Washington – Quinnipiac University’s polls have come under fire recently, especially from liberals supporting Hillary Clinton’s bid for the White House, for results that differ from those in other presidential polls this year.
The reason why is unclear, but several of the university’s polls have shown more support for GOP candidate Donald Trump than other polls have. Drawing particular scrutiny have been polls in crucial “swing states,” and a poll that measured Trump support among Hispanics.
The problem, some critics say, is that Quinnipiac has over-represented white voters, who are more likely to vote Republican than minorities.
Quinnipiac poll Director Douglas Schwartz defends his methodology, which he described as the “gold standard” in the field. He “categorically disagrees” that there are too many white respondents in Quinnipiac’s polls.
Polls use detailed methodologies to select a random sample of voters to call, and then weight that sample to match the demographic makeup of the population.
Quinnipiac, for example, says it uses census data to weight its sample to match the population by region, gender, age, education and race. But it does not weight by party affiliation because, Schwartz said “that something that can change.”
“Whatever we get, we get,” he said.
As far as geography, race, gender and other demographics, Schwartz said, “If you do a good random sample, you don’t have to weight your data a lot.”
Quinnipiac made headlines with a May 10 survey showing that Clinton was in trouble in three key swing states. It showed Clinton at 43 percent with 42 percent for Trump in Florida, Clinton at 39 percent, with 43 percent for Trump in Ohio and Clinton at 43 percent, with 42 percent for Trump in Pennsylvania, which meant the race was neck-and-neck when the poll’s margin of error was factored in.
“Quinnipiac University Swing State Poll Is First to Call Clinton-Trump Race Close,” the university said.
But other polls found less Trump support in those swing states.
Jason Easley, publisher and editor-in-chief of PoliticusUSA, a liberal news and opininon web site, blasted Quinnipiac’s methodology, saying it over-represented whites. He said the results would be different if Quinnipiac used actual turnout data from the 2012 election, based on exit polling, to adjust its sample.
“Any Republican is going to win a presidential election if white turnout increases and minority turnout drops,” Easley wrote.
“The reality of turnout statistics is that the opposite has been happening. White turnout has been decreasing while non-white turnout is growing rapidly,” he said in a July 13 posting. ”Quinnipiac is projecting an electorate that is nothing like who are likely to vote in November. “
Quinnipiac’s Schwartz said boosting the number of minority respondents based on exit polls from the 2012 election would be fraught with danger.
“If there was ever an election to be skeptical about assumptions based on exit polls from previous elections, it’s this one,” Schwartz wrote in a recent blog posting. “This presidential election already has defied more predictions than any in modern political history. There are so many ‘firsts’ that could affect turnout, nobody knows what will happen.”
Quinnipiac uses live interviewers to survey people they reach through random calling on both landlines and cell phones.
“We’ve been using the same methodology for over 20 years and have an outstanding record and outcomes,” Schwartz said in an interview.
Yet the New York Times on Monday said, “For whatever reason, the Quinnipiac polls consistently show Mr. Trump doing better than NBC/Marist polls do, even though they appear fundamentally similar.”
Lee Miringoff, director of the Marist Institute for Public Opinion, said he did not know why his polls and Quinnipiac’s differ.
“Interestingly, Quinnipiac and us both use live interviews and both landlines and cells,” he said. “But there may be a difference in the number of landlines and cells contacted.”
Minorities are more likely to use cell phones exclusively, and reaching more respondents by cell phones results in more minorities – who trend Democratic – in the sample.
In the most recent national poll of the presidential election, Marist pollsters reached 44 percent of their respondents by cell phone, while Quinnipiac pollsters reached 39 percent by cell phone.
Sixty five percent of the Marist poll respondents were white, 12 percent were black and 13 percent were Latino.
Meanwhile, Quinnipiac’s sample was 73 percent white, 11 percent black and 8 percent Latino.
Miringoff said he had a “hunch” cell phones had something to do with the differences in their poll results, resulting in a sampling group that is more Republican and less Democratic.
But he said there are other “variables,” including the quality of the questions asked and the competence of the interviewers, that can affect a poll’s results.
“You get different gas mileage from different cars,” he said.
Schwartz emphasized that his samples are representative of the demographics of the population.
Marist’s latest national poll on the presidential horse race, released on July 15 and covering a survey period of July 5 through July 9, had similar results to Quinnipiac’s latest national poll, released June 29 and covering the period June 21-27.
The McClatchy-Marist poll had Clinton leading Trump 42-39 percent, and the Quinnipiac poll showed Clinton leading 42 to 40 percent, but it was taken before U.S. Attorney General James Comey blasted Clinton for using a private email server to receive classified information.
Swing states stunners
Another difference with other pollsters is that Quinnipiac has not yet restricted its polling to “likely voters” in the Presidential race, although others, such as the Mason-Dixon poll have, because they say that leads to more accuracy. Quinnipiac says it will start polling likely voters in August or September, closer to the election.
“When you are within four to six months of an election, it’s a lot easier to ask people whether they are going to vote, said Brad Coker, managing director of the Mason-Dixon Polling & Research.
Many of the differences between the polls released by Quinnipiac and other pollsters center on surveys of voters in “swing states” that will determine the outcome of this year’s presidential contest.
Schwartz defends the number of minorities in Qunnipiac’s samples, but made a downward adjustment in June on how many white respondents it included in its samples from Pennsylvania, from 83 percent to 80 percent, and Florida, from 67 percent of respondents to 63 percent.
Schwartz said the change was needed to fix a “sampling error.” The percentage of white respondents in Ohio was kept the same, at 82 percent.
After the adjustment, Quinnipiac released another poll in those swing states that showed Clinton had “inched up” on Trump, especially in Florida, where she led Trump by 47 to 39 percent.
The poll showed Clinton tied with Trump in Ohio and holding a slim 42-41 percent lead over Trump in Pennsylvania.
Qunnipiac’s latest swing-state poll, released on July 13 and conducted in part after Comey scolded Clinton over the use of a private email server, showed Clinton losing ground again.
Perhaps the poll that raised the most eyebrows was one Quinnipiac released at the end of June that showed Trump had the support of 33 percent of the Hispanics surveyed.
Before he was fired, former Trump campaign manager Corey Lewandowski touted the Quinnipiac poll as proof of the billionaire mogul’s success among Latinos.
“Look at Q. He’s got 33 … 33 percent, he’s winning with Hispanics, which if you look at what Mitt Romney received or John McCain received, he’s way above where they were. Way above,” Lewandowski said.
But all other major polls, including an NBC-Wall Street Journal-Telemundo survey released this week, showed Trump having much less Latino support.
The NBC-WSJ poll determined Trump had the support of only about 14 percent of the nation’s Latinos. It surveyed 300 Hispanic registered voters, weighted the sample by ethnic origin – that is whether they were Puerto Rican, Cuban, Mexican or of some other ethnicity – to match national demographics, made sure there were equal numbers of male and female respondents and reached more than half of those surveyed by cell phone.
Meanwhile, Quinnipiac’s poll surveyed 103 Latinos as part of a national poll of 1,610 respondents and did not screen the respondents as the NBC-WSJ pollsters did as far as national origin and gender. The Quinnipiac poll had a 10-point margin of error for the Hispanic subgroup while the NBC-WSJ poll had a 5.6 percent margin of error.
“I would say the best way to look at our Hispanic poll is with caution,” said Schwartz, referring to the poll’s large margin of error. “Clearly (Trump) is way behind with Hispanics.”