A new op-ed published in The Hill from Ali Valenzuela, an Assistant Professor of Politics and Latino Studies at Princeton University, and Tyler Reny, a Ph.D. student at the UCLA Department of Political Science, summarizes their recent analysis of voting precincts in Florida to explore how the state’s Hispanic electorate actually voted in 2016. Despite the national exit poll claims to the contrary, Donald Trump actually fared worse than Mitt Romney among Hispanic voters in Florida, according to the precinct-level analysis, and by a margin close to the Latino Decisions election eve poll.
The new Florida precinct analysis mirrors similar precinct analyses in Arizona and Texas, both of which found that the exit polls missed the mark estimating Latino voters in 2016. As a reminder, the controversy over accurate Latino polling is not a mere academic exercise. Having an accurate sense of how Latinos actually voted and turned out is incredibly consequential, especially at a time when the Latino and immigrant communities are vulnerable and fearful in the aftermath of Donald Trump’s election. Selling the Latino vote short, as the national exit polls do, essentially diminishes – even disenfranchises – Latino voters at a moment of maximum peril and tells a factually incorrect story about what happened in 2016.
See below for The Hill op-ed from Professor Valenzuela and Tyler Reny, “Study: Trump fared worse than Romney in Florida Hispanic vote” (available online and with graphics here):
“The 2016 National Exit Poll, a proprietary survey conducted by Edison research, reports 29 percent of Hispanic voters nationwide supported president-elect Donald Trump, a two percentage-point increase over Mitt Romney’s 27 percent in the 2012 NEP.
We are not the first to argue that it seems implausible that Trump increased his Latino vote share despite his anti-immigrant rhetoric.
In contrast the Latino Decisions election eve poll reported Trump doing worse than Romney among Latinos, with a record low 18%.
Recently other social scientists have pointed out that official precinct data in Texas and Arizona both indicate that Trump received a record low Latino vote, and that exit poll data for Latinos seemed wrong.
Given the public availability of official precinct data, it is not difficult to extend this analysis to other states. Here we examine the Latino vote in the critical battleground state of Florida.
In addition to the exit poll numbers suggesting Trump won more Latino votes, pundits such as Harry Enten and Nate Cohn have looked at county-level voting patterns and also agreed that Trump seemingly did better with Latinos than Romney.
However, counties are far too large of a unit to allow inferences to individual voters. Miami-Dade, for example, had around 975,000 votes cast.
Looking just at counties is misleading, and smaller levels such as voting precincts provide much greater precision when attempting to make individual inferences.
Consistently, research has shown that exit polls under-samples voters who are more likely to support Democrats: those who live in majority-minority precincts, immigrants, non-college educated Latinos, and those who prefer to complete a survey in Spanish – all issues leading to bias in the results.
This is not a methodological point: if Trump can threaten to revoke birthright citizenship, establish a deportation force to round up and expel millions of immigrants and their families, and engage in racist accusations against an American-born judge on the basis of his parents’ ethnic heritage and still come away with more support than his predecessor did, why should either of the parties make positive appeals to the Hispanic community?
Here we look more closely at Florida, and demonstrate that even among the most conservative segment of the Hispanic community—Cubans in Miami-Dade—Trump did worse than Romney and Clinton did better than Obama.
Moreover, at the statewide level, our precinct-level estimates of Hispanic support look significantly worse for Trump than do the exit poll results.
We use this evidence to argue for renewed commitments to the Latino community and its policy preferences.
In at least one key respect, Florida’s Hispanic community is significantly different than nationally: 29 percent of Hispanic Floridians are Cuban, and they’ve voted for Republicans, historically.
This is due, in part, to their greater economic resources, privileged asylum status and historical patterns of support for Republicans fighting against Castro’s communism.
Still, the Cuban-American community is undergoing change. Younger U.S. born Cubans are less likely to cast a party-line vote for Republicans and some older Cuban immigrants are less convinced an isolationist-only approach to the island is worthwhile amid the opening of relations with Cuba during the Obama years.
So what was the real Latino vote in 2016, and did Trump best Romney among Florida Hispanics? To answer this question, we turned to official vote tallies reported in 3,216 individual precincts in Florida, the smallest geographic unit reporting vote results that are available.
Our approach is superior to county-level analysis because we have many more data points available for analysis and because we make fewer assumptions about their demographic composition.
Counties are large and heterogeneous, and even the most heavily Hispanic counties contain sizeable non-Hispanic populations that limit inferences one can make about Hispanic voting behavior.
By contrast, precincts allow us to identify much smaller units that are greater than 95 percent Hispanic and compare them to precincts which are 50 percent or 10 percent Hispanic, which brings us much closer to identifying the voting preferences of individual Hispanics.
The method is based on aggregate data, a tricky endeavor that we accomplish through ecological inference, a technique developed by Harvard researchers.
Note that our argument is not merely one about Hispanic voting in heavily Hispanic precincts: our data in Florida contain all available precincts for 15 counties, with large and small Hispanic populations.
In the interest of transparency, we have posted all data to this website for replication.
We used two approaches to answer the question of whether Clinton underperformed Obama in 2012 as Enten and Cohn have suggested. For each precinct in Florida we first took the net raw vote margin for Clinton in 2016 (i.e., Clinton votes – Trump votes) and subtracted from it the net raw vote margin for Obama in 2012 (i.e. Obama votes – Romney votes) – a similar approach taken by social scientists in their analysis of Texas precincts.
We then plotted the results by the percentage of all voters who are Latino in each precinct along the x-axis as shown in Figure 1. The data show that the larger the Latino population in a precinct, the more the mass of red dots shrinks while the mass of blue dots grows and gets darker.
In other words, for precincts that are 75 percent or more Latino, Clinton won more net votes than Obama in 88 percent of them.
In only 12 percent of Latino precincts did Clinton “run behind” Obama. Already this basic precinct data suggest Clinton outperformed Obama among Florida’s Hispanic population.
Second, we calculated the vote share in each Florida precinct for Trump, Clinton, Romney and Obama, and plotted each of these.
Each precinct is weighted by population size (and drawn accordingly) and is drawn four times, once for each candidate. To detect trends in this mass of data, we overlaid smoothing regression lines for each candidate.
Two results stand out. First, at every level above about 40 percent Latino, Clinton outperformed Obama and Trump underperformed Romney in Florida precincts.
The more heavily Latino a precinct was, the more Clinton outpaced Obama.
Second, Trump (and Romney) beat Clinton (and Obama) in the most heavily Latino precincts in Miami-Dade. Why?
Because of the high concentration of Cubans in Miami-Dade County. Zooming in to only precincts in Miami-Dade, the pattern is clear: large concentrations of Cuban-origin Hispanics voted for Trump.
But even here, at every level of Hispanic concentration above 40 percent, these precincts voted more for Clinton than did for Obama four years earlier.
So while Trump still won a reasonable number of votes in Cuban-American enclaves in Miami, there is no question that he won far less than Romney from Latinos in Miami.
Outside of Miami, high-density Latino precincts voted very heavily for Clinton at rates higher than Obama, and in Central Florida the total number of Latino votes increased substantially over 2012 numbers.
Finally, we used ecological inference (EI) to estimate the vote preferences of all Latinos in Florida and have broken out the results by Miami-Dade and the rest of the state.
In Miami-Dade County, across 750 precincts we estimate 48.7 percent of Latinos supported Clinton compared to 46.7% for Trump.
Looking at 2012 data, the same model estimates 44 percent for Obama to 55.2 percent for Romney.
So Clinton cut a deficit of 11.2 point to an advantage of two points among Latinos in Miami-Dade County. This is corroborated by countywide vote totals, which show Clinton bested Trump by 291,000 votes in 2016 while Obama beat Romney by 208,000 votes in 2012.
There is no question that Clinton won more votes in Miami-Dade than Obama and won more votes among Latinos across Miami-Dade.
Among Latinos in the rest of Florida, our EI analysis of 2,466 precincts from Orange, Osceola, Broward, Palm Beach, Hillsborough, Collier and other counties suggests Clinton won an estimated 84.2 percent of the Latino vote to just 6.4 percent for Trump.
Despite a very large media focus on Latinos in Miami-Dade, Latino voters outside of Miami account for about two-thirds of Latino voters statewide in Florida, and Clinton dominated among non-Miami Latinos.
When combined for an overall statewide Latino vote, the full set of 3,200 precincts estimates 66.9 percent for Clinton to 30.7 percent for Trump.
When compared to the exit polls and Latino Decisions election eve poll, our overall estimate of 66.9 percent is a match to the LD estimate of 67 percent for Clinton, while the exit poll estimate of 62 percent appears too low.
Thus, if the national exit poll is correct that Trump increased his Latino vote share over Romney, then it did not happen in Florida.
In Florida, the official precinct data is all to the contrary. Trump fared much worse than Romney among Florida Hispanics while Clinton improved over Obama. Similar analysis in Texas, Arizona and California has concluded the same —Trump did worse than Romney with Hispanic voters.
Given these findings, we find it implausible that the national exit poll is correct when it comes to the Latino vote — the numbers just don’t add up.”