The 2016 Democratic Presidential nominating process has become in part a competition over who can win over non-white voters. This is not surprising given that the Democratic Party is associated with representing the interests of minority voters. The Democratic Party needs these voters to turn out in November to defeat the Republicans. Heading into Nevada this weekend, Clinton and Sanders debated over whether Sanders could attract non-white voters in subsequent caucuses/primaries.
After the Nevada caucuses, the results of a CBS entrance/exit poll suggested that Sanders won the Hispanic vote in Nevada. According to the poll, 53% of Hispanics voted for Sanders while 45% of Hispanics voted for Clinton. This was good news for Sanders. However, after CBS released the poll, Nate Cohn at the New York Times questioned the results of the poll on many grounds. One criticism of note, that I think consumers of this poll must take into account, is the poll’s margin of error. I did not find it on CBS’s website, but as Cohn points out, a margin of error for this type of poll is +/- 6 percentage points. If this is the true margin of error for this poll, Clinton could have gotten anywhere between 39% and 51% of the Hispanic vote. Sanders could have gotten anywhere between 47% and 59% of the Hispanic vote. Hence, we cannot tell from the data who got the Hispanic vote from this poll. Since both candidates could have reasonably received 50%+ of the Hispanic vote, we cannot draw any conclusions from this poll about which candidate won the Hispanic vote.
Another one of Cohn’s criticisms of the poll results was based on precinct outcomes in Nevada. He displayed a map of the precincts Clinton won against a map with the demographic makeup of the precincts. He pointed out that Clinton won heavily Hispanic precincts, implying that this could mean she actually did win the Hispanic vote and the CBS poll was wrong.
At first glance, it may seem impossible that a candidate can win a precinct even if that candidate did not win over the key demographic in that precinct. However, it is very possible. Statisticians have long warned against using aggregate level data to make inferences about individual level behavior. The practice is known as the ecological fallacy.
An ecological fallacy occurs when we make conclusions about the relationships between individual level variables based on aggregate level data. In this case, Cohn is making a conclusion about the relationship between ethnicity and individual voting behavior based on aggregate level data. He is arguing that since precincts with large Hispanic precincts voted for Clinton, more Hispanics must have voted for Clinton over Sanders. However, as I will demonstrate below, even if Clinton did not get the Hispanic vote in some of these precincts, she could still win them based on how other voters in those precincts voted. In other words, it is possible that Hispanics did vote overwhelmingly for Sanders in some precincts, but their votes were not enough to overcome the vote share of other voters.
Take, for example, Clark Precinct 4551. It is a plurality Hispanic precinct. 47.8% of Democrats are Hispanic, 32.7% are white, and 19.4% are African American. If the individuals within each demographic group in this precinct voted based on what we expect from the CBS entrance/exit poll (45% of Hispanics, 47% of whites, and 76% of African Americans voting for Hillary), we should have expected Clinton to receive 51.6% of the vote in this precinct.
(47.8*0.45 + 32.7*0.47+19.4*0.76) = 51.6%
Clinton did in fact win this precinct. She received 7 delegates and Sanders received 2 delegates. Clearly, she won it by a higher margin than the predicted vote share. It is possible that this could be due to more Hispanics voting her- but it also possible that more whites or African Americans voted for her as well. It is also possible that Hispanics did not turn out in high numbers in this precinct. We simply do not know. We would have to look at the individual level data.
So, it is hard to say, based on the aggregate data if Clinton or Sanders won the Hispanic vote. And it is impossible to say based on the polls who won the Hispanic vote. We simply do not know.
(Thank you to L2 VoterMapping for access to precinct level demographic data in Nevada.)