I was outed three years ago. Johann Hari, now a contributor to Slate, declared me a gay writer. I'm not sure how he figured it out. Maybe it was my conspicuous interest in homosexuality. Maybe it was my writing style, a photo, or a TV appearance. Somehow, he knew.
Now Marcus Bachmann, husband of Republican presidential candidate Rep. Michele Bachmann, is setting off gaydar alarms. It started as a subtle joke among bloggers. Then it progressed to parody and overt insinuation. On Tuesday, Dan Savage said the Bachmanns' marriage was frigid because Mr. Bachmann may have "tiptoed down" the road to homosexuality "just a couple of inches … maybe six, maybe seven." As evidence, Savage cited Bachmann's "mincing" in a YouTube clip, plus "the sound of his voice." He concluded that Bachmann "appears to be a lying closet case." On Wednesday, Jon Stewart said Bachmann, who counsels homosexuals to overcome their urges, seems to be doing this "so he can hoard all the gayness for himself." Stewart said Bachmann "dances and sounds not only gay, but center-square gay."
There's nothing new about calling somebody gay based on a lisp or a girlish gait. We all saw, did, or suffered it in grade school. What's unusual is seeing grown-up gays and liberals do it in 2011 with such open ridicule. But don't worry: The new queer-hunters are progressive. They detect homosexuality based on science, not stereotypes. Savage cites a series of studies, written up two years ago in Scientific American, in which college students correctly distinguished gay from straight men based on facial features. He concludes: "Gaydar is for real."
Is it? Can gaydar identify Marcus Bachmann's orientation? Let's look at those studies.
The five key experiments, conducted by Nicholas Rule, Nalini Ambady, Reginald Adams Jr., and Neil Macrae, are impressive. In the first study, undergraduates were shown pictures of 81 men taken from online personal ads. The pictures were stripped of any giveaway context. The students were asked to guess quickly, based on what "most people" or "society" would say, whether each man was "very gay," "somewhat gay", "somewhat straight," or "very straight." Their performance was rated by a "correlation coefficient" (R) which, when squared, shows the extent to which variance in the students' estimates of the men's orientations can be explained by the men's actual orientations—a rough measure of accuracy. They scored an average R value of 0.31, which implies that accurate gaydar accounted for about 9 percent of the variance. *
In the second study, the research team removed various facial clues from the pictures to see whether their absence made a difference. The students still did well. With each man's hair removed, the students scored an R of 0.19. With his mouth obscured, they scored 0.22. With his eyes obscured, they scored 0.26. This suggests accurate gaydar accounted for 4 percent to 6 percent of the variance in estimates.
Even when the students were shown just one feature for each man—hair, mouth, or eyes—they outperformed random guessing. Given only the man's eyes, they scored 0.11 to 0.12. Given only his mouth, they scored 0.11 to 0.15. Given only his hair, they scored 0.24 to 0.27.
The researchers noted that "judgments based on hair only were significantly more accurate than those based on the eyes only or mouth area only." Why? Possibly because hairstyle "is a deliberate aspect of appearance that is groomed to look a particular way." It's culturally influenced and personally expressive. Reading somebody's hair isn't like reading his palm. It's more like picking up a behavioral signal.
And this is the problem, more generally, with using personal-ad photos in gaydar experiments. When you take or choose a photo of yourself for a personal ad, you're trying to send signals. From hair to eyes to mouth, you're conveying your sexuality, aiming at a particular audience, and trying to fit in. That's great for attracting a partner. But it's lousy for testing gaydar. You're making your orientation too obvious.
The researchers recognized this problem. "It is possible that differences in self-presentation may have led to some systematic differences in the appearance of gay and straight men when posting photos of themselves on personal advertisements," they conceded. So, in their final experiment, they tried to eliminate self-presentation. They used pictures from Facebook. They didn't use pictures chosen by the men whose faces were shown. They used pictures posted by friends of these men—pictures showing several people, to minimize the chance that the picture had been selected to emphasize anything about the man in question. The researchers noted that "these album photos are often candid, 'real life' photos (i.e., absent the target's awareness that the photo is being taken), rather than posed shots."
How did the students do with these photos? They scored an R value of 0.22—again, significantly better than chance, though not by much. According to the authors, this shows that "sexual orientation may be accurately perceived from male targets' faces in the absence of the self-presentation demands of personal advertisements." In this way, the results "extended the generalizability" of the evidence for gaydar.
But wait a minute. How did the researchers know which men on Facebook were gay? Answer: "We performed a search for men's profiles that indicated romantic or sexual interest in other men (i.e., self-identified homosexual men)." All the gay men were out. Not a closet case in the bunch.
That's a totally understandable constraint. You can't score the students' accuracy if you don't know which men are gay. But it also limits your ability to generalize the findings. Even in a candid group photo on somebody else's Facebook page, a man presents himself, consciously or not. His hair, his eyes, and his expression may all be influenced by his identity and the patterns of presentation he associates with that identity. If one of every five openly gay men displays a visual signal strong enough to pick up, that'll boost your gaydar accuracy to 60 percent.
What it won't do is extend your gaydar into the closet. A man who doesn't identify himself as homosexual might not absorb or project homosexual patterns of presentation. He isn't trying to fit in with gay men. He's trying to fit in with straight men. So when you hear a lisp or see a fussy walk, you can't infer that it's a signal. Subtract the people who lisp for effect, and you're left with the people who lisp because they can't help it.
Unless homosexuality comes with innate visible or audible characteristics, gaydar is signal-reading. And signal-reading ends where the signal fades: in most cases, at the limits of gay self-awareness. Yes, some closet cases give off a vibe. But it's wildly unlikely that they'll do so with the same frequency or clarity as openly gay men. Which leaves you with a much higher error rate than you might infer from these studies.
If you want to test gaydar in the closet, try showing your subjects pictures of a different group of men: ex-gays. You can find their names, sometimes with photos or videos, proudly displayed on missionary websites. These men have rejected gay identity and are trying not to send signals. If gaydar can pick them out in an experiment, that would be one hell of a story.
Maybe Marcus Bachmann is gay. Maybe I am, too. We're on opposite sides of the debate over homosexuality, but both of us are suspiciously interested in it, don't you think? I keep writing about gay marriage, fellatio, and anal sex. Isn't it getting kind of obvious?
Then again, I remember an acquaintance telling me years ago, with perfect self-assurance, that a fellow journalist was gay. I asked how he knew. "You can tell," he said. That was before the three of us came to work at Slate. Now we're all in straight marriages. Is one of us living a lie? I wouldn't bet on it. I wouldn't bet on it in Marcus Bachmann's case, either. My gaydar isn't that good.
Correction, July 19, 2011: I initially reported the R values as a linear representation of the guessers' accuracy. This was a huge ignorant goof. R values are the square root of the percentage of variance in one thing that can be explained by variance in another. So an R value of 0.31 doesn't signify that students were 65 percent (or any other percent) accurate in guessing men's sexual orientations from photos. It signifies that the men's actual orientations accounted for about 9 percent of the variance in the students' estimates of those orientations. The paper by Rule et al argues that the R values, statistically, are significantly better than chance. But no accuracy rate can be directly computed from the data presented in the paper. I'm indebted to Slate commenters Andrew Gottlieb, Ann Calhoun-Sauls, and Marcus Felson for pointing out the error and explaining what the R values do and don't mean. ( Return to the corrected sentence.)
(Readings I recommend: Awesome Slatecontributor Jesse Bering wrote the Scientific American column on the gaydar studies. Joe.My.God has the best blanket coverage of all things gay and Bachmann. Lauri Apple at Gawker rounds up the early gaydar alarms about Bachmann. Cher led the gaydar squad, then chimed in yesterday with this tweet: "M.Bachman should b strung up by her Testis..well Some-1 in her house has got 2 have a pair?" Mariah Blake at The Nationdetails an inside investigation, led by John Becker at Truth Wins Out, of Bachmann's homosexuality-purging therapy. Michelle Goldberg at the Daily Beast has transcripts of the therapy sessions. Bachmann tells the Star Tribune that his reference to "barbarians" in a widely distributed audio clip actually had " nothing to do with homosexuality," so the clip must have been doctored.)