Ruth Fremson / The New York Times
Tuesday, Feb. 23, 2016 | 2 a.m.
One of Saturday’s biggest election surprises was the entrance and exit polling measuring Hispanic voters in the Nevada caucus. It found that Bernie Sanders defeated Hillary Clinton by 8 percentage points among Hispanic voters, overturning months of conventional wisdom about Clinton’s strength among nonwhites.
But there are a lot of reasons to question the findings from the polls. They have a small sample of precincts and voters, and they simply were not devised to provide precise estimates of the Hispanic vote.
The actual election returns in Clark County hint at a different story. Analyzed neighborhood by neighborhood, they suggest that Clinton might have won the Hispanic vote by a comfortable margin. She won about 60 percent of delegates in heavily Hispanic areas, a result that calls the finding of the polling into question.
There is not much evidence, though, that Clinton won Hispanic voters by the sort of landslide margin that she did eight years ago. That’s a good sign for Sanders, who needs to make up for the huge swing among black voters, who have gone from uniformly for President Barack Obama to uniformly for Clinton.
But the finding that Sanders won the Hispanic vote is at best extremely questionable — and, at worst, wrong.
Clinton Won Heavily Hispanic Areas
The Hispanic vote in Nevada is overwhelmingly concentrated in Clark County, home to Las Vegas. In particular, Hispanic voters are concentrated on the east side of the city, where they make up the vast majority of the population but only a slight majority of registered Democrats.
In the 76 precincts in Clark County where we believe that a plurality of registered Democrats are Hispanic, Clinton defeated Sanders in the delegate count by a margin of 58 percent to 42 percent. In the smaller number of majority Hispanic precincts, she seemed to win about 60 percent of the delegates, and she won perhaps 65 percent of the delegates in the precincts where Hispanics appeared to be a particularly large share of registered Democrats. (For details on the estimates, see my note at the end of the article.)
Similarly, it was widely reported that Clinton fared extremely well in the precincts along the Las Vegas Strip.
Brian Fallon, Clinton campaign press secretary, tweeted that the campaign estimated that Clinton won 61 percent of the vote in Hispanic precincts throughout the state.
Does that prove the “entrance polls were wrong,” as Fallon said? Not on its own. It’s dangerous to infer much about the preferences of a demographic group based on the results of the areas where they live. Hispanic voters who live in majority Hispanic areas might be particularly likely to support Clinton. In other words, focusing only on the places where Hispanic voters are most prevalent could provide a misleading picture of Hispanic voters overall. We know that this is an issue in other contexts: Similar analysis would overestimate the Democratic share of the vote among Hispanic voters, or the Republican vote among white voters.
A Democratic pollster unaffiliated with either campaign said that Sanders does better among nonwhite voters in areas where whites represent a larger share of the population, which would suggest that a precinct-based analysis would understate Sanders’ strength.
There’s another problem: There were relatively few homogeneous precincts, where almost all of the voters are Hispanic. Even the majority Hispanic precincts in Nevada have large numbers of non-Hispanic voters, who themselves could have been likelier to support Clinton (though the findings still hold if one excludes the precincts where there appear to be an above-average number of black voters, who favored Clinton by a wide margin).
But it is a stretch to say that Sanders could have done so well among Hispanic voters in non-Hispanic areas that he overcame a deficit in relatively Hispanic areas, where a large proportion of the Hispanic population lives.
These Polls Are Hardly Perfect
In general, entrance/exit polls are not well suited to measure the Hispanic vote. This one is particularly problematic.
Overall, the poll included just 1,024 respondents — and just 213 Hispanic respondents. In a normal poll, that would imply a margin of error of plus or minus 6 percentage points.
But these types of polls are not normal. The margin of error is even higher, at 7 percent, and the true error can easily be much higher.
That’s because an entrance/exit poll isn’t a random sample of the population like a normal poll. It’s a random sample of precincts, usually between 15 and 60 in a state exit poll. This one had just 25 precincts. Race is not usually used as a criteria for selecting precincts.
Precinct selection introduces a whole new dimension of error into an entrance/exit poll for Hispanic voters, or any demographic group that’s concentrated in small areas. The 25 precincts could overrepresent an unrepresented group — imagine if a New York exit poll, by chance, included two Orthodox Jewish precincts in Brooklyn and showed a Republican ahead among Jewish voters — or completely miss a small, geographically concentrated group.
You can imagine how this could wind up creating the opposite risk of the precinct analysis: What if the entrance/exit poll simply doesn’t include a heavily Hispanic precinct where Clinton excelled? The danger is real with a sample as small as this, especially in a highly diverse state.
This type of problem is responsible for one of the biggest exit poll controversies in memory: the finding that 44 percent of Hispanic voters supported George W. Bush in 2004. Simply by chance, 3 of the 11 plurality Hispanic precincts were in Miami-Dade County, where Hispanic voters are particularly conservative. The 44 percent finding was 4 points higher than the result from a compilation of the 50 state exit polls — a difference well outside the formal “margin of error” for a sample of that size.
The exit pollsters are upfront about the issue, as they acknowledged in the exit poll’s postelection analysis in 2004:
“A National Sample of 250 precincts can do a good job estimating all of the broad characteristics of the electorate, but it is not designed to yield very reliable estimates of the characteristics of small, geographically clustered demographic groups. ... If we want to improve the National Exit Poll estimate for Hispanic vote (or Asian vote, Jewish vote or Mormon vote, etc.) we would either need to drastically increase the number of precincts in the National Sample or oversample the number of Hispanic precincts.”
And that’s for a national exit poll of 10 times as many precincts as the one in Nevada.
The exit polls have also been criticized for other issues measuring Hispanic voters, like whether pollsters conduct Spanish-language interviews. They generally do not.
In a December interview, Joe Lenski, who runs the exit polls for Edison Research, acknowledged that exit polls “are not the precise measurement of demographics that people would want them to be.” He said that he “cringes” when people say, “Oh, that’s the number,” and run with it as “absolute truths or precise numbers.”
Referring specifically to the Nevada results, Lenski added that there are additional errors with caucus and entrance polls. The entrance poll measures initial preferences — those can change at the caucus, and uncommitted is an option that received 3 percent support in initial preference. The delegate tallies don’t necessarily line up with actual vote counts, either; the delegates are allocated only after candidates get 15 percent of the vote. In other words, a candidate could win 60 percent of the delegates without winning 60 percent of the vote.
There were signs of problems Saturday in poll results themselves. Clinton and Sanders were tied in the initial entrance/exit poll estimates; she ultimately won by around 5.5 points. Clearly the entrance polls missed Clinton’s strength somewhere.
Is it possible that the precincts missed Clinton’s strength among Hispanic voters in Clark County? Quite possibly. Even the final entrance/exit poll numbers show Clinton leading in Clark County by just 5 points; she actually won there by 10 points.
A Clinton Win, but Not a Landslide
It’s tough to give a poll subsample of some 200 respondents much weight, especially when it’s for a clustered group like Hispanics.
It’s even harder to give it credit when there’s so much reason to wonder whether it’s right. Clinton fared well in majority Hispanic precincts. National polls show Clinton faring well among Hispanic voters — and Sanders basically finished in line with national polls among both white and black voters.
All considered, I’d err on the side of a Clinton win among Hispanic voters.
But it would be hard to argue that she won Hispanic voters by a lot. Sixty percentage points could easily be too high if she indeed fared better among Hispanic voters in non-Hispanic areas. The entrance/exit poll result may not be perfect, but it certainly seems consistent with this possibility. If I had to bet, I’d say she won Hispanic voters by a somewhat more modest margin.
It would still imply that Sanders made big gains over the last month. But it would still be short of what he would need to overcome Clinton’s vast strength among black voters. She won them by a huge margin, whether you look at the entrance/exit polling or majority black precincts.
• • •
In assessing delegate counts, the estimates used are not perfect. They were based by matching — visually — the precinct results reported by the Nevada Democratic Party to estimates of the racial composition of registered Democrats from L2, a nonpartisan voter file vendor. Even if the matching effort was perfect, the estimates of the racial composition of registered Democrats would not be: They are based on the racial composition of census tracts and the surnames of voters. The data is often used for analysis of Latino voting trends, including by the firm Latino Decisions. The L2 data shows that 23 percent of registered Democrats in Nevada are Hispanic, a hair above the exit poll estimates for the caucus vote.