Students who saw silent videos picked the right candidate 58 percent of the time, whereas those viewers who heard full sound or muddled sound were only right 52 and 48 percent of the time, respectively, no better than the results of random guessing.
Forget the campaigns. Disregard the position papers and attack ads. One of the best ways to tell who's going to win an election is to see the candidates on TV, watching them for 10 seconds and keeping the sound off. That's how more than 260 Harvard undergraduates, watching gubernatorial candidates in 58 races, compiled a rather impressive record of forecasting elections. They picked the winner an average 58 percent of the time, according to Thin-Slice Forecasts of Gubernatorial Elections (NBER Working Paper No. 12660). The students were more accurate than any economic measure that the paper's co-authors, Daniel Benjamin and Jesse Shapiro, tested. They were far more accurate than the Harvard students who actually heard what the candidates had to say.
If this gut-level, insta-pick method seems disturbing, take heart. At least Americans aren't alone in skin-deep politics. A study of the 1996 presidential race in Romania found that people could predict the outcome of the first round of voting based merely on photographs and video clips of the candidates. A study last year of Finnish elections found that ratings of candidates' physical attractiveness predicted their success better than ratings of their competence.
Following these foreign and other similar U.S.-based studies, the NBER paper offers several new insights. It quantifies how much of a role personal appeal plays in relation to other economic and political factors. It tries to single out the quality behind such appeal. It suggests, strikingly, that theless one hears a candidate, the better one can assess his or her chances of winning. That last finding "may help to explain why expert forecasters, who are highly informed about and attentive to policy matters, are often found to perform no better than chance in predicting elections," the authors write.
The thrust of the NBER research was to ensure that the subjects knew as little as possible about the candidates they were seeing. None of the 264 students in the study were shown videos of candidates from his or her home state. Since virtually all of the students were from Harvard, all Massachusetts races were eliminated as well. The authors dropped responses from any student who recognized a particular candidate.
To minimize any bias from lighting or staging, the researchers used 10-second videos of opposing candidates from the same televised debate. Sometimes these clips included full sound; sometimes the sound was purposely muddled (so students could make out the candidates' tone but not their words). Most of the videos were silent.
The results were consistent. Students who saw silent videos picked the right candidate 58 percent of the time, whereas those viewers who heard full sound or muddled sound were only right 52 and 48 percent of the time, respectively, no better than the results of random guessing. Moreover, the predictions from the no-sound videos closely mirrored the results of the actual elections. So, the larger the majority of students that a candidate "won," the larger the share of voters heor she was likely to have won at the ballot box.
Their forecasts were far more accurate than elections based on various economic measures of voter well-being, such as per capita income, unemployment, or state fiscal health. Even when such state and local data was significantly better or worse than national trends, the predictive power of economics was limited.
But if it's not "the economy, stupid" -- if, indeed, Bill Clinton was wrong about the key to a winning campaign message -- then what winning quality were the Harvard students detecting when they picked winning candidates?
The authors looked at the influence of candidates' race, gender, and height to see if students were swayed by these factors. Gender and race weren't a factor, since students were just as accurate predicting races involving two white males as they were in the races overall. Height played a small role but was statistically insignificant. So were qualities such as likeability and physically attractiveness. Candidates the students judged to be good leaders had a slightly better chance of winning than those not rated as good leaders, but the correlation was marginal. Finally, the researchers looked into whether a candidate's confidence influenced students. But in the 22 races that were considered close, where presumably the two candidates were equally confident (and where there were more than 30 student raters, constituting a large sample), the students' accuracy in picking winners was roughly on par with the overall sample.
Thus, the research suggests that some factor beyond the students' own preferences or the vagaries of a particular race -- the authors call it candidate charisma or personal appeal -- is communicable, even during a 10-second silent video clip.
Two political factors turned out to be more accurate than the students' picks. One is incumbency, which accounted for about 23 percent of the voting outcome compared with about 20 percent for the student predictions. Campaign spending was even more accurate, accounting for about33 percent of the outcome.
These findings come with their own chicken-and-egg complexity. If good fundraising causes election success, then candidates' charisma plays a smaller though still significant role in predicting their success. But if good fund-raising is caused by other factors, as other researchers have found, then charisma may play a larger role than this research suggests. The same dilemma conundrum applies to incumbency. The best that can be said is that charisma and ballot box success are related in ways that economic factors cannot come close to matching.
-- Laurent Belsie