The polling industry is again trying to understand why some of its predictions for 2020 went so wrong, particularly the Latino vote in key states. It may have to look no further than the racial makeup of the pollsters themselves.

“There’s not a lot of diversity leading the industry,” said David Wilson, associate dean for social sciences at the University of Delaware and a member of the American Association for Public Opinion Research, which represents more than 2,000 polling organizations. “It matters when it comes to interpreting the data and asking the right questions to different populations.”

Pollsters tend to view Latinos as a monolithic group. Experts say that broad-brush approach probably contributed to this year’s lapses, figuring in polling that showed Democratic challenger Joe Biden ahead of President Donald Trump by 1% in Florida and within striking distance of winning Texas, trailing by slightly more than 1%. Preliminary election results show that Trump gained as many as 12 percentage points in Florida’s majority-Hispanic counties and seven percentage points on average in similar counties in Texas, compared with four years ago. Those margins contributed to Trump’s decisive win in both states.

The AAPOR plans to perform an autopsy of this year’s election, just as it did in 2016 after missing the shift in White working-class voters that helped Trump defeat Clinton. Many of those same patterns also were apparent in this election, along with some new twists that showed some Latino voters giving undetected support to Trump. Pollsters like Morning Consult, Quinnipiac and Ipsos either don’t track the diversity of their workforce or don’t disclose that information.Polling has been beset by a host of challenges during the past few years. Response rates among minorities have been dropping for a while. But so have those of White Americans. More people block or don’t answer unknown callers, requiring pollsters to make many more attempts to reach respondents. And even as more Latinos become fluent English speakers, language can still be a barrier.

“Hispanic and Latino voters are especially hard to reach because you need a Spanish-first caller, which increases your labor cost by between 20% or 40%,” said Dan Cassino, a professor of political science at Fairleigh Dickinson University who also works at polling research center PublicMind. “And labor is already your biggest cost.”

Voters who identify as Latino or Hispanic represent about 18% of the U.S. population, making them the nation’s largest ethnic minority group. The puzzle is why pollsters didn’t accurately predict the winner in Florida and Texas while they seemed to get it right in Arizona and Nevada, states that also have large Latino populations.


“This year the question is if the ones we interviewed are representative of their community,” said Courtney Kennedy, director of survey studies for Pew Research Center. “There were a number of polls conducted in both English and Spanish, but others were just in English. In the upcoming weeks we’ll really take a microscope look into whether excluding Spanish speakers might have played a role.”

One issue that confounds pollsters is that Latinos as an ethnic group are actually an amalgam of many different racial and national identities who sometimes share a common language and often little else. University of Delaware professor Wilson, who is both Latino and Black, thinks that even though the science behind polling is sound, the industry needs to reconsider how it looks at race “in a way that is not all encompassing.”

Divergent views of Cuban Americans and Puerto Ricans in Florida, or voters of Colombian and Dominican heritage in New York, are often overlooked, he said.

“We like to use artificial categories because they are easy,” he said. “If we start thinking about all the diversity within the Hispanic and Latino community, then ethnic categories don’t become useful anymore.”

For some, this year’s polling faux pas reflects bigger deficiencies in opinion surveys of Latinos, said Matt Barreto, co-founder of Latino Decisions, an opinion-research company focused on Latinos, Hispanics and other minorities. It wasn’t just that polling of Latinos has been inaccurate; it’s that there simply wasn’t much polling at all. Barreto said he started the firm in 2008 amid “immigration marches, with millions taking to the streets. But there was no data on them in any mainstream polling. We couldn’t find anything.”

To Barreto, whose firm has consulted for the Democratic Party and recently for the Biden campaign, the latest election cycle reflected sampling errors in the design of the polls rather than a shift of opinion among Latinos.


“They are not asking what percentage should be immigrant, or Spanish dominant, or have less than a high school degree,” he said. “So, the samples they get are way overly assimilated, too highly educated, English dominant and far more suburban than the true population. And that does not give us an accurate read.”

As demographics change, polling methodology might need to as well. Online surveys can be reliable for contacting younger Latinos, but older ones still rely on mobile phones and even landlines. Wording is also important: questions about race need to be carefully constructed to avoid alienating Black respondents. Respondents whose families include undocumented immigrants could be wary of answering pollsters, afraid of how the data might be used.

With the 2020 election over, the polling industry has begun once more to catalogue its shortcomings, including a lack of diversity. Correcting that could help the next set of polls to more accurately reflect the electorate.