If you’re a fan of exit polls, you were probably surprised to see that according to the exit polls, Trump not only improved upon Romney’s numbers with whites without a college degree, but also did better than Romney among Hispanics. In 2012, exit polls showed Romney getting 27% of the Hispanic vote, compared to 71% for Obama. In 2016, exit polls showed Trump getting 29% of the Hispanic vote, compared to 65% for Clinton, a net gain of 8 points for Trump. Given Trump’s rhetoric towards Latinos, this Hispanic shift towards Trump seems incongruous with the prevailing narrative of Trump’s base being made up almost exclusively of white voters.
Of course, exit polls are often unreliable. In 2012, for example, exit polls in Ohio showed African American voters making up 15% of the Ohio electorate, which would represent an unprecedented 88% turnout rate among black voters, a huge uptick in turnout compared to 2008. Actual turnout statistics in black neighborhoods, however, showed turnout either remaining constant or falling. Black turnout in Cleveland, for example, dropped to 55%. Indeed, I suspect that a similar error occurred in the 2016 exit poll Hispanic vote estimates, and I’m not alone in my suspicion.
To understand why, let’s take a look at the most hispanic counties in the country. For our search, we’ll look at counties that are at least 65% Hispanic, and had at least 10,000 votes cast in 2016. Below, all the counties that met our criteria:
County | State | Population | Obama Vote% | Romney Vote% | Clinton Vote% | Trump Vote% | Net Shift Towards Trump |
---|---|---|---|---|---|---|---|
National Exit Poll | NA | NA | 71.0% | 27.0% | 65.0% | 29.0% | 8.0% |
Santa Cruz | Arizona | 46461 | 68.2% | 30.4% | 71.6% | 24.5% | -9.3% |
Imperial | California | 174528 | 65.2% | 33.1% | 68.2% | 27.1% | -9.0% |
Miami-Dade | Florida | 2693117 | 61.6% | 37.9% | 63.7% | 34.1% | -5.9% |
Dona Ana | New Mexico | 209233 | 55.9% | 41.1% | 53.7% | 35.9% | -3.0% |
Rio Arriba | New Mexico | 40246 | 74.7% | 22.1% | 64.4% | 24.2% | 12.4% |
San Miguel | New Mexico | 29393 | 76.9% | 20.0% | 67.8% | 21.5% | 10.6% |
Cameron | Texas | 406220 | 65.0% | 33.9% | 64.6% | 32.1% | -1.4% |
El Paso | Texas | 800647 | 65.5% | 33.1% | 69.1% | 25.9% | -10.8% |
Hidalgo | Texas | 774769 | 70.4% | 28.6% | 68.6% | 28.1% | 1.3% |
Jim Wells | Texas | 40838 | 58.1% | 41.2% | 54.2% | 43.8% | 6.5% |
Maverick | Texas | 54258 | 78.6% | 20.6% | 76.5% | 20.7% | 2.2% |
Starr | Texas | 60968 | 86.3% | 13.0% | 79.1% | 19.0% | 13.2% |
Val Verde | Texas | 51047 | 52.0% | 46.6% | 51.4% | 43.5% | -2.5% |
Webb | Texas | 250304 | 76.6% | 22.6% | 74.4% | 22.8% | 2.4% |
Weighted Average | NA | NA | 71.0% | 27.0% | 73.2% | 24.8% | -4.3% |
Latino Decisions Poll | NA | NA | 71.0% | 27.0% | 79.0% | 18.0% | -17.0% |
As you can see, there’s no evidence of a Latino shift towards Trump in the most heavily Latino counties in the country. Indeed, our county-based exercise showed a 4 point shift towards Clinton among Latinos, almost exactly in between the exit poll’s 8 point shift towards Trump and the Latino Decisions poll’s 17 point shift towards Clinton.
Another way of looking at the Latino vote is by looking at Nothing But Numbers’ Demographics Model. By setting the non-Latino demographics to the exit poll results, you can then isolate the Latino vote’s impact on the accuracy of the Demographics model in predicting the actual election results. Indeed, by comparing the National Exit poll with our County Based estimate and the Latino Decisions pre-election poll, you can actually test which Latino vote estimate was the best predictor of the final election results. Below, the results of said comparison. Remember, a higher R^2 Correlation and a lower Root Mean Square Error points to a more accurate Latino vote estimate:
Latino Vote Estimate | Latino Clinton% | Latino Trump% | R^2 Correlation | RMSE |
---|---|---|---|---|
Latino Decisions | 79.0% | 18.0% | 0.9295 | 3.47% |
Exit Poll | 65.0% | 29.0% | 0.9269 | 3.71% |
Nothing But Numbers County Estimate | 73.2% | 24.8% | 0.9292 | 3.54% |
As the chart makes clear, the Exit Poll does pretty horribly in this comparison, having the lowest R^2 Correlation while having the highest Root Mean Square Error. In addition, the Latino Decisions poll ends up being the most accurate estimate of the Latino Vote, with both the highest R^2 Correlation and the lowest Root Mean Square Error. Indeed, it seems that not only was the exit poll incorrect in it’s estimation of the Latino vote, it’s likely that Hispanics voted against Trump en masse, with roughly 80% of Latinos voting for Clinton.