In this recently published paper (gated; ungated), Jeremy Bailensen, Shanto Iyengar, Nick Yee, and Nathan Collins digitally “morphed” candidate photographs with the photographs of experimental subjects. They conducted one study involving Charlie Crist and Jim Davis, another involving George Bush and John Kerry, and a third involving (then) potential 2008 presidential candidates that are both more and less familiar to voters (Democrats Hillary Clinton, Edwards, Granholm, and Bayh; Republicans Elizabeth Dole, Giuliani, Kay Bailey Hutchinson, and Robert Ehrlich).[1] Subjects were obviously unaware of any morphing.
Their findings are interesting. To me, these findings also demonstrate the important limitations on how much candidate appearance can influence vote choice and election outcomes.
The main finding is that facial similarity does make voters more likely to support a candidate. But the caveats are just as important:
* The effects are not always large — a “moderate but consistent effect.” The largest effect emerges from the Crist-Davis experiment. Bailenson et al. created a preference measure for each candidate that ranged approximately from -2 to +2.[2] They then subtracted one measure from another, so that negative values indicated a preference for Crist, and positive values a preference for Davis. Presumably, then, this measure ranged from -4 to +4, although few respondents would have been at these extremes. Respondents morphed with Crist had a preference score of -0.60; those morphed with Davis had a score of 0.37. That’s about a 1-point difference on this 8-point scale. Bailenson et al. refer to this as “an extremely large effect.” That’s in the eye of the beholder, I suppose. I’m happy calling it a moderate effect.
* The effects of similarity with regard to party and policy are larger than the effects of facial similarity. That’s true in the study of Bush and Kerry and the 2008 candidates. In the study of the 2008 candidates, familiarity with the candidate was also important. Other things matter more than looks.
* The effects of facial similarity, at least in highly visible races, may be limited to independents and weak partisans. That is their finding in the experiment involving Bush and Kerry. Bailenson et al. combine independents and weak partisans. But given that weak partisans often act like strong partisans, I wonder if the effects might be limited only to “pure” independents (who make up only 10% of the population, and an even smaller fraction of voters).
* The effects of facial similarity appear limited to unfamiliar candidates. When Bailensen et al. analyzed their experiments pooled together, facial similarity did not affect evaluations of Bush, Kerry, Clinton, Dole, Edwards, or Giuliani.
Here’s the upshot, to broaden their results from the impact of facial similarity to that of physical appearance more generally: if “faces matter,” they are likely to matter more in precisely the kinds of races where, ironically, voters are less likely even to know what the candidates look like in the first place. If the ballot actually printed candidate photographs, that would be different. (See this study of another set of low-visibility races — the New Deal for Communities partnership board elections in Great Britain — where such photographs existed.) While it is clear that voters can form snap judgments of competence based on minimal exposure to candidates (see Alexander Todorov’s work; and Andy’s thoughts), it’s not clear how many voters have actually done so.
One other thought concerns the practical implications of facial morphing for candidate strategy. In this news story, Bailensen suggests a possible scenario:
bq. Bailenson said he is unaware of any campaign using the technology to manipulate voters, and said follow-up tests showed that people reacted negatively when they knew their picture was being melded with a candidate’s.
bq. But with free and public sources of mug shots easily found on photo-sharing websites and in motor vehicle department databases, it’s easy to imagine a candidate targeting a potential voter with an ad featuring a morphed photo.
bq. “From an ethical standpoint, I’d hope we never see that happen,” Bailenson said, adding that it takes about 15 minutes and $20 to make a morphed photo with a computer. “Candidates spend seven-, eight- or nine-figure budgets on their campaign. So it’s not outrageous to think that in a swing state such as Ohio or Pennsylvania, you can have 2,000 people sitting in a room morphing every single citizen in the state. That’s a job that’s going to take three weeks and not three years.”
But here’s the problem: nothing besides insufficient resources prevents both candidates from doing this simultaneously. And if I get a piece of direct mail from the Republican with his or her faced morphed with mine, and a similar piece of mail from the Democrat, what’s the net effect? The experiments in this article don’t tell us. My bet is that the effects would wash out.
I hope that more work will be done on this subject, but, even more, I hope that the limitations on the effects of candidate appearance will mitigate some of the hyperventilation in news coverage of this research.
fn1. Granholm is, of course, not eligible to be president. She was born in Canada.
fn2. The measure is standardized with a mean of 0 and a standard deviation of 1. About 95% of subjects should lie between 2 and -2.