Update on study of alternatives to exit polling

With a $250,000 grant from the Knight Foundation, AP hired GfK Custom Research to conduct experiments in Florida’s Democratic and Republican primaries as part of its efforts to explore alternatives to exit polling.

David Pace, news editor for race calls and special projects, explained the results:

National Election Day is looking more and more like national election month. Voters who cast their ballots before Election Day accounted for 35 percent of the total vote in the last presidential election. And in 35 states, the percentage of absentee and early voters exceeded 20 percent of all votes cast. The increasing numbers of pre-Election Day voters over the past decade is one of the biggest challenges facing the news media in conducting election polls to help call winners and explain the demographics and voter attitudes behind the numbers. Simply put. The 45 million voters who didn’t go to the polls in 2012 weren’t available to exit poll interviewers. AP and the networks that make up the National Election Pool had to resort to costly telephone polls to estimate the impact of those non-Election Day voters. And telephone polls are facing huge methodological challenges as residential phone lines disappear and potential voters use caller ID to screen out calls from pollsters.

In response to these challenges, AP commissioned an experiment to see if an online survey can reach both early voters and Election Day voters at the same time. For survey participants, GfK used the Florida residents on its probability-based, 55,000-member national online panel. And they added a larger nonprobability-based sample of Florida residents, using known differences between the two groups to blend them into one sample for each party. The results were intriguing. The average error in the vote proportion estimates for each of the five Republican and two Democratic candidates was 1 percentage point, both for the probability-only sample and for the combined probability and nonprobability sample. In contrast, the average error on the combined exit poll and telephone estimates of the same seven candidates was 2.2 percentage points. The experiment also compared the online estimates of the demographics and attitudes of each party’s primary electorate with the exit poll/telephone poll findings. The question-by-question comparisons of the web survey results to the exit poll version released at 5 p.m. Election Day can be found here. The comparisons to the final version of each party’s exit poll can be found here. GfK’s full report on the Florida experiment can be found here.
It should be noted that results from these experiments are not being used in AP’s election coverage or race-calling decisions this year. Instead, AP hopes the experiments will show the way to more accurate and less expensive survey methods for the 2017-2020 election cycle.