At a campaign rally in Tampa on Wednesday, Donald Trump returned to one of his favorite topics of conversation, polling. He hasn’t been talking about his good poll numbers much recently because his poll numbers haven’t been very good. But he singled out a couple of points of data during his Tampa speech that he figured were worth sharing.

“I don’t know if you’ve seen what's happening, but over the last three weeks, the polls with African American folks and Spanish-speaking folks, the Hispanics, Latinos, have gone way up. Way up. They’ve gone way up,” Trump said, to applause. It seemed like it was validation of the time he'd spent over the last week making somewhat awkward appeals to voters in those demographics.

But it’s not, because those numbers are wrong.

The problem almost certainly began with a headline at Trump’s favorite website, Breitbart.com. On the heels of Trump’s speech near Milwaukee last Tuesday, the site ran a story with a headline that would seem to make Trump's case on black voters.


“The Los Angeles Times/USC Dornsife national tracking poll of the presidential campaign picked up a sharp uptick in African-American support for Republican nominee Donald J. Trump, which it now pegs at 14.6 percent,” the story began. “When the ‘Daybreak’ poll began July 10, Trump’s support among black voters was 4.6 percent and he fluctuated between 5.4 percent Aug. 1 and 2.5 percent Aug. 11.”

We wrote about the Times/USC poll on Monday, for the simple reason that its results were consistently more favorable to Trump than most polls. It’s the blue line below, while the average of a number of polls compiled by RealClearPolitics is the yellow one. The Times/USC poll has always been at least 2 points friendlier to Trump -- but it's usually much more favorable than that.


The fair assumption is that Trump saw the Breitbart headline, or at least was aware of the point it made, and looped that into an argument that his numbers have “gone way up.” That’s only black voters, sure, but it's still a big upswing, tied to when he started talking to black voters. That the Times/USC poll is a weekly average of poll results, and therefore didn’t suddenly jump immediately after the Wisconsin speech because of the Wisconsin speech is the sort of detail that those egghead pundits can worry about.

Unfortunately for Trump, though, his timing couldn’t have been worse. That upward spike in the Times/USC poll vanished entirely in the most recent iteration of the survey.


You can see the brief bubble on the chart above. It’s not an uptick “over the last three weeks,” but it was there, briefly. Those gray bars at the bottom also show that the bubble coincides with a period when the margin of error for black respondents had increased, probably due to a smaller pool of black respondents being included in the survey.

[Update: Jill Darling, survey director for the Times/USC poll, explained the change over email:

In any survey, estimates among subgroups are more sensitive to shifts based on relatively small changes in the population’s average opinion. So in our case, strongly held Trump support among even a very small number of participants can widen the margin of error and affect the estimate, as we’ve seen over the past few days. Therefore, we urge caution in over-interpreting these shifts or any others in our – or any poll's - sample subgroups as a reflection of change among the group overall.

In other words, there wasn't a big shift -- there was a small one, in the right place.]

The Times/USC poll also doesn’t show any significant increase in support for Trump among Hispanic voters. He’s up a bit over the past 10 days -- but that’s only after he sank a bit for a week or so before that.


Nor do other tracking polls. The NBC/SurveyMonkey tracking poll released Tuesday showed Trump at 8 percent with black voters and 22 percent with Hispanics. On Aug. 5, those numbers were 8 percent and 24 percent, respectively. In the Reuters/Ipsos poll, Trump went from 5.6 percent with black voters on Aug. 1 to 2.5 percent on Monday. (There weren’t enough Hispanic respondents to separate out.)

Other than tracking polls, there aren’t any major national polls that have been conducted both right after the conventions ended and more recently, meaning that this is all we’ve got in terms of figuring out what Trump’s argument might be. It’s why that Breitbart story — which got a lot of pickup — is the most likely culprit.

There has been no updated story at Breitbart with the new, worse Times/USC numbers on black voters, by the way. But it still provides an excellent lesson in the dangers of picking out one bit of data from one poll. You shouldn’t do that! Or, at least, you shouldn’t do that without mentioning the dangers of doing that. And you certainly shouldn’t do that and then assume it holds for other demographic groups and assume it represents a real change and assume that it took place over a longer period of time and then figure that it is worth mentioning in a major televised speech.

But maybe that’s just me.