Should we trust the US presidential election polls?
An interview with Lorenzo Pregliasco of YouTrend, to understand where the "polls" of four years ago went so wrong, and what they say today
Next Tuesday 3 November is US Election Day. We are less than a week away so we can finally try to take stock of the situation. First of all, one thing that seems rather strange to us non-Americans: the United States has already been voting for some time. There have been voting 22 days before November 3, like Texas. Others where you have voted for 29 days before, such as California and Iowa, and still others where you have voted as early as 45 days before Election Day, such as Minnesota, Michigan and Virginia. In short, the US elections will take place shortly, it is true, but it would be better to say that they are already happening. At the moment, it is estimated that over 60 million eligible voters have already expressed their preference, both physically, thus going to the polling station in person, and by voting by post.
As the fateful date approaches, there is more and more talk of the United States and its elections, so it is better to arrive at this date prepared with numbers, statistics and possible scenarios. To do this, we interviewed Lorenzo Pregliasco, an expert in polls and political strategy, as well as co-founder of YouTrend and the Quorum agency.
After Clinton's surprise defeat to Trump in 2016, confidence in the polls apparently waned: yet Hillary Clinton got 2 million more votes than Trump (so the polls weren't too wrong). Today, four years later, can we trust numbers and projections?
"Absolutely not, because there are some methodical reasons why the numbers could be more accurate, but overall any measurement of public opinion, just like four years ago, can have an error of one sign or another.
The interesting thing is that many observers almost take it for granted that if there is a poll error it will be favorable to Trump. In reality today we have no reason to believe it. If anything we can say that, as there was an error of a few points four years ago in favor of Trump, this year it is possible that there is an error in favor of Trump himself, but also that there are no substantial errors or that there is a mistake in Biden's favor. We have no reason to think that the mistake necessarily goes in the same direction as four years ago. What needs to be reiterated is that any measurement, any survey or predictive model, has a margin of uncertainty just like any human activity. In some fields even more than in this one, with the only difference that we are less aware of it ".
The turnout at the moment is very high. The so-called early voting (early voting) has record numbers, we are talking about over sixty million eligible voters who have already voted: is there one of the two candidates who can consider this data as favorable?
“In general, the question is not how many votes but who votes. And then, of course, how you vote. So knowing how many voted compared to 2016 does not in itself indicate a favorite among the candidates. However, there are two elements that we can consider: the first is that those who vote today do so in a certain context, in an ecosystem where public opinion is very favorable to Biden. The national vote, as we know, does not matter too much in the United States, but having Biden with a nine or ten percentage point advantage means that whoever is voting now will not be influenced by any developments that may have taken place in recent days, perhaps in favor of Trump. . So the fact that the turnout figure is high, given Biden's current numbers nationally and in individual states, makes us say that it can be good for Biden, because it's a kind of stock of votes crystallized in a moment to him. favorable.
There is another aspect, which however would require very detailed analyzes, and that is that some states provide various types of data with respect to early voting. Some tell us the party registration, or the party with which that person has registered on the electoral roll, other states tell us the ethnicity, still others tell us the age of those who are voting early. In this sense there are indications of some interest. In Texas, for example, there is very high early voting participation in Austin County, which could be a good sign for Democrats. In Florida there is a different situation, looking at the "party registration" we note that among the voters there is a higher share of people registered as Republicans than in 2016. However, these are all so-called indirect analyzes: we use age or registering with a party to assume it could be pro-Biden or pro-Trump votes, but the truth is we can't know for sure. A voter could easily have registered as a Republican, maybe twenty years ago, and vote for the Democratic Party today. Or viceversa".
Speaking of Texas, on the fivethirtyeight website today the polls on voting in Texas essentially give Trump (at 47.5%) and Biden (at 47.6%) on a par for a chance to win. In March 2020, Trump had a 3.9 point lead. How can figures such as these affect the real vote?
“Well, they can impact as well as a thousand other factors. Let's say that polls can be part of the scenario that every voter knows when deciding how to vote and whether to vote. The same goes for other information, in the United States for example fundraising data can influence voting choice, because one candidate is known to have raised more or less money than another.
The polls, like any other fragment of the public debate, from what the newspapers write, to the contents proposed by the media or the conversations held on social media, are aspects that affect the voter's reasoning and therefore can have an effect. These are things we know: there is evidence that there may be a conditioning in favor of the candidate given the advantage, the famous bandwagon effect, but there is also the opposite risk, so showing a projection on a close election could mobilize voters of the candidate at a disadvantage. I think that in the end there is an effect, but that it is only one of the thousand ingredients that build the perception of the voter.
Specifically in the Texan percentages, it is true that an effect of this data in Texas can be there, but it can be in both directions: it is true that it can mobilize the Democrats, who have not won in Texas since 1976, but it is also true that it could mobilize the Republicans, perhaps those who did not vote four years ago ”.
Analyst Dave Wasserman puts together some data on the current situation of the race for the White House: "Biden is leading with 52% -43%, with a more stable base of votes than Clinton had in 2016". He also writes that "there are fewer undecided or third party voters", and that "district polls (which caused Clinton problems in 2016) this time confirm national polls." It would almost seem that Democrats can sleep peacefully ...
“Wasserman lines up very correct and interesting points. If we look at the curve of national polls in 2016 and compare it with that of 2020 we see that in the first case it is fluctuating, that is, support for Clinton and Trump rises and falls: there were at least three moments of rapprochement between the two (one after the convention, one in late September and then one in the last days of the election campaign). The 2020 curve, on the other hand, shows us that approval for the two candidates is stable: Biden remains at around 52% -53% and Trump is always at 42% -44%. This is a decisive factor: the 2020 campaign was a stable campaign for many months, at least as far as polls are concerned. Another thing that Wasserman rightly notes is that in 2016 Trump won also because there was a pool of minor candidates, like Gary Johnson, who eventually deflated and many of those votes went to Trump. This time there is a smaller pool of undecided, and the number of people who say they want to vote for a third candidate is also reduced. So it is more difficult for those at a disadvantage to come back. To answer your question: you can never sleep soundly in an election campaign, and I think the Democrats learned this four years ago. However, if we seriously compare the 2016 election and today's election, we can imagine that Trump has some chances of winning but they are not as high as four years ago. This, in the light of the data we have, is a fact.
Another thing that changes from four years ago is that today we have more polls on the decisive states, which are also quite stable - I am referring mainly to Pennsylvania, Michigan and Wisconsin. In these states there is an important methodological difference compared to the past: in 2016, polling institutions did not weigh the sample by level of education, because this had never been a decisive factor. But four years ago it was seen that it is, because white voters without degrees voted en masse for Trump. The lack of this specification weakened the polls of the time, which is why today many institutions include the level of education among the key factors, and today's samples are much more accurate. Despite this, in these Midwestern states, Biden's advantage ranges from five to seven points ”.
One of the most prominent analysts in American politics, Nate Silver, writes that “there are various theories as to why Trump could win that are most likely not true. But adding up the longshot possibilities, you still get a 10-15% chance, which is something to take very seriously ”. According to the numbers, where does Trump's chance to win come from?
“Surveys are a measurement, they tend to be accurate but have a certain margin of uncertainty. This margin can go in one direction or another and we cannot exclude that, with an error of even a few points, Trump could return to being close to being competitive. The third point is the American electoral system. If this system were national today we would have little doubt about Biden's victory, but since the American electoral system is based on the great voters that the candidate wins state by state, it would be enough that the polls are wrong in only three states and this could lead to a Trump victory.
Another factor of uncertainty comes from the fact that survey errors may be related. We saw it in 2016: if, for example, Trump's appeal among undergraduate whites is underestimated, this favors him, say the case, in Pennsylvania; but demographically, Wisconsin and Michigan are very similar to Pennsylvania and here it becomes probable that the same error affects at least those other two states, and perhaps Minnesota as well. Let's say that it is enough to make a mistake, even if not very much, in a specific segment of the American electorate for this mistake to have a ripple effect. Despite everything, therefore, it can be said that there are real, albeit not very high, possibilities of a Trump victory ”.
A YouTrend survey measured the perception of Italians on how much the US presidential elections affect our country. The result is very clear, there are peaks of 66.5% among those who think that the election of the US president is "very or quite" important. It is often said that the influence of the United States is in decline: in the international scenario, in your opinion, is the weight of the United States as strong as it once was?
"I would say that this is one of the few elections followed and studied at a global level, and indeed the poll shows that even in Italy there is a perception of the considerable importance of the American vote. In my opinion this is also linked to the fact that the figure of Trump is, for better or for worse, a figure quite extraordinary by the standards of contemporary politics. Trump inevitably increases interest in what is happening in US politics, even in Italy.
Beyond what the survey shows, the type of leadership and attitude of the United States on the international level still have significant repercussions, such as on the relationship between the US and Europe, or on the relationship with China and Russia. It is not surprising, therefore, that the United States remains, despite the frequent talk of decline, a country to be followed with particular attention, especially during a vote that potentially changes their president ”.
