Donald Trump found a new talking point after his Nevada win - but it has some big flaws
Trump touted one subgroup in particular in his victory speech: The real-estate magnate won the support of 45% of Latino caucus-goers, easily beating Marco Rubio's 27% support among that group, according to a CNN entrance poll.
"And you know what I'm really happy about? No. 1 with Hispanics," he said. "I'm really happy about that."
But there are some pretty major caveats to this result. A new poll out Wednesday found that Trump is viewed most negatively by Hispanic voters of any Republican hopeful.
And second, entrance polling is based on taking a small sample of caucus-goers, ideally at random, and asking them about their demographic identities, political affiliations, and - most importantly - who they support in the caucus.
Entrance and exit polls tend to do a decent job of giving us an overall picture of who participated in a caucus or an election. And they can sometimes (but not always) give an early clue as to the results of that caucus or election.
But they also often run into difficulties in discerning the behavior of smaller subgroups.
The CNN poll had a total of 1,573 respondents. Of those, only 8% self-identified as Latino. (This compares to 24% of voting-age Nevadans overall who the Census Bureau identified as Latino or Hispanic.)
This gives us a sub-sample of around 126 Latinos. Since 45% of this sub-sample voted for Trump, according to the poll, the end result is about 57 Latino Trump voters.
Small samples bring with them statistical uncertainty. An extremely conservative estimate for a margin of error for our estimate gives us around plus-or-minus nine percentage points.
Assuming that everything went perfectly with getting a representative sub-sample of Latino voters in the entrance poll, Trump's actual level of support among Latino Republican caucusgoers in Nevada could have been as low as 36% - or as high as 54%.
This level of uncertainty happens whenever we look at small samples, and is one of the major limitations of entrance and exit polls. Drawing sweeping conclusions about populations based on tiny sub-samples is a potentially dangerous game.
Aside from issues of uncertainty when looking at statistics with small sub-samples, it's important to bear in mind that Trump won a plurality of Nevadans who actually went to a Republican caucus. Given that overall turnout for the Republican caucuses was about 75,000 people, this is a very small and likely non-representative fraction of Nevada's population.
The general election in November will certainly have a much larger electorate that will very likely look quite different from the subset of Nevadans who attended a caucus.
Tuesday's results, then, have very little predictive power about how Latinos, or any other group of voters in the state, will vote this fall.