Hispanic Heterogeneity in the 2016 Election

When exit polls broke on the night of the 2016 General Election, many found it difficult to believe their findings that every minority racial and ethnic group swung toward Trump compared to Romney, particularly in light of Trump’s highly-charged rhetoric on illegal immigration and Islamic travel restrictions. According to the final exit polls, Trump received 28% of Hispanic votes compared to Romney’s 27% in 2012, while Clinton received 66% compared to Obama’s 71%. Asians showed a similar shift, with the Democratic vote share falling from 73% to 65% and the Republican share rising slightly from 26% to 27%.

Hispanic-advocacy group Latino Decisions vehemently contested the claim of a Hispanic shift toward Trump on multiple occasions, claiming Trump received only 18% of Hispanic votes. The Washington Post used precinct-level results for Texas to argue for this number as well. In particular, their analysis relies heavily on the following argument:

Instead, we draw on 4,372 precincts across Texas. These precincts cover all regions of the state and more than 75 percent of its Latino population. You can download our data here.

The first key finding: compared with Obama, Clinton won more votes in almost all precincts — and especially in heavily Latino precincts. The graph below shows these patterns. In the 864 precincts in which 75 percent or more voters are Hispanic, Clinton won more votes than Obama in 723 of them, fewer votes in 130, and tied in 11.

If we compare Clinton’s vote margin over Trump to Obama’s margin over Mitt Romney four years ago, Clinton had a higher margin than Obama in 692 of these 864 precincts — or 80 percent.

There are a number of problems with this methodology, but the largest is this: Texas had close to 8,800 precincts in the 2016 General Election, far more than the 4,372 used in their analysis. In fact, their 4,372 precincts cover only 16 of Texas’s 254 counties, all of which but one are highly urbanized. I have no doubt that urban Hispanics likely did vote far more Democratic than Hispanics on average, just as urbanites in general voted more Democratic than the average American. To claim this is representative of all Hispanics is patently absurd, especially in an election so polarized on lines of population density. Hispanics may or may not have swung toward Trump, but they certainly did not give him only 18% of their votes.

Their claim also does not address the following facts: counties in Texas’s heavily Hispanic Rio Grande Valley almost unanimously swung toward Trump as did rural counties in similarly-proportioned north-central New Mexico. Simulatenously, Hispanic areas of Houston swung strongly to Clinton. This potential urban-rural divide has been noted by some authors, but was rapidly discarded by the Post‘s article as being only on a county-level of analysis. In truth, the urban-rural divide is not the whole story of the Hispanic vote; highly Hispanic precincts in urban areas such as Denver, Albuquerque, San Antonio, Phoenix, and Los Angeles all either swung toward Trump or swung toward Clinton by substantially smaller margins than neighboring majority-white precincts.

A more comprehensive analysis of swings in precincts across the country, both urban and rural, can shed some light on the matter. Here are the average swings between 2012 and 2016 in majority Hispanic precincts across the country based on 2015 ACS census tract demographic data:

Positive number indicates pro-Clinton swing Above Precinct Average Below Precinct Average
Proportion with a Bachelors Degree 4.94% 0.56%
Median Household Income 4.85% 0.36%
Median Age 1.13% 0.97%
Disabled Population -1.68% 2.30%
Area Classified by the Census Bureau as Urban 1.24% -1.98%

 

Broken down by state:

Positive number indicates pro-Clinton swing Overall Urban Rural >0.38 Coll. Degree <0.38 Coll. Degree
Arizona 2.23% 4.01% -1.42% 4.60% 2.04%
California 3.18% 3.14% 3.45% 8.19% 2.79%
Colorado -11.85% -11.27% -14.19% -5.10% -12.20%
Connecticut -7.09% -7.09% N/A 3.01% -7.43%
Florida 4.06% 4.87% -9.69% 11.64% -0.29%
Georgia 6.36% 6.36% N/A 14.44% 5.81%
Illinois 1.30% 1.32% -3.46% 4.40% 1.08%
Kansas -0.37% -5.98% 3.41% N/A -0.37%
Nevada -4.13% -4.18% -3.10% N/A -4.13%
New Jersey -7.34% -7.34% -7.28% -6.22% -7.43%
New Mexico -6.39% -3.77% -8.67% -0.27% -8.24%
New York -4.71% -4.64% -29.03% -1.90% -5.14%
Oregon -2.92% -5.33% -0.52% N/A -2.92%
Pennsylvania -3.48% -3.48% N/A N/A -3.48%
Texas 3.24% 5.73% -4.08% 10.25% 2.47%
Washington -0.12% -4.89% 2.51% N/A -0.12%

 

Texas, the basis of the Post‘s urban-based Hispanic analysis, shows a large urban-rural divide, but while urban Hispanic areas generally swung more toward Clinton than rural Hispanic areas, the effect was not as universally true as might be expected. All three Pacific Coast states where Hispanics tend to be heavily employed in rural agriculture saw a reverse phenomenon, for example. The college degree columns were more consistent in the story they tell; in every state, more educated Hispanic areas swung more to Clinton than less educated regions. However, there does not appear to be any consistent reason explaining the between-state discrepancies. The most heavily urbanized Hispanic population was in New York, where such precincts swung toward Trump, while the most heavily rural Hispanic population was in Washington, which had close to zero average swing. Colorado, which saw a massive swing in Hispanic areas toward Trump, had an average level of college education among Hispanic precincts in these states, as did Texas and California which saw swings toward Clinton. While the theory does tend to hold in the aggregate, the story still feels rather lacking.

The observation of large swings toward Trump in the Rio Grande Valley and in north-central New Mexico provides another possible hypothesis: in both of these areas, a large proportion of Hispanics are what’s known as Hispanos, direct descendants of Spanish colonists who began settling these areas over twenty years before the Mayflower landed at Plymouth Rock in 1620. These groups have little natural inclination to care about immigration policy, as from their communities’s prospective most other Americans are the immigrants. Hispano culture developed disconnected from broader Mexican-Hispanic culture, developing their own dialect of Spanish which mixes Native American terms with circa 17th century colloquialisms, not completely unlike the isolated Elizabethan-speaking population of Tangier Island in Virginia and other areas of the Tidewater and Outer Banks.

Using ACS ancestry data, we can look at the average swings of precincts heavy in particular Hispanic subgroups:

Positive number indicates pro-Clinton swing Average Swing Avg Urban % Avg Degree %
>40% Hispanic, all precincts 0.78% 83.40% 22.21%
>40% Hispanic, majority of whom are of Cuban descent 8.66% 97.88% 35.78%
>40% Hispanic, majority of whom are of Mexican descent 1.73% 80.15% 21.15%
>40% Hispanic, majority of whom are of South American descent -1.15% 99.67% 33.03%
>40% Hispanic, majority of whom are of Dominican descent -4.48% 100% 22.76%
>40% Hispanic, majority of whom are of Puerto Rican descent -4.84% 99.21% 21.68%
>40% Hispanic, majority of whom are of Hispano descent -9.21% 51.86% 28.72%

 

The strong swing among heavily Hispano precincts toward Trump remains even after accounting for turnout changes, urbanization, and a host of other demographics, while the others are more ambiguous. The swing for Puerto Ricans is of note for several reasons, but especially in light of the large immigration off the island to the greater Orlando area; while Orange and Osecola Counties swung toward Clinton, this may be due more to a compositional effect rather than the existing population swinging toward the Democratic Party. With the large disparity between Cuban and Hispano regions, painting the entirety of Hispanic voters with a single brush would seem overly simplistic.

To what does all this add up? How exactly did Hispanics vote? Did they swing toward Trump or Clinton? Some did and some didn’t, but it’s very difficult to identify precisely. The Post‘s article uses a court-sanctioned statistical technique known as ecological inference to estimate the (urban) Hispanic vote at 18% Trump, despite the fact that ecological inference often produces extremely unreliable estimates. The average more than 40%+ Hispanic precinct saw a 7.1% increase in turnout from 2012, 60%+ Hispanic a 9.1% increase, and 80%+ Hispanic a 12.5% increase in turnout. They averaged 66.1% Clinton, 71.7% Clinton, and 74.1% Clinton with +0.8%, +1.4%, and +2.1% swings, respectively. Was this from more Hispanics voting or from Hispanics changing their votes or both? Perhaps the fact that it’s quite difficult to say itself says something: either any change in Hispanic voting on the aggregate level likely wasn’t that large to begin with or Hispanic subgroups behaved in nearly off-setting ways such that identifying a singular “Hispanic swing” borders on meaninglessness.

I’m not going to patronize you by providing a flimsily-justified overall estimate based on flawed statistical techniques. Ultimately, these numbers don’t prove how different groups voted, which you can’t do without microdata, and the search for definitive numbers in this realm is both impossible as long as the secret ballot exists (political analysis and punditry in general could benefit from a healthy dose of epistemic humility) and perhaps misguided at its core: how much of the hemming and hawing over the exact Hispanic swing comes from a desire to know the mind of the Hispanic community versus assuaging the consciences of elites and advocacy groups who couldn’t come to terms with people behaving as they didn’t expect? “Hispanics couldn’t have possibly swung toward Trump or else I might have to acknowledge my worldview and priorities aren’t universal and what I care about others don’t. My community wasn’t responsible for electing Trump, it was the uneducated rubes and deplorables who brought us here.” I doubt they appreciate the irony of those who twist statistics to fit a narrative and discard all other information considering themselves the “better educated” ones.