Donald never lets a mere fact stand in the way of good propaganda, but the rest of us should be skeptical of his opinion poll-based claims.

In the last few days, Donald has been touting polls that claim a significant increase in support among Hispanics. Today, he touted a 50% approval rating in the latest Marist/NPR/PBS poll, and if you review the poll results, it does indeed say that. But that’s not the end of the story. Not by a long shot.

Marist/NPR/PBS Poll shows President Trump’s approval rating among Latinos going to 50%, an increase in one year of 19%. Thank you, working hard!

— Donald J. Trump (@realDonaldTrump) January 22, 2019

First, according the methodology, the poll was based on 1,023 mobile and land line phone interviews, of which 15% of respondents were Hispanics. That’s only 153-154 people. When I put that into Raosoft’s opinion poll sample size calculator for a 95% confidence level and a population of 58 million people, it calculated that the margin of error was 7.9%. So for a theoretically perfect opinion poll, that 50% approval rating could be as low as 42.1%, or as high as 57.9%.

But we know that the poll’s not perfect. In fact, we know it’s imperfect in ways that likely inflated Donald’s approval rating among Hispanics.

First, the percentage of Hispanics in the US was 17.6% in 2016. That means the 15% of respondents is a slight undercount, and that results in some small error. It’s possible that this error was calibrated out using comparison to other, more complete datasets, but it’s not explicitly stated in the methodology statement.

Second, the methodology says that the interviews were conducted only on personal, non-business phones. While lots of Hispanics have phones, many of those phones serve as both personal and business phones. Since we don’t know exactly how the business phones were screened out, we can’t say how much of an impact this would have had. But it very likely biased the sample somewhat, adding some indeterminate amount to the error.

Looking at the demographic profile of the respondents, the respondents tended to be both older and wealthier. As of 2017, the median age of Hispanics in the US was only 29 and 73.2% were younger than 45. That compares to the fact that the respondents were 55% 45 and older. Without information on the precise age demographics of Hispanic respondents we can’t estimate the error, but clearly this could be a strong source of bias in the results.

The median income for Hispanics is just shy of the $50k breakover point that the poll uses. But 57% of respondents had an income over $50k, indicating that this could be yet another source of bias in the poll.

Educational attainment is yet another source of possible bias. The poll skewed 55/45 toward people with no college degree, but 84% of Hispanics in 2017 didn’t have a college degree.

Finally, the poll exclusively interviewed in English, yet in 2016 about 32% of Hispanics didn’t speak English. This alone is a massive source of bias in the poll results.

We can’t know how much additional error all these biases introduced into the results for Hispanics, but it’s likely to be pretty significant. So instead of relying on polls with small sample sizes like the Marist/NPR/PBS poll, we should instead look at polls specifically of Hispanics. And the latest large poll of Hispanics that I could find (1500 Hispanic respondents, conducted in English and Spanish, etc.) found that only 22% approved of the job Donald is doing.

With polls, the devil is always in the details. And with Donald, he never lets a fact get in the way of a perfectly good propaganda opportunity.