Thursday, January 05, 2017
By Roy Fuchs
Doug Schwartz, Quinnipiac University poll director, told the Ys Men of Westport/Weston today that though few expected the presidential election’s outcome, “the polls were not as bad as is commonly assumed.”
The “quality” polls, those done by the major television networks, by Pew Research and Quinnipiac, projected a 4-point popular vote victory by Hillary Clinton. She won by 2. “If you’re off by two or three points on the margin, that’s not bad,” he said.
The biggest surprises—Michigan, Wisconsin and Pennsylvania—is where the pollsters missed. Fewer than 80,000 votes — less than 1-point—decided these states. Had their 46 Electoral College votes gone to Clinton, she would be president-elect.
Polls in those states all tightened at the end. In Pennsylvania, for example, they showed Clinton in the lead, but none by wide margins.
Quinnipiac released its final poll one week before the election, one showing Clinton up by 5-points. She lost by less than 1-point, essentially a tie. Unfortunately they were on the wrong side, Schwartz said.
In Florida, Quinnipiac’s final poll showed Clinton winning by 1-point. Trump won by 1-point. Again they were on the wrong side. But Quinnipiac noted in its releases that the race was too close to call, Schwartz said. Unfortunately, the media ran with their numbers, but not their caveat.
What did they miss? Schwartz’s answer: late deciders in the swing states — about 10 percent of the voters — fell overwhelmingly for Trump. Had they voted more like the final outcome, again Clinton would be president elect. Was it Comey? Or? “Polling data doesn’t answer the question,” he said.
Asked during Q&A whether they missed a “certain kind of voter?” Schwartz noted that the white working class, a key Trump’s victory, was not under sampled. Rather, they missed on the extent of their vote for Trump. “He just did better in that segment,” he said.
Perhaps one explanation is the “social desirability bias.” Some people did not want to admit they were voting for Trump even though they had made up their minds, Schwartz said.
Likewise, Hispanic voters, who were projected to go almost entirely for Clinton, but went for Trump more than they had for Romney.
Similarly, African-American voters, though in the other direction. They were correctly sampled, but voted in smaller numbers than in the two previous elections.
Schwartz noted that poll consumers often expect pinpoint accuracy. Not so. Polls have a built in sampling error, a statistical problem that makes them less predictive the closer the projected outcome, he said.
Posted 01/05/17 at 06:25 PM Permalink