The surprise was total, that night of November 8 to 9, 2016, after Donald Trump’s victory in the US presidential election. Even on the eve of the election, most national polls put Hillary Clinton in the lead, sometimes nearly 4 points ahead.
Particularity of the American electoral system, even if Hillary Clinton collected nearly 3 million votes more than Donald Trump, the latter had won the majority of the large voters thanks to the rule of winner-take-all (“The winner takes all the money”) which applies in 48 of the 50 states. The leading candidate thus pockets all the votes, the other two including a dose of proportionality.
Four years later, pollsters have changed their practices and seem more confident in their estimates. According to the specialist site FiveThirtyEight, Joe Biden has more than 87% chance of winning against the Republican president. In 2016, this same site estimated, for example, at the end of the campaign that Hillary Clinton had a 71.4% chance of winning.
Mathieu Gallard, research director at Ipsos France, reviews the methods used in these opinion surveys and their reliability.
Page Contents
Can we trust national polls in a country where the vote is so scattered, considering that there are 50 elections?
Mathieu Gallard: The American voting system implies that these national polls must be taken with caution. We know that the disproportion between the popular vote and the electoral college exists, and that the latter is favorable to the Republicans. But with less than three weeks before the election, national polls give Joe Biden an average of 10 points. It’s a lot. It is almost impossible to be beaten in the Electoral College with such a lead in votes. It would be taken with caution if it was 4 or 5 points, as it was for Hillary Clinton in 2016.
This year, national polls have been very stable since the start of the year and give the Democratic candidate a victory. This really means that no event has a hold on public opinion, nor the coronavirus, nor conventions, nor the economic crisis, nor even the last debate.
In 2016, the debates had moved the lines, the reopening by the FBI of the investigation into Clinton’s emails too. At times she was 10 points ahead of Donald Trump, at others she was neck and neck with him. Joe Biden is in a much more favorable position because of this stability.
What mistakes could have been made in 2016 by pollsters that did not allow the victory of Donald Trump to come?
A criterion which had not been sufficiently taken into account and which is now proving to be important, is the level of education of the respondents. The polls had polled too many people with higher education degrees, which favored Clinton in surveys as those with less education voted more Trump, especially in deindustrialized states.
But that has been corrected today, the pollsters make sure to have representative samples from the point of view of the level of diplomas, an extremely important data to understand the behavior of individuals.
Another reason why the polls could have been wrong in 2016 is that in the Rust Belt (“rust belt”) there were 8% to 10% undecided on the eve of the election who ended up overwhelmingly voted for Trump, something pollsters did not anticipate.
Wisconsin and Pennsylvania, both located in this Rust Belt, were taken for granted by Hillary Clinton by many observers in 2016, but ultimately switched to the Republican side. Are we now monitoring more states than before, likely to create a surprise in addition to the famous “swing states” (“swing states”)?
Today we see that Biden is in the lead in these states, but with a much larger lead than Clinton had, and it is difficult to see how he could lose this lead in such a short time given this crystallization of the public opinion.
Overall, the swing states of 2016 remain closely watched, now there is also the South, those of the Sun Belt (“Belt of the sun”) who are going to be watched. Arizona, Georgia, Texas and North Carolina could switch for demographic reasons to Democrats. There is a whole white middle class population out there, formerly moderate Republicans, who might turn to Joe Biden.
One of the limits of the polls can still be explained by the undecided electorate, the very one who had tipped the election and foiled the polls in 2016. How do pollsters calculate this part of the unknown?
When we question a potential voter, we ask him if he is sure to go and vote, then if he thinks he can change his mind about his choice or if he is already sure. Some institutes choose not to take into account these undecided, while others will distribute them equally between the two candidates. There is no absolute rule, no agreement on one method more effective than the other.
It is impossible to know towards which side an undecided one tilts. In 2016, a large chunk of undecided voters voted for Donald Trump, but four years later, those undecided are not necessarily the same and could very well react quite differently. This year, around 5% of voters who think they are voting say they are still undecided. It is very weak. The vote seems very polarized.
