When Donald Trump won the presidency on Nov. 8, pundits, reporters and political wonks turned to the national exit poll to find out exactly what had happened. How had various groups in various states voted?
One of the exit poll findings was particularly surprising: Although Hillary Clinton triumphed among Latinos overall — with 66% of their votes — she got fewer of their votes than Barack Obama got in 2012. And although Trump lost among Latinos, his share of their votes — 28% — was 1 percentage point higher than Mitt Romney's in 2012.
If those figures were accurate, they would represent a major reversal in Latino voting trends.
Since the 1990s, scholars tracking Latino votes have seen the group's increasing muscle at the polls and that, in general, they've been putting it to work for Democrats. Was the largest ethnic group in the nation suddenly starting to shift its party allegiances? And could that have happened even with a Republican candidate who called immigrants rapists, promised to build a border wall and claimed that a Mexican American judge couldn't be impartial because of his heritage?
As it turns out, a closer analysis in states with large Latino populations indicates that the national exit poll — which is still being cited as an authoritative source — is wrong. In Arizona, Texas and Florida — and in California, where we just finished analyzing the data — the exit poll has substantially overestimated Trump's success among Latinos.
The national exit poll is a state-by-state survey conducted by Edison Research and paid for by news media. It samples selected precincts in an attempt to gather an accurate overall count quickly on election night. But, like all polls, it isn't fail-safe. In California, the Edison exit poll data suggested that 71% of Latinos cast their votes for Clinton, and that 24% voted for Trump. According to our research, the numbers are significantly different: 83% for Clinton, and just 11% for Trump.
Instead of polling a sample of voters, our analysis is based on actual votes: The official 2012 and 2016 results in 10,121 precincts in 16 California counties (the most populous counties and the ones that certify their election results quickly). That data is combined with census information that tells us the demographics of each precinct.
The accompanying graph shows you what we discovered. The dots represent the vote percentage for each candidate in each precinct. The more votes he or she won, the higher the colored dots rise on the vertical axis. Along the horizontal axis, the vote totals are sorted by the size of the Latino population in each precinct. And the colored lines track the overall estimates of precinct vote shares for all four candidates in both elections, arrived at by a separate statistical analysis.
We know that Obama won California by 3 million votes in 2012, and Clinton expanded this margin to 4.3 million in 2016, so we would expect to see Clinton doing better and Trump doing worse than their counterparts in 2012. In fact, that is exactly what the dot pattern shows. Trump underperformed Romney across almost every precinct we analyzed, and the deficit widens as voting precincts get more heavily Latino. Clinton's vote share largely mirrors Obama's except in precincts where more than 75% of the population are adult Latino citizens. In these precincts, she captures a greater proportion of the vote share. (That increase is most likely the reason Clinton won the state by such a large margin over Obama in 2012.)
One key finding from all of our analysis is this: It is virtually impossible that 24% of California Latinos cast their vote for Trump as purported by the exit poll. Instead of winning 71% of Latino votes, Clinton won more than 80%.
The discrepancy between the exit poll and our findings is similar to discrepancies found in precinct and census research in Texas, Arizona and Florida. Taken together, these results suggest that the findings of an "entrance poll" that was conducted in the days before the Nov. 8 election are much closer to reality than the exit poll. The research firm Latino Decisions did the survey. It put the Latino vote for Clinton at 80% in California, and for Trump, 16%. Nationally, 79% voted Democratic, and only 18% Republican.
More state level analysis needs to be done on the voting patterns in the 2016 election. At the very least, we expect scholars and policymakers will ultimately recognize that the trend in Latino voting — more turnout and more votes for Democrats — was buoyed and extended by Clinton's candidacy, not interrupted or reversed.
Voting patterns drive policy and party choices in Washington and in state capitals. And exactly how the vote stacks up in California is particularly significant. The demographics of the Golden State — where Latino voters are an increasingly influential bloc at the polls — are likely to be replicated in much of the rest of the nation in coming years. We know that democracy needs good data. When it comes to the Latino vote, the national exit poll isn't good enough.
Francisco I. Pedraza is assistant professor of political science and public policy at UC Riverside. Bryan Wilcox-Archuleta is a doctoral student in poltical science at UCLA.