By ALAN J. ORTBALS
Right up to Election Day in 2012, Mitt Romney was sure he was going to be victorious based on his internal polling. He lost by 5 points.
In 2014 polls showed the question of Scottish independence was a tossup. It got clobbered at the polls.
Last month, Bernie Sanders won the Michigan primary, while polls predicted he would lose to Hillary Clinton by 21 percentage points.
Why the incongruity?
“Polling has gotten a lot harder than what it was about 10 years ago,” said Ken Moffett, associate professor of political science at Southern Illinois University Edwardsville.
Moffett points to multiple factors at play. One is that so many people today have caller identification and don’t answer the phone, requiring more calls to make connections. But, even when a pollster connects with a potential respondent, the percentage of those who will take the survey has dropped dramatically.
“It used to be when you contacted people to ask them to participate in a poll you’d get about 30 percent who would agree to participate,” Moffett said. “Now you’re down to about 5 percent or so. I think part of the decline in participation is people being over-surveyed. More companies and organizations are doing more surveys in a variety of ways and you’re getting respondent fatigue.”
Another factor driving down the respondent percentage is the hectic lifestyles of many Americans. Moffett said that there are certain demographics that are particularly hard to survey, one of which is males between the ages of 35 and 50 because when Mom and Dad are both working, picking up the kids from school, ferrying them to soccer or band practice and trying to get supper on the table, about the last thing they want to do is spend five minutes on the phone taking a poll.
The difficulty with reaching people and getting them to take telephone surveys creates a couple of problems for pollsters. One, it’s difficult to tell if your sample is representative of the target group. Two, time is money and the more time it takes to collect a statistically significant sample, the more it costs to do it.
Another issue at play is the growing number of people who have gone to cellphones only, says Moffett. Federal privacy regulations prohibit the kind of automated random-digit dialing that is often done with landline polls. The pollster has to physically dial the number and that makes cellphone-only people harder to reach, thus harder to poll.
“Cellphone-only households do in fact systematically differ from those who have both cellphones and landlines,” Moffett said. “They’re primarily younger and also primarily lower-income levels. A lot of pollsters are pointing to those two things to explain why the polls went the way they did in the Michigan primary election.”
Those problems aren’t going away and may get worse, says Moffett, and pollsters are having to change some of their methodology.
YouGov/Polimetrics and GFK Knowledge Networks are two companies that use Internet-based panels rather than phone calling. Moffett said these companies recruit millions of people to be a part of their databases and survey samples of these people via the Internet. They randomly select respondents from their database who accurately reflect the target population and then entice them to participate with survey incentives. They might pay a respondent $5 for taking a 10-minute survey, for example. Incentives can vary. They may even go so far as to pay for lower-income people for their Internet and provide a computer for them to participate.
With these problems, complications and changing technology, polling companies are having to change on the fly during this election season, said Moffett. As election results come in, the post mortem immediately begins to determine what went wrong with the polling and what needs to be done differently. But, he said, with so many candidates some voters aren’t making up their minds until they mark their ballots.
“And I would say that this election, in terms of how it’s gone to this point, is very much an anomaly compared to recent elections,” Moffett said. “For one thing, on the Republican side, you not only have a higher number of candidates in the primaries than what you would normally get, but you’re also getting different types of candidates, and I’m referring in part here to Donald Trump. But, even on the Democratic side it’s become harder to predict than normal. That could be partly because they are attracting voters who weren’t voters before and there are some political scientists who are in the beginning stages of testing exactly that.”