This piece was originally posted at HuffPost Pollster, October 6, 2014
Georgia has more than 900,000 Latinos and more than 270,000 Latinos eligible to vote in 2014. For good reason, analysts have begun to identify Georgia as a battleground state where Latino votes could help determine the outcome of the gubernatorial and U.S. Senate elections this November. However in order to understand Latino political preferences in Georgia, or anywhere in 2014, election polling must be accurate, culturally competent and unbiased. Recently, NBC-11 in Atlanta produced what seems to be a “news story” with the headline, “Hispanic voter buck assumptions, back GA GOP candidates.” {screen shot of NBC-11 story}
The problem is that NBC-11 poll they rely on is based on a sample size of 38 survey interviews with Latinos, conducted entirely in English, and partially collected via smartphone app or tablet response {screen shot of Survey USA methodology}. The margin of error on a sample that small is a whopping +/- 16%. Moreover, since approximately 35% of all Latino registered voters nationally are Spanish dominant–and even more in immigrant heavy states like Georgia–the absence of Spanish language interviews skews the results significantly since it structurally fails to interview the most Democratic portion of the electorate. The Latino sample, and any inferences made from it, are both unreliable and biased.
This is not the first time that so-called “mainstream” pollsters have misunderstood the Latino vote. Most famously polls in Nevada in 2010 all forecast Democrat Harry Reid would lose to Republican Sharon Angle, and that Latinos were backing Angle at surprisingly high numbers, given that, you know, she ran one of the most anti-immigrant campaign in modern history. In the end, all the pre-election polls in Nevada were terribly wrong. Rather than lose by a predicted 5 points, Harry Reid actually won by more than 5 points. Even though election forecasting guru Nate Silver predicted Angle would win with 83% confidence, his own post-election analysis suggested that his prediction could have been based on biased polls in Nevada, with bad Latino samples in particular. The Nevada example is just one of many in which English-only, automated voice robopolls with tiny samples of Latinos introduce significant bias. Specifically, these polls tend to produce Latino samples with three-to-four times higher the income and education levels than what census data on Latino registered voters reports. Rather than review the length history of evidence of really bad polls of Latinos (which you can read here), let’s examine the Georgia example in a bit more detail.
First, assuming the NBC-11 poll followed all the best practices of having a reliable Latino sample — which it did not — a sample size of just 38 completed interviews yields a margin of error of +/- 16%. The poll reported Latino voters in Georgia supported Republican Nathan Deal over Democrat Jason Carter 40% to 29%. Critically important, the 11 point difference is well within the 16 point margin of error. Taking the margin of error into account, the Deal support levels could be as low as 24% and the Carter levels as high as 45%.
However, we have strong reason to believe the poll introduced additional bias that would lead to a conservative result.
Second, automated voice robopolls are known to produce a Republican bias. Academic and controlled-experiment studies have confirmed that conservative leaning respondents are more likely to participate in these robopolls and thus be over-represented in the polling data. With respect to Latinos, robopolls have an even harder time getting adequate response rates that are representative of the larger population.
Third, the poll was conducted only in English, and excluded any possible Latino voter who was more comfortable taking the survey in Spanish. According to the Latino National Survey, which interviewed 400 Latinos across the state of Georgia, using live callers, who administered the survey in English or Spanish, about 80% of the Latino adult population in Georgia is foreign-born. According to the LNS data, Latino respondents in Georgia preferred to be interviewed in Spanish at rates even higher than the national average.
Fourth, the reported results of this poll just don’t pass the common sense test. Republican Nathan Deal has a long history of supporting anti-immigrant positions. The Latino adult population in Georgia is overwhelmingly immigrant. Does anyone actually believe that plurality of Georgia Latino voters would support an anti-immigrant candidate? Of course not. In 2009, then Congressmen Deal proposed a bill to repeal birthright citizenship from U.S. born children whose parents are undocumented. His previous record in the House was voting yes on a bill that would require hospitals to check immigration status before they could deliver health care to suspected unauthorized immigrants. He consistently voted to build more fence and to further militarize the U.S. Mexico border. And Mr. Deal received a perfect 100% rating from the anti-immigrant group FAIR. Returning again to the Latino National Survey, 76% of Latinos in Georgia support providing in-state tuition rates to undocumented students who graduated from state high schools. As Governor, Deal has opposed extending in-state tuition to undocumented students. What’s more, only 4% of Latinos in Georgia support an effort to seal or close off the border to stop illegal immigration, instead more than 80% support a pathway to citizenship for those who are undocumented.
Given Deal’s history supporting anti-immigrant policies, and Georgia Latinos preference for pro-immigration policies, could it really be true that Deal leads Carter 4o to 29 among Hispanic voters in Georgia? No way.
Our conclusion? The poll reported by Survey USA for NBC-11 is not appropriate for estimating Latino vote preference in Georgia and, as a consequence, is definitely underestimating not just Latino support for Carter and Nunn and, by extension, their statewide standing.
