In case you didn’t know, I have a thing for Pennsylvania. So when POLITICO dropped this story about the Democrats’ polling problems with rural whites, I returned to my recent collaborative presentation with Varad Mehta and Miles Coleman, which explored how the Presidential Primary in 2016 had offered a clear warning to Clinton, had her staff bothered to track their results state by state. Before I dive into our own work, let me give you a telling bit from POLITICO:
The same strategist added that many of these voters also may choose not to participate in polls “because they don’t like the establishment and they don’t want to take a survey.”
The yawning education gap among white voters’ preferences — Trump clobbered Clinton among white voters without a college degree, while the two ran neck-and-neck among those with a degree — means that nonresponse bias may have been determinative, said Democratic pollster Nick Gourevitch, a partner at Global Strategy Group. And it may have been going on for some time.
“I think it’s very plausible that for years pollsters have been over-representing educated voters, and that it only came back to bite us recently because it was a key driver in vote preferences this time,” Gourevitch said.
Now, back to our recent presentation at Saint Anselm. For Pennsylvania, we broke down the state into it’s various municipal divisions (in Philadelphia county, its sixty-six wards). Using 5-year estimates from the American Community Survey, we were able to then identify wards and municipalities whose adult populations were majority-whites-without-college-degrees, plurality whites-without-college degrees, majority whites-with-college-degrees, plurality whites-with-college-degrees, majority-nonwhite, and plurality non-white (note: when you get this granular, ACS error rates grow- after all, they’re estimates, and we’ve zoomed way down from statewide and county- but in general, the ACS data and the cycle-to-cycle election changes line up well). When we sorted them out, here’s how the Keystone State’s general election vote shook out:
- 71.83% of ballots cast in the Presidential election were from places where whites without a college degree were a plurality (13.09%) or outright majority (58.74%)
- 12.28% of ballots cast were from places where whites with a college degree were a plurality (6.46%) or outright majority (5.82%)
- 15.88% of ballots cast were from places where African-Americans, Hispanics, and Asians make up a plurality (5.8%) or outright majority (10.08%)
Areas dominated by one demographic aren’t monolithic. Using ACS estimates and applying them to each municipality and ward’s vote would yield the following breakdown of ballots, assuming that the voters reflected the adult population:
- 25% of ballots cast by African-Americans, Hispanics, Asians, and other non-white voters
- 23% of ballots cast by whites with a college degree
- 52% of ballots cast by whites without a college degree
Voters with college degrees are more inclined to vote than voters without, so these numbers are probably off by a bit. But turnout rates aren’t dramatic enough to blunt the enormous share that whites without a degree make up.
Public polling was sampling an electorate divorced from reality. Morning Call’s final poll, which found Clinton leading Trump by six points, surveyed an electorate almost half-filled with college graduates (48%, Q 18). Franklin and Marshall, which had the biggest miss of the pollsters releasing within 10 days of the election, surveyed the same percentage, though it’s unclear if that number is the unweighted or weighted sample. The only pollster to whiff Pennsylvania even worse was Marist, conducting surveys for NBC and the Wall Street Journal. Digging through their likely voter sample, only 44% of those surveyed were whites without a college degree. Marist, unsurprisingly, found Clinton up by twelve in early October, which would make sense in a scenario where whites with degrees made up 37% of electorate.
Having detailed results from the election would do both parties good: county level tallies simply don’t cut it, and exit surveys have problems. Deeper analysis can be time-consuming and costly, but the further down you drill, with careful demographic studies, the easier it is to identify your problems and improve targeting. In a state like Pennsylvania, counties mask serious divisions: in Bucks County, which overall barely moved between 2012 and 2014, white working class areas like Bensalem and Levittown shifted hard towards the Republicans, while post-grad clogged communities in the middle of the county moved towards Clinton.
Going back to dividing up results by municipality demo type, you can see how Nate Cohn’s post-cycle analysis published today in the New York Times holds up strongly. Donald Trump won Pennsylvania, and the Midwest, by winning over a significant number of white working class voters that backed Barack Obama.
- In municipalities where African-Americans, Hispanics, Asians and other non-whites are an outright majority of the adult population, then-Senator Barack Obama won 88.75% of the two party vote against Senator John McCain. Hillary Clinton won 89.44% of the two party vote against Trump.
- In areas where racial minorities are a plurality, Obama won 72.31% of the two party vote in 2008, Clinton earned 71.57%
- In places where whites with a college degree are a majority, Obama won 56.4%, Clinton 61.08%
- Where whites with a degree are a plurality, Obama won 55.53%, Clinton 58.13%
- Even in municipalities and wards where whites without a degree are a plurality, Obama won 54.54%, and Clinton carried them too, with 54.08% of the vote.
With these sorts of numbers, Clinton needed to totally collapse among the last category to fall so short of Obama’s 2008 tallies, and collapse she did. In the municipalities and wards where whites without a degree are an outright majority, Obama lost, but still earned 47.17% of their vote. Eight years later, Donald Trump would beat Hillary Clinton in these communities by a 62.28% to 37.72% margin.
Months before the general election, the Pennsylvania primary saw Clinton shed 455,000 primary voters in white working class communities and wards. They didn’t vote for Sanders- his numbers in these communities were smaller than Obama’s in 2008- they stayed home. Six months later, a large number of them had been persuaded by Trump, and pull the lever for the Republican. Many of these very same voters came home to Obama in 2008 after backing Clinton, and stuck with him in 2012. These voters aren’t committed Republicans: one hundred and fifty communities and wards that voted for Trump happily voted for all three of the Democrats running for statewide office. These voters are very much up for grabs, and polling firms and political parties will do well to reach them.