinochet, a right-wing incumbent, to retain the charter — stalled by 16 percentage points. Marta Lagos of the Latinobarómetro Institute says there are “systematic errors” in Latin American research because “in general the statistics are not good, they are averages”.
To offset this, he says, more investment in consultancy is needed to scale up models. However, given its institutional difference, an internal audit is needed to understand what happened:
“But they told me I was crazy,” Lagos said. – Research in the region was most accurate in times of stability, and inaccurate in times of change. We should be careful enough.
