Wednesday was a pretty good day for Donald Trump in the polls. A Quinnipiac poll of Florida, Ohio, and Pennsylvania, released in early morning to help shape the news cycle, showed the Republican slightly ahead of Hillary Clinton in Florida, tied in Ohio, and slightly ahead in Pennsylvania. Given that Mitt Romney lost all three key swing states in 2012, and that recent polls have shown Trump trailing in them as well, it was good news for the Republican.
Polls released later Wednesday, from NBC/Marist and Fox, showed Trump behind Clinton in Colorado and Virginia — two more states Romney lost. Like Quinnipiac, the NBC/Marist poll showed Trump and Clinton tied in Ohio, but in Pennsylvania, NBC/Marist had Clinton up by a significant margin — a decidedly different result than Quinnipiac’s.
But the Quinnipiac poll, focusing on the magic formula of Florida, Ohio, and Pennsylvania — the three states that would give Trump the presidency, if he won them on top of the states Romney won in 2012 — attracted the most attention. And almost immediately, some on the left and right — anti-Trumpers all — began attacking the polls and the pollster.
“Here’s why Quinnipiac’s polls showing Trump leading Clinton are totally wrong,” read the headline at the lefty website PoliticusUSA.
“Shoddy Q poll strikes again, and media lap it up,” read the headline at the lefty Blue Nation Review.
“If you believe Trump is leading Clinton by 6 in Pennsylvania, Quinnipiac University has some real estate in Arizona for sale,” tweeted Leon Wolf, managing editor of the conservative RedState, referring to “this joke Q-poll.”
None, left or right, could accept the possibility that at this moment Trump and Clinton might be tied in the three electorally critical states, or that Trump might be slightly ahead in one or two. So the unskewing began.
The lefties objected to the racial mix in Quinnipiac’s samples. The polls counted too many white people and not enough non-white people, PoliticusUSA and Blue Nation Review said. “Since their polling is built on flawed assumptions that are not going to happen on Election Day, no one should take Quinnipiac’s pro-Republican polls seriously,” wrote PoliticusUSA founder Jason Easley, charging that Quinnipiac is “rigging their polls for the best possible Trump outcome.”
Easley said whites made up 70 percent of the Florida Q-poll sample, while they made up just 67 percent of the 2012 electorate in the state. Easley said whites made up 83 percent of the Ohio Q-poll sample, while they were 79 percent of the 2012 electorate. Easley said that whites made up 83 percent of the Pennsylvania Q-poll sample, while they were 78 percent of the 2012 electorate.
PoliticusUSA’s numbers were incorrect for all three states. “The numbers cited in the PoliticusUSA piece are wrong,” wrote Doug Schwartz, director of the Quinnipiac poll, in an email exchange. “In Florida, our sample was 63 percent white. They incorrectly said our Florida sample was 70 percent white. In Ohio, our sample was 82 percent white. They incorrectly said our sample in Ohio was 83 percent white. In Pennsylvania, our sample was 80 percent white. They incorrectly said our sample in Pennsylvania was 83 percent white.” (By the way, Quinnipiac published the correct numbers right at the bottom of its polls, for all to see.)
Blue Nation Review cited the same wrong numbers as PoliticusUSA. So much for that unskewing.
On the anti-Trump right, the objection to the Quinnipiac poll had to do with the sample’s partisan mix in Pennsylvania. “In 2012, the PA electorate was 45D/35R/20I,” Wolf tweeted. “This joke Q-poll has it 35D/34R/25I.”
“The essential problem with these Q-polls is that they assume an electorate that looks like 2004 from a party ID standpoint,” Wolf continued. “Meanwhile, all the actual evidence shows that Republican party ID has dropped hugely relative to Dem party ID since 2004.”
That’s a highly debatable proposition. Consider this, from the New York Times’ Thomas Edsall two months ago:
Take three battleground states: Ohio, Pennsylvania and Florida. In 2008, Gallup found partisan advantages favoring Democrats in these states by 18, 16 and 9 points respectively. But in 2015, the most recent year for which Gallup state data is available, the Democratic advantages had virtually disappeared in two of the three states, falling to 0.2 points in Ohio and to 0.7 points in Florida. In Pennsylvania, the 16 point Democratic advantage fell to 3 points.
I asked Doug Schwartz about Wolf’s objection, and about partisan ID. “We have been using the same methodology for more than 20 years and have an outstanding record for accuracy,” Schwartz responded via email. “In the 2012 presidential election in Pennsylvania, the Quinnipiac University Poll came within one percentage point of the actual result. One example that highlights the folly of relying on past exit polls to predict the future is the partisan shift from 2004 to 2008. In 2004, the national exit poll showed the electorate was evenly divided between Democrats and Republicans. In 2008, Democrats had a 7 point advantage.”
The bottom line is, one could make a perfectly arguable case that the Quinnipiac poll is either right, or well within the ballpark of plausibility, about the actual situation right now in Florida, Ohio, and Pennsylvania. So why the strident denunciations?
In 2012, the poll unskewers, Republicans in that case, let their longings get in the way of their judgment. That led them into what Nate Silver calls “The Valley of Unskewed Polls.” There’s no reason to believe this year’s poll unskewers are any different.
The odd part this year is that, overall, the polls favor the unskewers’ preferred outcome: Trump’s defeat. The about-to-be-nominated Republican is trailing in RealClearPolitics averages of polls nationally and in a bunch of swing states. If the election were held today, Trump would lose. Of course, that could change — that is what campaigns are for. But whatever happens, the urge to deny polls that disagree with one’s political views remains strong.
