The team behind Economist 2020 election forecast They designed their model to consider many factors: surveys, economic conditions, presidential approval ratings, political polarization, the presence of a headline.
What was not factored?
“We are not trying to model the probability of a meteorite impact,” said Andrew Gelman, a professor of statistics and political science at Columbia University who helped design the model.
Of course, the probability of a meteorite hit radically disrupting the electoral landscape between now and November is quite slim. But again, not too long ago, few would have guessed that the campaign would be on the rise amid a global pandemic.
Former Vice President Joe Biden has a huge advantage over President Trump in polls nationwide and in key key states. But with memories of Trump’s surprise 2016 victory still fresh, professional pollsters, as well as average voters, spend a lot of time wondering if something unexpected might stumble over forecasts again this year. Many of the uncertainties involve the impact of the global health crisis.
Four years ago, a mixture of methodological errors and political circumstances led many analysts with trust predict a Hillary Clinton presidency, only for Trump to win in his place.
That mistake came after an electoral cycle that seemed unique at the time: a political stranger and former reality show host on the one hand, the first woman candidate of one important political party on the other. This time, the circumstances feel equally novel, albeit for very different reasons.
But Gelman points out that his model, which currently gives Biden a probability of winning better than 9 out of 10, is less concerned with what’s special about the current moment than with “the general range of specialty that may occur,” such as it is captured by historical information.
“People have said, ‘Well, 2020 is special,'” he explained. “But you go back to every election: 1948 was the first election without Franklin Roosevelt running in a long time, right? 1952 had Dwight Eisenhower, who was a singularly nonpartisan figure. 1960 had [John F.] Kennedy, who was the first Catholic to obtain an important ticket. You can go ahead; in almost every election, there has been something special and unusual. “
The pandemic, then, could be unusual, but historically speaking, it was unusual it is the normal.
Even if the forecasters did wants their models to explicitly explain the coronavirus, there doesn’t seem to be an empirical way to incorporate any additional errors introduced by the pandemic, said G. Elliott Morris, the Economist data journalist with whom Gelman worked.
“Your [would] You have to make assumptions about how much error you’re going to add, “Morris said.” Those assumptions are really difficult to make in a way that improves your model. “
But with months before Election Day, there is plenty of time for things to change.
Courtney Kennedy, director of survey research at the Pew Research Center, said she does not know what impact the coronavirus will have on the ability of the polls to predict the outcome of the election, but it is reasonable to expect some kind of effect.
“I think it’s one of the great reasons that people really can’t take [a poll], especially a national poll, in June or July right now, and suppose that translates to votes, “he said.
Beyond whatever uncertainty the pandemic adds, polls still have the difficult job of estimating who is actually going to vote. In hindsight, experts think many of the problems in 2016 involved state surveys overrepresent college graduates, who were disproportionately in Clinton’s favor.
Charles Franklin, director of the Marquette Law School Survey in Wisconsin, said the “vast majority” of pollsters now evaluate their results to make sure that the proportion of college-educated voters in their survey sample matches the proportion of the population. But, Kennedy said, not everyone does.
This time, Morris said, problems with poll accuracy could arise if the virus creates a systematic gap between people pollsters think are likely voters and those who actually turn out.
If one difference affected one party significantly more than the other, if Biden’s supporters did not vote disproportionately because they were more concerned about the virus, for example, surveys that did not take that into account would give inaccurate results.
Morris is skeptical of that happening.
“I think you’re starting to get into the land of conspiracy theory if you’re going to say that the coronavirus will kill 30 million more Democrats than Republicans, or make them more afraid to vote,” he explained.
Nathaniel Rakich, an electoral analyst for the leading prediction outlet FiveThirtyEight.com, said the virus could disrupt the electoral process in other ways.
“You have seen it in the primaries, there have been voters who have been unable to vote due to problems with absentee ballots, there have been long lines in some primaries,” he said.
Franklin said the economic impact of the pandemic could also complicate the surveys. A problem would be if a large number of people “lose their homes and having to move, because that may mean re-registering to vote, “which is not always easy, he said.
Ironically, the pandemic has also produced benefits for the surveys. Marquette is seeing more people agree to participate in the surveys, Franklin said, attributing that to more people who are at home when the phone rings. There are also indications that more men are responding than usual.
And as pandemic conditions do mail ballots Increasingly common, surveys can enjoy an increase in accuracy. As the election approaches, the fact that a higher number of people have already voted than normal could reduce the risk that voters who make late decisions will change the result.
In 2016, voters who decide late disproportionately broke right in key states.
This time, when “Trump is now a familiar commodity,” Kennedy said, there are good reasons to wonder if another pivot like this would occur. But early voters might be consistently different from election day voters.
“I don’t think we have good empirical evidence on how much [early voting] It matters, because we’ve never been in a situation like this, with such a big increase, “Franklin said. Pollsters will be watching” to see if those who have already voted are somehow different from those who have not yet decided, or if they still have to vote in person. “
A final question hanging over the survey research is whether the 2016 mistakes predicted a permanent shift toward less accurate surveys. Overall, however, pollsters remain cautiously optimistic.
In the 2018 midterms, the poll proved highly accurate, Kennedy said. “My big 2018 conclusion was that it showed … that fundamentally, the polls are not broken,” he said. Despite how things felt after 2016, “carefully conducted surveys can still work well.”
