HAS every day or almost, its new survey. According to the latest Susquehanna study, Kamala Harris is 5 points ahead of Donald Trump. But, according to Rasmussen Reports, it is the Republican who is in the lead by 2 points. Between the two, many studies give Trump and Harris within the margin of error, both nationally and in swing states. The American presidential election on November 5 therefore promises to be daggers drawn. Provided that the pollsters do not fail like in 2016 and 2020
8 years ago, the institutes – and the journalists who blindly followed them – completely missed the rise of Donald Trump in the polls. Nationally, Hillary Clinton won the popular vote by 2 points, just one less than the polling average on the eve of the election. But the American presidential election does not take place by direct universal vote. To accurately predict the winner, you have to measure opinion in 50 different states – or at least in the half-dozen swing states where everything is at stake. It is on this point that the failure was bitter. So much so that statistics expert Sam Wang ate a locust live on CNN as penance.
Regional polls often catastrophic
The polls had, in fact, overestimated Hillary Clinton’s score by 3 points in Pennsylvania and Michigan, and by 6 points in Wisconsin and Ohio. In the autopsy of this fiasco, the pollsters explained that they had underestimated in their samples the share of low-educated white voters, who had mobilized en masse for Donald Trump, who ultimately won the election by 75,000 vote difference in three states.
In 2020, Joe Biden’s victory hides an even bigger failure. The American president won the popular vote by 4 points while the polls announced him at +7. The institutes missed out again in Wisconsin, but also in Florida, where they mismeasured Donald Trump’s breakthrough among the Latinos, especially men. Joe Biden was elected, but by only 45,000 votes in three states (Arizona, Georgia and Wisconsin).
ALSO READ These details on which the duel between Trump and Harris will be played outSince then, pollsters have tried to correct the situation, sometimes too much: in the 2022 midterms and in last year’s elections, the Democrats, buoyed by the battle for abortion rights, did better than expected.
And this year? We won’t know until November 5. As Nate Cohn explains in the New York Timescertain institutes, in order not to underestimate Donald Trump again, have decided to weight their samples according to the vote of those questioned in the previous election – a criterion which has long been considered unreliable. Worse, some pollsters adjust their results so that they do not deviate too much from the average of competing studies, which tends towards more uniformity. And increases the risk of collective failure.
A close duel among the punters
Some experts therefore decide to do without surveys. Historian Allan Lichtman, who successfully predicted nine of the last ten US elections (including Trump’s in 2016, although he missed Biden in 2020, underestimating, according to him, the impact of Covid), bases its system on thirteen keys which mix economic, political and personal data. He is categorical: Kamala Harris is the favorite.
Thomas Miller, data scientist at Northwestern University, prefers to analyze online betting. According to him, “prediction markets look into the future, and research shows they are better than polls at predicting election results.”
According to his model, we were heading towards a huge Trump wave against Biden, then towards a clear victory for Kamala Harris after the televised debate. But, in the final stretch, everything is tightening up. We are going through, he says, a period of “great uncertainty”. It remains to be seen whether we will be treated to “an October surprise” capable of tipping the scales.
Related News :