There are some basic things the data can tell us, of course: Joe Biden is ahead in the most important states in the electoral college. Two polls show him up seven points in Pennsylvania, and the New York Times’ data analysis column, the Upshot, puts him ahead by five in Florida. National polls also show a stable Biden lead: Both the poll averages at FiveThirtyEight and RealClearPolitics give him an 8-point lead. Using this data, forecasts give Biden a roughly 80 to 90 percent chance of victory.
But the data that’s fueling these observations is, inescapably, incomplete. Given how fast the news is changing, and how many major stories are breaking at once, it’s impossible to capture how the public is responding to multiple overlapping events — and thus to say with any certainty how an individual development will affect the election.
As Times analyst Nate Cohn recently pointed out, modern polling can’t handle the current pace of events. Most high-quality survey houses spend days collecting information from their pool of respondents, aggregating it and preparing for publication. But news around President Trump’s health is changing by the minute. Reporters are still in the process of untangling the White House’s confusing timeline about Trump’s diagnosis and symptoms, and there are still questions about the state of Trump’s health. Public opinion may change as news continues to break, rendering even the best polls outmoded soon after publication.
And a confluence of major stories may make it difficult for voters to figure out what they think as they try to absorb an enormous amount of new information.
Think of it this way: The New York Times published its bombshell story about Trump’s taxes on Sept. 27, but other outlets kept publishing news and commentary on Trump’s taxes throughout the week. While voters were still reading those stories, the news cycle shifted to a raucous and unprecedented debate. And then, while Americans were still watching clips of Trump interrupting Biden and Biden telling him to shut up, Trump tested positive for the novel coronavirus and was hospitalized. Oh, and a new jobs report came out.
Even if I could pinpoint the exact moment when public opinion moved, it’d be tough to say which of these overlapping events caused the movement. There’s no way of knowing whether these events pushed and pulled the public in different directions at the same time. It’s possible that, say, the jobs report had no effect on the horse race — or that it would have caused a boost, but Trump’s covid-19 diagnosis pushed the numbers out of the news cycle.
That doesn’t mean polls are useless. They’re still our best tools for understanding public opinion, and they’re the most accurate predictors of election results.
But, at a time like this, we need to read them cautiously. If you see a new poll, check the field date — when pollsters interviewed voters — rather than the release date to see if it’s current. As always, compare individual polls to aggregates maintained by outlets such as FiveThirtyEight and RealClearPolitics, to see whether a given poll is an outlier. Treat surveys like a thermometer: They can take the public’s temperature, but that number is neither a perfect forecast of November’s temperature nor a complete explanation for the source of that the temperature.
Reading polls with this degree of diligence might seem like a lot of work. But most of these extra steps aren’t permanent necessities. Someday, the news will slow down, and we won’t have to worry about disaggregating the effects of a job report, presidential illness and a debate. And, when the election is over, we won’t need minute-by-minute updates on voters’ presidential preference. We’ll be able to, for a year or so at least, breathe easier — until the next horse race starts up again.