Last week, we cautioned about the effects of faulty polling of Latinos in battleground states on pre-election predictions and post-election analyses.
In a new piece for Univision, David Adams outlines pollsters’ common mistakes when interviewing Latino voters, including small sample sizes and targeting the ““wrong Latinos”…By not offering interviews in Spanish and relying on interviews via internet and fixed home phone lines, they end up with a biased sample of more assimilated, native-born, higher income and higher educated voters, according to internal poll research conducted by Latino Decisions.”
For more information about what constitutes a good vs. bad Latino poll and why both many pre-election polls and the national exit polls miss the mark when it comes to Latino voters, check out a recent panel discussion from Latino Decisions, NCLR, Catalist and Professor Alan Abramowitz.
Read the entire piece “Do polls underestimate the Democratic Party’s Latino vote?” below or online here.
“Hillary Clinton may be popular among Latino voters, but Democrats have expressed concern for months about a traditionally poor Hispanic turn out hurting her chances of election.
However, some Latino pollsters say an overlooked factor in mainstream polling data could produce a welcome surprise for Democrats on November 8, especially in several states with large Hispanic populations, such as Florida and Texas.
“In many states with large Latino populations, polls underestimated the Democratic advantage,” according to Gabriel Sanchez, a political scientist at the University of New Mexico who is one of the principals at the polling firm Latino Decisions.
That’s because the national polls aren’t really focused on the Latino voters, and only have a small sample of Latinos, resulting in a larger margin of error, he explained.
When a sample is only 100 to 200 voters, the margin of error can leap between 10-14 percent. Most polls tend to sample at least 1,000-1,500 voters, with a margin of error of only 3 to 4 percent.
The polls also target the “wrong Latinos,” making their sample unrepresentative of eligible voters, said Sanchez. By not offering interviews in Spanish and relying on interviews via internet and fixed home phone lines, they end up with a biased sample of more assimilated, native-born, higher income and higher educated voters, according to internal poll research conducted by Latino Decisions.
First-generation Hispanics speak less English, rely more on cellphones and often don’t have internet, he noted.
Latino Decisions, some of whose team are working for the Clinton team, did a webinar earlier this month in an effort to alert the media to two key mistakes it has detected in polling. The webinar was conducted by Sanchez, who is not working for the Clinton campaign.
It cited statistical analysis this year by David Damore, a researcher at Latino Decisions, who found “clear evidence” that Latino respondents who appear in mainstream national polls are statistically more likely to be Republicans.
Eduardo Gamarra, a Miami pollster focusing on the Latino vote, agrees and blames the faulty methodology on cost. “It is too expensive to draw large Latino samples if you also are trying to grasp the national market,” he said.
More targeted Hispanic polling conducted for media outlets such as Univision and Telemundo has consistently found stronger support for Clinton than the mainstream national polls.
For example, a recent Bloomberg poll that showed Trump leading Clinton by two points in Florida found the Democratic candidate had only 51 percent of the Sunshine State’s Hispanic vote. The survey of 953 registered voters included only 148 Hispanics.
That contrasted sharply with a poll this week conducted by Gamarra and Adsmovil for Voice which found Clinton leading among Florida Hispanics 64.4 percent to 26.1 percent for Trump. Her lead was even greater nationally, 75.7 percent to 16.1 percent.
A Univision poll of Hispanics found Clinton leading Trump by 30 points (58-28 percent) in Florida in early October. That survey consulted 400 voters with a 4.9 percent margin of error.
Hispanic pollster Fernand Amandi pointed to a recent Nevada survey by Marist Poll for NBC and the Wall Street Journal that had Clinton and Trump tied 43-43. A poll by Amandi’s firm, Bendixen & Amandi International, had Clinton leading by 7 percent, 48-41.
“It’s not necessarily about faulty methodology. It’s more about general challenges associated with reaching the entire Latino community,” said Geoffrey Skelley, at the University of Virginia Center for Politics.
“An obvious problem for some pollsters is the lack of interviewers who speak Spanish. That’s an additional cost that many may not be able to afford. And this language problem could also pop up if a pollster is using internet panels because they may not offer questions in Spanish,” he said.
Such polling errors may not have been a big deal a few decades ago, but the rapid growth of the Latino population has made measuring it accurately much more significant nationally. “In states where a sizable number of Hispanic voters will cast ballots (i.e. Colorado, Florida, Nevada), a large underestimation of Latino support for Clinton could throw an entire poll’s topline result off,” said Skelley.
In Florida, Clinton could have a larger lead than most polls show. In Texas, Trump’s shrinking lead could be even thinner.
Lee Miringoff, director of the Marist Institute for Public Opinion recognized that Latino polling lacks precision. In the case of Nevada he noted the sample was more male and more Republican than he would have liked. He noted that the sub-sample of Latinos had an 8-9 percent margin of error, while the overall poll of 985 voters had a 3.1 percent margin of error.
Unlike some polls, Miringoff said Marist did use live operators to make bilingual phone calls, and also included cellphones in the survey. Even so, the results could be misleading due to the small sample, he conceded.
He expected future polling to improve as the Latino population grows requiring larger samples.
Latino Decisions began analyzing polling methodology after the 2010 elections when they noted a sharp discord between pre-election polling, exit polls and the real results. When they examined the methodology, they found “dead giveaways,” he said, including weak sampling of Latinos and methods of selecting respondents.
In 2010, more than a dozen pre-election polls showed Sharon Angle led incumbent Senate Majority Leader Harry Reid by 3 percent. But Reid ended up winning comfortably by almost 6 percent. Similarly, in Colorado, pre-election polls showed incumbent Democrat Michael Bennet trailing Republican Ken Buck, yet he won by almost 2 percent.
In Nevada, Latino decisions found that most polls had only small Latino samples and woefully underestimated Reid’s support among Hispanic voters by as much as 20 to 35 points.
Post-election analysis at the time by Nate Silver at polling analysis site FiveThirtyEight showed that many states with large Latino populations polls underestimated the Democratic advantage.
Silver found that Democrats outperformed their polls by 2.3 points in 15 races in the eight states with the largest share of Latinos in their population: Arizona, California, Colorado, Florida, Nevada, New Mexico, New York and Texas.
While Silver did not reach a firm conclusion about possibly inadequate sampling of Latino voters, he detected “the beginnings of a pattern.”
Considering how rapidly the Latino population is growing, he added it was a pattern “that pollsters are going to need to address in states like Nevada, California and Texas if we’re going to be able to take their results at face value.”
Latino Decisions found similar issues in the 2012 presidential election. In late October the respected Monmouth University poll had Republican Mitt Romney ahead nationally by 3.5 percent, with Obama leading among Hispanics by only 6 percent.
By contrast, most Hispanic pollsters, including Univision and NBC-Telemundo, had Obama with a commanding 48 percent lead among Latinos, giving him a 1.5 percent edge nationally.
Obama would go on to win the election 3.9 percent, a margin of 5 million votes.
“We are seeing the same thing with Donald Trump and Hillary Clinton,” said Sanchez.”