If that’s confirmed by additional higher-quality polling as primary season gets underway, Mr. Trump is no front-runner.
Why such a large gap? (Wonkiness rating: 6/10)
There are many reasons polls can disagree, but most of the usual explanations don’t add up to a huge 30-point gap:
-
It’s not about telephone versus online polling. Almost all of the polls have been conducted online, so the difference can’t be attributed to a so-called mode effect — like the possibility that Mr. Trump’s supporters won’t divulge their preference to a live interviewer.
-
The way pollsters define the Republican primary electorate (say: self-identified Republicans versus people who say they’ll vote in a Republican primary) doesn’t explain what’s happening, either. An analysis of New York Times/Siena polling last fall suggests that these choices do have effects, but that they are fairly modest on the scale of the 30-point gap in question. And multiple pollsters with similar definitions of the Republican primary electorate nonetheless show fundamentally different races.
-
The gap persists regardless of the number of Republican candidates listed by the pollster — including in head-to-head polling between Mr. DeSantis and Mr. Trump.
-
It’s probably not weighting, the statistical adjustments made by pollsters to ensure a representative sample. Most of the polls are weighted by roughly the same set of demographic characteristics, including by self-reported education.
-
It’s probably not the timing, but I’ll hedge a little bit on this one. Most of Mr. Trump’s worst polls were conducted in November and December, after the midterm election. That’s potentially relevant because the disappointing Republican showing in the midterms is the likeliest explanation for the apparent decline in Mr. Trump’s support. On the other hand, none of those pollsters have returned to the fray, and most of the polls that have conducted multiple surveys since the midterms have shown no change or no loss of ground for Mr. Trump in this period.
If it’s not the mode, the population, the timing, the question or the weighting, there’s really one explanation left: the sample itself. For some reason, some pollsters are getting a vastly more Trump-friendly group of Republican respondents than others.
Or, to be more blunt about it: Someone’s data could be extraordinarily and unacceptably inaccurate — inaccurate to a degree we would have never guessed until pollsters started asking about a new race.
So which pollsters are right?
It’s really hard to tell which of these polls might be “right” or “wrong.” There are countless ways to collect survey data online and, in general, there’s very little transparency about the process. Even when there is transparency, there aren’t well-established best practices that make it easy to evaluate whether a given approach is a sound one.
But there are two reasons to err toward the polls that are showing Trump weakness.
First, the so-called probability polls have uniformly showed relative weakness for Mr. Trump.
Probability sampling is where the respondents are more or less recruited at random, such as by calling random telephone numbers or by sending a mail invitation to random addresses to participate in an online poll. It’s traditionally considered the gold standard in survey research. A nonprobability sample, by contrast, isn’t selected at random. It might instead be recruited from banner ads on certain websites.
The five probability samples — from Ipsos, Suffolk, Monmouth, Quinnipiac and Marquette Law (fielded by SSRS) — tend to give Mr. Trump relatively bad news. Ipsos and Monmouth found him trailing Mr. DeSantis with just 25 percent and 26 percent of support in a multicandidate field. Suffolk University and Marquette Law/SSRS found Mr. Trump at just 36 percent and 33 percent in a one-on-one matchup (and did not ask a multicandidate question).
