In the dramatic aftermath of last week’s election, many Latino leaders and Democratic Party strategists were stunned when it was revealed that fewer Hispanics voted for Hillary Clinton in 2016 than for Barack Obama in 2012.
How could that be, given the wall-building, anti-immigrant rhetoric of Donald Trump?
Some questioned the exit poll data used by mainstream media, citing a different survey of Hispanic voters that had showed Clinton doing far better.
“Is this really the Hispanic community?” asked Matt Barreto, with Latino Decisions, a Seattle polling firm that did work for the Clinton campaign and conducted its own exit poll.
He questioned the methodology of the New Jersey firm that conducted the exit poll in question, Edison Research. “The exit poll is just another poll. And not a particularly good one,” he said. “It’s just one of many surveys, yet they treat it like the Bible.”
For his part, the co-founder of Edison Research, Joe Lenski, defended the exit poll in an interview with Univision and accused Latino Decisions of a “jihad,” adding that Barreto had a partisan agenda "to make Hillary Clinton look good.”
He added: "Latino Decisions is ... an advocacy group and they will disagree with any finding that does not match their findings."
So why all the bad mouthing over the poll numbers?
Univision delved into the data and found that the issue between the pollsters raises tough questions about the size and ethnic identity of the nation’s fast-growing Latino community.
Pollsters cannot even agree on who they consider to be a Latino voter. Then there are issues regarding the geography and demography of the sample of voters interviewed, including their level of education, wealth and how recently they arrived in the United States.
Part of the reason for the shrill language is the wide variation in the voting results pollsters came up with.
According to the Edison exit poll—known as the National Election Pool and commissioned by big media companies including CNN, AP and Univision—Clinton won the Latino vote by a whopping 36 points: 65-29 percent.
Sounds impressive, right? Not exactly. In 2012, President Barack Obama beat his Republican challenger Mitt Romney by 44 points: 71-27 percent, according to the exit poll that year, also carried out by Edison.
Edison surveyed 25, 000 people around the country. Of those 2000 were Hispanics. The interviews were 60 percent at precincts and 40 percent by phone
“I’m suspicious of the (2016) exit poll,” said Alan Abramowitz, a professor of political science at Emory University in Georgia. “It just doesn’t make sense that Trump would do better with Latinos than Romney, who didn’t go out of his way to insult them in the campaign.”
“The numbers defy common sense,” added Frank Sharry, director of America’s Voice, an immigrant rights group in Washington D.C. “The analysis shows that Latinos turned out in record numbers and Trump got fewer votes than Romney nationally,” he added, noting that Romney was much more respectful of Latinos.
What's at stake
More than 27 million Latinos were registered to vote this year, but only half - some 13 million - actually cast ballots, up slightly from 2012. Clinton's advantage with Latinos was overwhelmed by the wide margin of white, non-Hispanic and less educated people in several key states – Florida, Pennsylvania, Wisconsin, Ohio and Michigan - who supported Trump.
To be sure, most pollsters were offering mea culpas the day after the election, as most had predicted a Clinton victory. With their reputation in tatters, polling firms are busy examining where their methodology let them down.
For Latinos, there is much more at stake than the reputation and methodology of Edison Research and Latino Decisions: immigrant rights groups stress that an accurate assessment of the Latino vote in 2016 is crucial to the perception of Hispanic electoral power in the wake of Trump’s victory.
“If a narrative takes hold that Latinos underperformed at a moment of maximum peril for the community, that is not only inaccurate, but dangerous,” said Sharry. “The Republicans will feel emboldened about attacking them and Democrats will be less likely to defend them,” he said.
"We need to know the truth," Xavier Becerra, chairman of the Congressional Democratic Caucus, told Univision.
Becerra said that due to concern about “widespread misrepresentation by top media outlets” of the exit poll, he was reaching out to Edison Research, to request disclosure of the data and methodology.
Others say Latino Democrats are crying over spilled milk. "They are trying to push this narrative that there is a monolithic Latino voting block. That's not so," said Alfonso Aguilar, a Republican immigration lawyer in Washington and president of the Latino Partnership for Conservative Principles. "This is ridiculous. They did well. Hillary won the Latino vote," he added, noting that he did not vote for Trump.
Critics of Edison's exit poll argue that it failed to accuretely represent Hispanic voters due to faulty methodology that missed high-density, low income precincts where Hispanics heavily favored Clinton.
Instead, they hail the rival poll by Latino Decisions, which found that Hispanic voters backed Clinton by a much larger 60-point margin, 79-18 percent. That was nine points better than Obama’s 2012 margin over Romney (75-23 percent), according to polling by Latino Decisions at the time.
Latino Decisions focused its effort entirely on Latino votes, with a sample of 5,600 respondents, all interviewed by phone
So who's right?
The two companies use different methodologies, which makes that question hard to dechipher. Also, both companies continue to crunch precinct data as the final election numbers come in.
A clearer picture will become available in another five months from the 2016 American National Elections Study, a detailed voter survey by the University of Michigan, Stanford University and the National Science Foundation. The U.S. Census Bureau will also issue its own voter survey that collects data on turnout, age, race and origin.
Edison surveyed 25,000 people across the country: 60 percent at precincts and 40 percent by phone. Of those, 2,000 were Hispanic, according to Lenski. The number of respondents varied by state, with 1,466 in New York and as many as 4,000 in Florida.
Edison found that 8% of all respondents choose Hispanics as their race and 3% identified as Hispanic/Latino descendants
Latino Decision’s election eve poll, on the other hand, focused its effort entirely on Latino voters, with a sample of 5,600 respondents, all interviewed by phone.
Defining a Latino voter
Part of the confusion stems from the definition used by Latino Decisions and Edison Research to identify Latino voters in their questionnaires. Latino Decisions said it asks all respondents to self-identify using one simple question about whether they are White, Hispanic, African-American, or Asian.
Edison Research said it asked the same question, but then added a follow-up about ancestry: “Are you of Hispanic or Latino descent?”
Edison found that 3 percent of all respondents did not choose Hispanic in the race question but did choose Hispanic in the Hispanic/Latino descent question, including 2 percent of whites, 4 percent of blacks, 1 percent of Asians and 15 percent of “others,” likely those who self-identify as multi-racial.
Lenski said Edison believes this two question method helps to capture more assimilated Hispanic citizens, who live in the suburbs, don’t speak Spanish and are less impacted by immigration issues as they have a longer family history in the country, perhaps going back several generations.
The results of the two questions are revealing. Those who self-identified as Hispanic in question one voted 70-25 percent for Clinton. But that margin dropped dramatically among those who said they were of Hispanic descent in question two: 50-43 percent for Clinton. When the two were merged, Clinton’s support fell from 70 percent to 65 percent and Trump's rose from 25 percent to 29 percent.
Clinton's support fell with the Hispanics descendants when the two were merged
Lenski said the Edison method reflects a new standard for measuring race and ethnicity, one introduced by the U.S. Census Bureau in 2010.
Critics say Edison is casting too wide a net.
“That’s the first I have heard of that,” said Barreto. “It goes against social science. Almost everyone uses the first definition of what you consider yourself.”
Some pollsters agree with Barreto.
“You never ask that (ancestry) question. We just don’t do that,” said Eduardo Gamarra, a political scientist at Florida International University, who also ran a Hispanic focused election polling operation, New Latino Voice, in partnership with a mobile phone advertising company, Adsmovil. “In polling we always ask people to self-identify.”
New Latino Voice's final election eve poll gave Clinton 71 per cent and Trump 23 percent.
A Los Angeles Times/USC post-election poll - which correctly predicted a Trump victory - was much closer to Edison's exit poll and gave the margin among Latinos to Clinton by 67-28 percent.
The director of Hispanic research at Pew Research Center, Mark Hugo Lopez, said the Edison method raised legitimate, long-standing issues regarding Hispanic identity.
“We talk about racial cleavages in the United States, but it’s not as straightforward as it might seem,” he said.
Addressing the exit poll, he said: “Some people may answer ‘White’ for the race question and then Hispanic for ethnicity. It sounds like Edison is making an attempt to have as wide a measure of ethnicity as possible. Is it too wide? I don’t know.” (In the 2010 Census over half of the Hispanic population identified as White.)
Race and ancestry
Issues of ancestry were generally considered separate from polling the Latino vote, Lopez said.
Lopez said that Pew’s own polls ask one simple question: “Are you Hispanic or Latino?” A follow up question asks where the person was born and if they are a citizen.
“Ancestry.com style is not what we do. There is a possibility that there are people of Hispanic ancestry who don’t consider themselves Hispanic. We leave that up to them,” he said.
The U.S. Census Bureau does indeed ask multiple questions about race and ethnicity, though it uses a slightly different method from Edison, as it does not consider Hispanic or Latino as a race.
Instead, in its once-a-decade Census it asks first about Hispanic “origin,” with a list of countries to choose from. A second question asks about race, with a list including White, Black, Asian, and Chinese, but not Hispanic or Latino.
The 10-year Census does not specifically ask about a person’s ancestry.
A more detailed annual socioeconomic survey conducted the Census Bureau, the American Community Survey, does ask the question: “What is your ancestry or ethnic origin?”
It is sent to a small percentage of the population on a rotating basis throughout the decade.
The Census Bureau standards, which are set by the U.S. Office of Management and Budget in the White House, recognizes that race is a confusing issue with overlap between white Europeans and Hispanics, as well as blacks and Hispanics.
“We recognize that race and ethnicity are not quantifiable values,” the Census Bureau said in a 2013 report. “Rather, identity is a complex mix of one’s family and social environment, historical or socio-political constructs, personal experience, context, and many other immeasurable factors,” it said.
But, Latino Decisions says what concerns them even more than the Latino identity question is the selection of precincts by Edison.
“They handpick precincts and cluster them. They did not put a blindfold on and put their hand in the fishbowl. When you do that you haven’t let random chance come into play,” said Barreto. “If that precinct is an outlier and not representative of the general public you end up replicating an error, magnifying it.”
Lenski said Edison's exit poll was conducted at 350 clustered locations around the country, with additional surveys in 28 states. In total, Edison had interviewers at 933 locations nationwide. However, he declined to specify which precincts had been selected.
"We do not want those lists to get out and encourage outsiders either from the campaigns or other groups to send people to our exit poll locations to try to influence the outcome of the survey," he said. The company also wanted to protect the privacy of the respondents, whose data is kept in an archive.
That lack of transparency worries some critics who fear the exit poll may have focused too heavily on culturally mixed neighborhoods in an effort to capture a cross-section of the overall national population. Those precincts tend to be whiter, wealthier and better educated than the rest, said Barreto. Poorer, high-density minority neighborhood precincts were likely overlooked, he said.
“Unlike the exit poll, we have a distribution across all possible geographies.”
He pointed to Edison's own critique of its 2004 exit poll.
"A National Sample of 250 precincts can do a good job estimating all of the broad characteristics of the electorate, but it is not designed to yield very reliable estimates of the characteristics of small, geographically clustered demographic groups. These groups have much larger design effects and thus larger sampling errors," the company wrote.
Barreto cited U.S. Census data that showed 48 percent of Latinos live in majority-Hispanic neighborhoods, and doubted that the exit poll sample came close to matching that.
Edison told Univision that 40 percent of the Hispanic voters in the exit poll reported that they had a college degree. While that was higher than the national average - 30 percent - "there was no significant difference in our survey in how college educated Hispanic claim to have voted compared to non-college educated Hispanics," said Lenski.
He said the sample selection in the Latino Decisions survey targeted Hispanic surnames and focused too heavily on urban areas filled with Hispanic immigrant voters, who were more likely to support Clinton.
Barreto said Latino Decisions built its sample using ethnic voter file data to build a sample pool, and examined surnames of respondents. “About 80 percent of our respondents have a Spanish surname, and about 20 percent do not. This is exactly in line with Census estimates,” he said.
The human factor
Experts say it is virtually impossible to have a poll with perfect sample selection due to demographic uncertainties as well as human factors involving the voters and the exit poll workers.
The media often fails to recognize that, according to Jason Husser, director of the Elon University Poll at Elon University in North Carolina. "A major problem polls faced in 2016 was not in their data specifically, but in how those data were interpreted, either by pollsters themselves or by the media, Husser wrote in an op-ed published by Univision.
"At the end of the day, polls are but rough estimates of public opinion. They are the best estimates we have available, but they are still estimates – ballpark figures, not certainties."
Barreto doubted that enough exit poll interviews with Latinos are conducted in Spanish, noting that 30 percent of Hispanic voters in the United States prefer Spanish.
That could affect the exit poll’s accuracy as English-speaking Hispanics are more likely to vote Republican.
Edison told Univision that, depending on the state, between 4 percent and 14 percent of its exit poll interviews of Hispanics were conducted in Spanish.
“We offer Spanish-language phone interviews and we offer English- and Spanish-language questionnaires at all polling locations with greater than 20 percent Hispanic population according to the U.S. Census,” Lenski said.
Interviewers receive training that includes a manual, a 15-minute long training video online and a quiz. All interviewers also take part in a phone rehearsal the week before the election where they practice the procedure of reporting the results. “Supervisors monitor the locations to make sure that the interviewers are properly following procedures,” he said.
Regarding Spanish-language callers, Gamarra says it’s not how many are Spanish speakers but whether that option is available to every respondent. "Every single Latino voter must be given the option," he said, otherwise some will drop off.
It's also important that the caller have a neutral Spanish accent, as regional dialects can also result in response rate dropping. Cuban respondents who arrived in the country more recently are more mistrustful of government and also have a lower response rate, added Gamarra, who is Bolivian-born and has studied calling centers for langauge issues.
"As soon as you insert the human element in there you are bound to have human error," he added.
Edison said the exit poll had a 44 percent response rate on election day, which Lenski said was "higher than just about any other survey methodology out there."
Even so, he recognized that the 56 percent of voters who declined to participate may have differed from the 44 percent who agree to fill out the survey.
Experts say the debate between Latino Decisions and Edison illustrates that more needs to be done to accurately reach Latino voters. "We're just not very good at polling Latinos," said Trevor Tompson, director of the Associated Press-NORC Center for Public Affairs Research at the University of Chicago, who worked on the exit polls with Edison.
"(Latinos) are a confluence of a lot of groups that current polling just isn't good at reaching," he added, noting the language difficulty, as well as Latinos being a younger, more mobile demographic. "The industry needs to be more sensitive and adapt its methods to reach an increasingly diverse population."
The American Association of Public Opinion Research, the largest professional association of pollsters in the country, has already convened a panel of experts to examine the 2016 results.
Lenksi concedes Edison's poll is only a survey, so its results may not be accurate. But neither does he trust Latino Decision's numbers.
The real result may lie somewhere in between the two polls, given the +/- 4 percent margin of error, he suggested.
“There could be an overlap with the low end of their poll and the high end of ours,” he said, speaking of Clinton’s Latino support.
"This is not straightforward. There’s a lot of different ways to slice it."
(The exit poll data will be archived at the Roper Center for Public Opinion Research and is available for review by any subscribers to the Roper Center, which includes most academic institutions in the country. All exit polls going back to 1972 are available there. The 2016 exit poll data will be archived there in 6 to 12 months.)