A study recently accepted for publication in a peer-reviewed journal makes a startling suggestion about so-called "robo-polls" in the 2012 Republican presidential primaries, raising the question of whether these automated surveys may have been adjusted to match live-interviewer polls.
The paper, produced last spring, says its findings mean that aggregated results of pre-election polls - the "poll of polls" summaries popular on some political news websites - may be "misleadingly precise," if many polls included in their averages were manipulated to match preferred estimates.
The authors, Joshua Clinton, a political science professor at Vanderbilt University, and Steve Rodgers, a Ph.D. candidate at Princeton University, note that their results "are necessarily suggestive rather than definitive" - indicating the possibility that robo-polls, in which phone calls are automatically dialed with questions posed by a recorded voice, were adjusted to match previously released traditional polls, not proving it. They also say that given a lack of data they did not test the reverse possibility, that traditional polls were altered to match automated ones.
Previous reports have found pre-election horse race estimates from robo-polls (also known as IVR polls, for "interactive voice response") to be as accurate as those obtained using traditional polling methods. Clinton and Rodgers confirmed that, in part: Examining 159 pre-election results, they report that robo-polls were as accurate as traditional surveys when a traditional survey in the contest was released first - but were less accurate when that was not the case.
"There is no difference in the accuracy of IVR polls and human polls when IVR polls occur after a human poll, but IVR polls do significantly worse if human polls are not conducted first," they report. "The apparent equivalence of IVR polls and human polls in the 2012 Republican primary appears to depend on human polls being conducted prior to the IVR polls."
They say this outcome, obtained through a series of statistical analyses, "suggests, but certainly does not prove, that at least some IVR polls may use earlier human polls to adjust their results to ensure that they are not notably different from existing polls and beliefs."
Robo-polls are automated telephone surveys in which recorded instructions direct participants to press numbers on their telephone keypads in response to polling questions. They depart from traditional, live-interviewer polling techniques in a variety of ways, and generally make only landline telephone calls, despite the large number of adults reachable only by cell phone. ABC News policy is not to report robo-poll surveys.
The Clinton/Rodgers study has limitations. It's based on a small sample - of 106 IVR polls included in the study, just 17, in a total of five states, were produced before live-interviewer surveys in the same contest. "This is why we're a little tentative in what we're doing," Clinton told me. "It's possible that what's going on is something goofy in those five states."
Nonetheless, the results are statistically significant, and per Clinton the paper has been accepted for publication as a research note in the April edition of "P.S.: Political Science and Politics," a peer-reviewed journal of the American Political Science Association. More's on the way: Clinton says he and Rodgers are in the midst of repeating their analysis on a much larger dataset, including pre-election polls from 2008 and 2010 alike.