The experiments will build on similar online surveys GfK conducted for AP during last year’s midterm elections in Georgia and Illinois. Candidate estimates from the two GfK surveys were significantly more accurate than the exit poll, both among Election Day and absentee voters. In addition, the online survey in Georgia did not require an expensive telephone survey of absentee voters to supplement the exit poll.
The news industry is becoming increasingly receptive to online surveys as polling’s longtime gold standard, the telephone survey, faces more and more methodological challenges. AP started using GfK’s probability-based, online Knowledge Panel for its political surveys two years ago. Other media, including The New York Times, CBS News and NBC News, also have begun using online surveys in some cases. None of the organizations, however, has tried using online panels to replace any of the dozens of telephone polls needed to supplement exit polls in states with high levels of absentee voting.
The AP decided to experiment with online panels in Kentucky and Mississippi to address one potentially serious issue with traditional exit polls: the interaction between the interviewer and the voter. Right or wrong, that interaction has been blamed for the disproportionate refusal by Republican voters to participate in the exit poll. Those GOP refusals have consistently led to a systematic overstatement of the Democratic vote. But what happens if there is no interviewer-voter interaction? The Kentucky and Mississippi experiments will answer that question, as survey participants will be selected from both GfK’s Knowledge Panel and other online volunteer panels. All participants will complete their questionnaires online.
A major downside to online surveys is they cannot guarantee that actual voters are being interviewed. Online participants are screened into the survey’s voter pool based on their answers to a series of questions. Exit poll participants, on the other hand, are asked to fill out questionnaires as they leave their precincts after voting. Of course, telephone polls of absentee voters can’t guarantee that participants are actual voters either, and they make up more than a third of the national exit poll sample. To address this issue, AP has asked the National Opinion Research Center at the University of Chicago to explore the possibility of using GPS tracking on cellphones of online panel participants — with their permission — to verify that they have voted before asking them to participate in election surveys.
Further experiments around these and other issues are planned for next year’s presidential primaries and the fall general election. The AP’s goal is to develop innovative and more accurate ways to identify and survey voters who elect national and state leaders, both to help in calling election winners and in explaining the vote.