“President Trump was going over new polls — some internal, some not — showing him tied or leading Joe Biden in key swing states. ‘Pennsylvania tied. Florida, up one. Wisconsin, up one. Texas, up five. Arizona, Trump 49, Biden 45; North Carolina, Trump up three. And then Montana: Trump up a lot — 52-38,’ he said.”
Over and over, we’ve heard Trump wave away the idea that he’s in trouble in November, citing unspecified polls that show him doing well. And here some are — a couple without attribution and presumably internal, but a number with links helpfully included by Thiessen. The Wisconsin poll is from the Trafalgar Group; the Arizona and North Carolina ones from Gravis Marketing, commissioned by One America News; the Montana poll is from the University of Montana.
Actual polls, allowing us at last to evaluate whether Trump is right to feel confident about November.
He is not.
Let’s first walk through what the numbers say, in context. Here’s the margin Trump enjoys in the listed polls relative to former vice president Joe Biden, his likely Democratic opponent.
If you remember 2016, you’ll recall that Florida, Pennsylvania and Wisconsin were particularly close (and therefore particularly important) in determining the outcome. Trump leading or running close to Biden would suggest another tight race.
But it’s worth comparing the results Trump shared with the 2016 results. That result from Montana, where Trump boasted that he was up “a lot,” is a bit like LeBron James bragging about leading me by a wide margin in a game of one-on-one. Or, really, as if a starter for a Division II college were making the same boast: It’s not necessarily that he’s definitely going to win, but it certainly would be a surprise if he didn’t, and by a healthy amount.
After all, in 2016, Trump won Montana by more than 20 points. Meaning that the 52-38 result in Trump’s poll is actually significantly worse than he did four years ago. Specifically, it’s a swing of seven points, with Trump losing four points and his opponent gaining three. In Texas, a more subtle shift, with a five-point lead now comparing unfavorably with Trump’s nine-point win in 2016.
The rest of the results are similar to 2016, which is interesting in its own right. We’ll come back to this.
Savvy election watchers understand that individual polls provide a less-accurate picture than polling averages. Every poll has a margin of error; aggregating the polls into an average reduces that error. So how do Trump’s polls compare with the current averages in the identified states?
The average in Arizona has Biden doing eight points better than Trump’s poll — and leading. In Florida, Biden’s doing six points better in the average, also winning. There’s no average in Montana because it’s considered solidly red, but there is in North Carolina … where Biden’s doing seven points better in the average than in Trump’s poll. And, again, leading in the state.
In Pennsylvania, Biden is doing seven points better and winning. In Wisconsin, Biden is doing eight points better and winning. Only in Texas is Trump’s poll close to the average, which has Biden doing only two points better.
You’ll notice a pattern between the last two graphs. Excepting Texas and Montana in each, the differences between Trump’s polls and 2016 are fairly consistent across states, as are the differences between Trump’s polls and the polling averages.
Why? Perhaps because the polls Trump is citing are using a model of who’s likely to vote that closely mirrors the 2016 electorate.
When CNN released a poll showing Trump trailing Biden by double digits nationally, the president was furious, tweeting out a laughable memo attempting to undercut CNN’s results. (Those results have since been replicated numerous times by other pollsters.) Among the reasons that CNN’s results weren’t to be trusted, it asserted, was that they weighted the results according to the actual distribution of partisanship in the country and not to the 2016 electorate. In other words, CNN tried to emulate the electorate as it is and not as it was when Trump won, according to exit polls.
If the polls Trump is relying on now actually did weight their results to the 2016 electorate, that might help explain the similar results. The polls from Gravis indicate that they “are weighted by voting demographics,” which seems to support that idea. This also might explain why pollsters weighting according to party registration end up with results a consistent distance away from those results.
It’s worth noting that neither Trafalgar nor Gravis score particularly well in FiveThirtyEight’s rankings of political pollsters. Gravis gets a C; Trafalgar, a C-minus. Both pollsters were off by more than five points on average in their polling. This doesn’t mean their results are necessarily wrong. It does suggest, however, that one might be wary of accepting their results as more accurate than averages of a broader range of pollsters.
We’ve seen behavior like this before. In 2012, Mitt Romney’s presidential campaign was confident that it knew who would turn out to vote. Its internal polling showed Romney faring well in key states — because it was making incorrect assumptions about the electorate. The campaign was truly surprised when its polls turned out wrong.
The polling average that year was less generous. FiveThirtyEight’s analysis of polls gave Romney an 8 percent chance of winning on Election Day.
He did not win.
Correction: A previous version of this story mischaracterized what Romney’s internal polling showed.