Philip Klinkner, writing for Vox, cites the 2016 American National Election Study (ANES) pilot survey as the source of this number.
But, the stated purpose of the study is to test questions for future studies, not to draw general conclusions about the population:
The ANES 2016 Pilot Study was conducted for the purpose of testing new questions and conducting methodological research to inform the design of future ANES studies.
Also, the conclusion Klinkner draws is based on a very small sample: only 328 respondents identified themselves as "White" and said "Muslim" in response to the question "Is Barack Obama a Muslim?". (See below for more on this number.)
The entire data set includes 1,200 samples, collected in an online questionnaire:
Data collection was conducted between January 22 and January 28, 2016. The sample consisted of 1,200 individuals who were part of an opt-in internet panel and who completed the 32 minute (median) questionnaire online.
You can register and then download the data yourself to verify the math that Klinkner did. You may refer to the User Guide for details of how answers to survey questions are coded.
First, you can see that 875 respondents identified their race as "White". These are coded with a 1 for the
race variable in the data set.
One of the questions on the survey (the
bo_muslim variable in the dataset) was:
Is Barack Obama a Muslim, or is he not a Muslim?
and among the 875 respondents who identified themselves as "White",
- 328 chose "Muslim" (coded as 1)
- 543 chose "Not a Muslim" (coded as 2)
- 4 skipped the question (coded as 8)
So now we're down to 328 respondents.
Among the questions asked of the respondents were:
- How would you rate Donald Trump? (
fttrump in dataset)
- How would you rate Hillary Clinton? (
fthrc in dataset`)
where respondents could specify any value from 0 to 100.
Of 328 respondents who identified themselves as "White" and also answered "Muslim" to the question about whether Obama is Muslim, 285 gave Donald Trump a higher rating than Hillary Clinton. 43 did not.
Finally, let us weight the samples as directed by the survey organizers (using the
weight variable). As per the User Guide:
The variable called “weight” on the ANES 2016 Pilot Study data file is the weight for analysis that is intended to generalize to the population. ANES strongly recommends utilizing the weights provided with this data set when making inferences to the target population of U.S. adult citizens.
The sum of the weights of the 328 respondents who identified themselves as "White" and also answered "Muslim" to the question about whether Obama is Muslim is 332.8443. The sum of the weights of the 285 members of this group who gave Trump a higher rating is 295.5019. 295.5019/332.8443 = 0.8878082
That's where Klinkner got the 89% number. But, this number is based on 328 respondents, on a survey that was really only intended for testing new questions for another survey, and may not generalize to the rest of the population.
Also note that
You can ask just one simple question to find out whether someone likes Donald Trump more than Hillary Clinton: Is Barack Obama a Muslim?
is not really an accurate characterization of the ability of this question to distinguish between those who like Trump more than Clinton and those who don't.
Among the 543 respondents in the survey who gave Trump a higher rating than Clinton, about 60% said that Obama is a Muslim: the sum of the
weight variable for respondents where
fttrump is greater than
fthrc and who didn't skip the Obama Muslim question (
bo_muslim not 8) is 565.0238, and the sum of the
weight variable for the subset of those that answered "Muslim" to the Obama Muslim question (
bo_muslim is 1) is 337.6951. 337.6951/565.0238=0.5976653.
So while 89% of those who answered "Muslim" to the Obama Muslim question rate Trump more highly than Clinton, only 60% of those who rate Trump more highly than Clinton answer "Muslim" to the Obama Muslim question. If someone answers "Not a Muslim" to the Obama Muslim question, this is only a weak indicator that they rate Clinton more highly or the same as Trump.