With only six weeks left before Election Day, it's open season for polling. Eager pollsters are dialing away, trying to get people on the line who will tell them which candidate they are likely to vote for. But what about the people who don't answer their phones? Or those who say "no thanks"--or worse--and hang up? Is there a certain type of person likely to refuse the probing calls? And does that affect polling numbers?
In a January op-ed in The New York Times, Andrew Kohut, president of the Pew Research Center, warned that the disparity between polls and the outcome in the New Hampshire Democratic primary--Clinton beat Obama despite polls showing him with an advantageous margin--could have been due, in part, to the fact that less affluent whites are more likely to hang up on pollsters. "These whites who do not respond to surveys tend to have more unfavorable views of blacks than respondents who do the interviews," Kohut wrote.
So with the media now reviving questions about how race affects voters' decision-making, should we be worried that polling numbers under-represent racists (or anyone else, for that matter)? Several pollsters I spoke to this week said there isn't a notable disparity between the types of people who answer questions and those who do not. John Zogby, president and CEO of Zogby International, told me that over the past few decades, there has been a "democratization of refusals" and that there is a 95-percent confidence rate in polls' accuracy. "In the early days--this was before answering machines and all of the other screening devices--the refusal... used to be the purview of the affluent, and now the blocked call that comes in just runs across the board demographically," Zogby said. Similarly, Charles Franklin, co-founder of Pollster.com, said there isn't a notable partisan divide between people who agree to polls and those who refuse. "If every Republican hung up and every Democrat did an interview, the profession would be in a crisis," he said.
Scott Keeter, director of survey research at Pew, added that while there was evidence a decade ago that people with racially conservative views were likely to put their receivers down, that trend has diminished. "In 1997, the first time we did [an experiment], we did find a small--I emphasize small--degree of potential bias in under-representing racial conservatism, but we are talking about a percentage point or two, barely statistically significant," he said. "This problem, if it was a serious problem in the 1980s or early 1990s, is not a problem any longer." He said the inaccurate polling in New Hampshire was probably the result "a collection of intense political forces unlikely to be repeated anytime soon," ranging from Obama's victory in Iowa to Clinton's infamous misty eyes just before the New Hampshire vote.
Still, response rates overall have declined from more than 50 percent 40 years ago to often under 25 percent today. Pollsters have had to adjust their techniques to deal with this higher rate of refusals, which is due to the rise of call screening and the burgeoning telemarketing sector. (The latter has boosted the likelihood that people just don't want to talk to random callers.) To adapt, survey organizations have expanded the number of people they call per survey, and they often call numbers with no answers or hang-ups as many as a dozen times to try to get responses. Moreover, Kathy Frankovic, director of surveys for CBS News, said that before conducting their final analyses, pollsters assess and adjust their pool of results to make sure too much weight isn't given to those people more likely to be polled. (For example, people who have more than one phone number in their homes--and thus stand a higher chance of being randomly selected for a call--are likely affluent.) Pollsters also select a sample to analyze that reflects the nation's adult population demographically and geographically.
So, whether refusals are hard (the receiver slam), soft (the polite decline), or forced (the non-answer, or the midway interruption), pollsters agree they are unlikely to affect survey numbers adversely.