AP file |
| Undecided voters may not actually be all that undecided, scientists say. |
Scientists say a five-minute computer test could help pollsters figure out which way undecided voters will go, even before the voters themselves know.
The test got a successful tryout in a study discussed this week in the journal Science, and one of the researchers behind the experiment said similar studies will likely be conducted during this fall's presidential campaign.
"What our findings show is that these measures of automatic mental association have the potential to perhaps improve the prediction of election outcomes," said Bertram Gawronski, a psychologist at the University of Western Ontario.
Figuring out how the undecideds will decide - and getting the right kind of undecideds to the polls - is of paramount concern to political campaigns. In this week's NBC/Wall Street Journal poll, 13 percent of the survey respondents said they weren't siding with either Barack Obama or John McCain. (The typical responses were "not sure," "voting for neither" or "it depends.")
Come November, those undecided voters will spell the difference between victory and defeat. That's why the decision process for undecideds is of more than academic interest.
To learn more about that process, Gawronski and two colleagues from Italy's University of Padua, Silvia Galdi and Luciano Arcuni, used the same kind of test that other researchers have employed to find evidence of hidden racial bias. The idea is that people have subtle information-processing filters that lead them in directions they can't even explain to themselves. Implicit-association tests attempt to get at those unacknowledged filters.
The newly reported research looked at 129 local residents' attitudes toward the expansion of a U.S. military base in Vicenza, Italy. Each subject sat down at a computer and watched as a series of pictures and words were flashed on the screen. For one part of the test, they were told to press a particular key if a positive word (say, "joy") or a picture of the military base appeared. Then, they were told to press a different key if a negative word ("awful," for example) or a picture of the base appeared.
Researchers figured that the subjects would do better at the task if the pictures meshed with words that conveyed the feeling they had toward the base. For example, people who subconsciously didn't like the military base would be quicker to associate the pictures with negative words, even if they hadn't consciously made up their mind. The opposite pattern would hold true for people who were predisposed to favor base expansion.
"This is based on earlier research in our lab and from other people that automatic mental associations have the potential to distort reality," Gawronski said. "It's this distorted interpretation of information that has the potential to affect future decisions."
The first round of interviews and tests turned up 30 undecideds. Then all the subjects were given background materials on the base expansion plans, pro and con, and asked to come back a week later for a second round.
The researchers found that the undecideds' computer test results in the first round were a strong predictor for the way they eventually voted in the second round - the prediction success rate was 70 percent. But the test wasn't as good for predicting vote shifts among the first round's decided voters.
"One should not make the mistake of thinking of these measures as a replacement for the standard methodology," Gawronski told me. "This didn't do anything for decided voters. It's more of an addition, and this addition is particularly useful for figuring out where the undecided voters might go."
The differences in the computerized responses are not huge - they're on the order of 200 milliseconds. But that's enough of a difference to predict the probabilities for future behavior, Gawronski said. Currently, it takes five minutes to administer the test on a computer, and that time could be trimmed down to as little as two minutes, he said.
"A lot of researchers are actually doing studies with these measures over the Internet," Gawronski said. "So if you recruit a representative sample in the population, and in addition to your phone interviews, you direct people to a Web site, that's certainly feasible to do."
There are a few details yet to be worked out: In order to predict the election outcome, you'd need to develop a scale that would tell you the percentages of undecideds who ultimately vote one way or the other, as well as those who decide not to vote at all. The current research isn't fine-tuned enough to give you those proportions.
"You can compare it to a thermometer that gives you higher or lower numbers, but at this point you can't match it to a point where the water is freezing or boiling," Gawronski said.
For now, a good political operative could still probably do better than a test-wielding psychologist at predicting how the undecideds will decide. But that could change in the months and years ahead as other researchers follow up on the Italian study.
"My assumption is that a lot of my colleagues in the United States may actually be collecting data right now for the upcoming election, so we might see that this is an effect that also emerges in North America," Gawronski said. "In a couple of months, we will know more."
Update for 12:20 p.m. ET Aug. 22: The Los Angeles Times' report on the study notes that Virginia-based TargetPoint Consulting experimented with the implicit-association test during the Republican presidential primary campaign, and that a research team from the University of Virginia, the University of Washington and Harvard University is offering an Obama-McCain test you can take yourself (along with other tests on racial / religion / age bias, attitudes toward career and education, and more).
For more about the study, click on over to the University of Western Ontario news release or listen to Science's podcast. For more about the political road ahead, check out msnbc.com's Politics section, and don't miss our Gut Check coverage focusing on battleground economics.
