Polling Overload

There are increasingly more pitfalls complicating the work of pollsters. What if polling results mean nothing anymore?

Pollsters do not have an easy life. The results of their work are perhaps the most consumed product during an election season. It is a product often misunderstood not only by the general public who drink from the results, but also by the media who comment on it.

The pollster’s task, not easy to begin with, has become increasingly more difficult over time.

The Draconian fall of landline phones (along with nearly all voters being listed in the telephone Whitepages) means that now it is next to impossible for polls to produce a random and, thus, representative sampling.

The refusal of a growing number of voters to participate in such opinion polls, particularly among certain segments of the population who are more distrustful of government institutions, introduces an ongoing risk of nonresponse bias.

In the context of the Nov. 5 election, featuring Donald Trump as a candidate once again, there is a real likelihood that individuals being polled will hesitate to admit they support the Republican candidate. This fact has been apparent since 2016. Here, too, there is potential for nonresponse bias.

The same question always comes back to haunt American pollsters: Of those who agree to respond to a poll, who will vote on Election Day? It is not a mundane question. Pollsters have brainstormed and developed various methods to determine who will actually show up at the polls. Some rely on the voting history of the person polled (a history itself based on responses that may or may not be truthful), while others pose the question directly: “Do you plan to vote in November?”

With all this in mind, it is understandable that pollsters are having a tough time. And, for the third consecutive election, many of them are getting it terribly wrong.

It is impossible to know whether Kamala Harris or Trump will win the Nov. 5 election. The candidates don’t know, and neither do their campaign teams. As statistician Nate Silver pointed out earlier in the week, whoever claims that the polls are in their favor is selling you pixie dust.

Harris could sweep the seven states currently at play, which would give her the largest Electoral College victory since Barack Obama in 2008.

Trump could do the same, which would give him the biggest margin in the Electoral College of any Republican candidate in nearly 40 years.

And the two candidates could plausibly split those states in a way where Harris wins with 270 electoral votes to 268, which would be the closest margin since the 1876 election, the closest election in American history.

In short, uncertainly reigns over the outcome of the vote.

Sometimes polls can lead to aberrations. For example, if Harris really does hold the lead in Nevada by seven points (as Bloomberg reported), she cannot be behind by six in Arizona (as USA Today affirms). And vice versa.

And if Harris is leading Trump by just about three points in Minnesota (as Rasmussen maintains), a state not won by a Republican presidential candidate since 1972, she cannot be ahead by five points in Pennsylvania, the most hotly contested state in the country (as Bloomberg also reported).

This is not about simple deviations attributable to “margins of error.” It is about portraits of a world that does not exist.

Nor is this about a few polls that make up a handful of normal “aberrations” that occur when the number of samples is multiplied. It is about a persistent issue over several weeks.

In August, the same pollsters at the University of New Hampshire gave Harris a 5-point lead in New Hampshire, and an 18-point lead in Maine. As with the case of Arizona and Nevada above, here it is about two states that are not only neighbors geographically, but also politically. In the last two presidential elections, the results from these states were within 1 percentage point of each other.

It took an accumulation of senseless data like this for Nate Cohn, the excellent New York Times analyst, to ultimately write a piece shedding light on the fact that certain things were not quite right.

Among them, that the appetite of the American media for profit encourages polling firms (equally in a quest for profit) to churn out polls, with an emphasis on quantity over quality.

As revealed by a trio of political scientists 20 years ago in a classic article, nothing is more popular with consumers of election-related news than polling results. As the details of the legislative process can be boring at times, the “horse race” — a metaphor used by the authors of this article to illustrate polling results — sells.

Of course, polling firms must protect their reputations. But the fact remains that, given the quantity of polls thrown into the public arena, the collective memory tends to be short. Who remembers that pollsters from ABC gave Joe Biden a comfortable lead of 17 points in Wisconsin barely days before he won the state by a meager 0.6%?

Additionally, campaign teams have their own in-house pollsters. Their main goal is to provide the campaign, and the candidate, with the most extensive and reliable portrait of the true state of the race.

For weeks, sources close to Team Harris maintained that internal polling does not correspond to what is being disseminated in the media, who are, according to these sources, painting a rosier picture for Harris in states like Michigan. Time will tell who was right.

For the moment, two things remain.

First, polls, despite their errors and overabundance, continue to be the best tool for measuring public opinion. Second, with or without polls in the final five weeks of the race, it is impossible to predict who will win the White House on Nov. 5.