Shy Trump voters are likely a myth, but polls aren’t perfect
It’s the latest poll from Iowa that has the race about even with perhaps the slightest edge for Trump. Trump won the state by 9 points in 2016.
What’s the point: This Iowa poll, like many others, shows that Biden is doing significantly better than Hillary Clinton in a given contest. As I have noted previously, Trump’s deficit has made folks wonder if the polls are undercounting Republican support. Previously, I’ve argued against such a phenomenon.
This poll reinforces that there is little sign of a vast supply of hidden Trump support in the electorate. That said, it is possible, especially at this early juncture, that Trump outperforms his current polling and goes on to win.
The great thing about Monmouth University polling is that it weights its results by party registration. Unlike party identification seen in most public polling, party registration is not what a person considers their political affiliation to be on the day she or he was polled and is not subject to a margin of error. Rather, it is a measurable fact produced by government agencies in a number of states.
Currently, the Iowa Secretary of State reports that 34% of voters in the Hawkeye State are registered Democrats and 34% of voters are registered Republicans. The aforementioned Monmouth poll matches that exactly.
Monmouth also matched the party registration figures pretty much perfectly in Pennsylvania when they polled there. Like in Iowa, they found Biden vastly outperforming Clinton’s baseline by more than 5 points in the state.
The same holds true if you look at other pollsters like the University of New Hampshire and the New York Times/Siena College polls. Both weight by party registration, and both have given Biden substantial advantages in key swing states.
Indeed, as The New York Times’ Nate Cohn has pointed out, Republicans were not any less likely to answer their polls than Democrats were once we control for demographics. (If those controls had not been in place, the polls would actually have too many Republicans relative to their registration percentages.)
View 2020 presidential election polling
A slightly hopeful sign for Trump backers looking for hidden Trump support is an examination of polls that don’t use live interviews. Most of these polls are done online or with an automated voice.
An accepted polling practice to see if voters are afraid to give a certain answer they deem to be socially undesirable is to compare the results when a live interviewer is present and when one isn’t.
Biden’s national lead is still 8 points in polls that don’t use live interviews at this point. That is a little lower than in polls that do use live interviewers and call cell phones, though that gap has only recently appeared and may just be a statistical artifact.
No matter how you measure it, Biden is ahead and by a not insignificant margin.
All of this said, it is still possible Trump still wins this election. For one thing, the election is still about 90 days away. We have to get through both conventions and the debates. In previous years, the difference between the polls at this point and the result have been wide enough to not feel overly confident that Biden’s advantage will hold.
Even if we were at the end of the campaign, Biden would need to be leading by a lot to feel very confident he’d win. The true 95% confidence interval for polling error is a lot more than just the margin of error. That’s even true when we average the polls.
Although most of the larger polling misses occurred before modern polling techniques were implemented after 1948, large errors can still occur. The average national poll still missed by 7 points in 1980. That’s actually much less than the 2 point miss of the national average in 2016.
On the state level, the errors have been even greater. Since 1972, the 95% confidence interval for state polling at the end of the campaign is about +/- 9 points for races within 10 points. While it’s quite unlikely all the polls from the key states would miss by that much, polling errors tend to be correlated with a given year. For example, most of the state polls in the Great Lakes underestimated Trump in 2016.
The good news for poll watchers is that big errors are rare. The average national error since 1968 has been less than 2 points. Since 1972, the average state poll aggregate has been off by just over 3 points in races where the polling margin was within 10 points.
In other words, polls don’t often miss, but when they do miss, they miss spectacularly. And to be clear, there’s no guarantee than any miss would underestimate Trump again. The direction of polling errors tend to be uncorrelated from year to year.
The bottom line is that Biden would very likely win if the current political atmosphere holds until November. Still, it wouldn’t be a guarantee.