And how exit polling is likely to get the Latino vote wrong, too
Many are surprised by the high turnout rates and evident determination of Latino voters to make their voice heard in this election. There are a number of factors for this surge: the growth of the Latino community; the hard work and sustained efforts of on-the-ground citizenship drives, voter registration drives and voter mobilization efforts; Trump’s nativism; Republican extremism; and the Clinton-Kaine campaign’s embrace of the new politics of immigration.
But to understand why this story is a surprise, we have to consider another factor: bad polling. Most general election polls simply do not adequately capture the Latino community. The main culprits:
Small Latino samples that don’t reflect their proportion of the electorate;
English-only operators who aren’t able to communicate to the roughly 30% of the electorate that prefers to communicate in Spanish;
Call-back methods for Spanish-dominant respondents that have notoriously poor success rates; and
Turnout models that discount and ignore first-time and low-propensity voters.
The result is polling that undercounts Latinos, and since most Spanish-dominant and low-propensity voters lean Democratic, undercounts the Democratic share of that vote (click here for a set of recent presentations by polling experts for analysis and examples).
For example, in-depth, high quality Latino polling conducted by Latino Decisions and Bendixen & Amandi for Univision/Washington Post have Hillary Clinton with a Latino lead in the high 50s, while most general polls have the lead in the 30s. According to Professor Alan Abramowitz, a political science professor at Emory University, “the difference between a 30-point Clinton margin among Latinos and a 58-point Clinton margin among Latinos would be approximately 3 points of the national vote margin – the difference between a 2-3 point national margin and a 5-6 point national margin.” (Click here for his PowerPoint presentation).
So, if Hillary Clinton wins the popular vote today by 5-6 points instead of the 2-3 point margin predicted by most polling averages, this could very well be one of the main reasons why.
The problem of capturing Latino voter behavior extends to the media-sponsored exit polling. Exit polling does many things well, but capturing the behavior of Latino voters is not one of them. This is because exit polls systematically misrepresent minority voters (in many states, the exit polls don’t distinguish Latino from African American). For example, in 2010 exit polls had minority voters with higher education levels than all other demographic groups in the country, something that is not supported by the research literature. In Arizona, the exit polls said 45% of minority voters had graduated college when the actual level is 24.5%; exit poll: 57.9% of Colorado minority voters with a college degree, actual level 28.3%; exit poll: 40% of Florida minority voters with a college degree, actual 27.4%.
Another example: In 2014, exit polls showed that Texas Governor Greg Abbott received 44% of the Latino vote. When it became known that the Edison positioned no exit poll surveyors in the majority Latino counties of the Rio Grande, researchers studied the actual voting support for Abbott from counties with Latino populations of 89% to 96% of the total population. The result: Abbot received approximately 25% of the Latino vote in the counties where most Latinos live — not 44%.
With respect to bad general polls and unrepresentative exit polls, Gabriel Sanchez of Latino Decisions sums it up in this way: “those missing in these polls are most likely to be Democrats: immigrants, Spanish-dominate, lower socioeconomic status, and younger Latinos.”
Leading political and media observers are picking up on the storyline that 2016 polling may have been missing or skewing Latino voters. See these pieces in Bloomberg, Univision, New York Magazine, and Politico, and check out recent Twitter observations from longtime Washington political observers Ron Brownstein and Norm Ornstein; Clinton campaign press secretary Brian Fallon; election forecasters and polling gurus like Nate Silver of FiveThirtyEight and Dave Wasserman of Cook Political Report and FiveThirtyEight; and other leading 2016 chroniclers like Jeet Heer of The New Republic.
Attempts by political scientists and polling experts to work with the Edison to improve Latino exit polling have been rebuffed to date. This is why we sponsor our own version of exit polling. Thousands of voters who are sure to vote, or have already voted, are surveyed in the three days leading up to Election Day. This year such polls will be carried out by Latino Decisions and Asian American Decisions, firms with expertise in capturing the voting behavior of their communities. The Latino Election Eve poll is a 5,600 person poll which will include both national figures and results from these twelve target states: AZ, CA, CO, FL, IL, NC, NV, NY, OH, TX, VA, and WI. A separate 2,150 person Asian American Election Eve poll will include both national figures and results from these eight target states: CA, FL, IL, NV, NC, PA, VA, and TX.