People identify the sexual orientation of strangers as fast as 50 milliseconds

ResearchBlogging.orgNalini Ambady has become famous for her research on "thin slicing," the idea that ordinary people can make accurate judgments about others amazingly quickly. We've discussed work from her lab showing that people can accurately predict teaching ability by watching just six seconds of video of a teacher at work. Other judgments, like gender, race, and age, can be made even faster.

But what about less obvious traits? Nicholas Rule and Ambady designed a study to see if college students could accurately identify gay men based on photos alone. They selected 90 photos of men from dating websites, carefully choosing only headshots that didn't feature facial hair, jewelry, glasses, or other accessories. Half the photos were of men seeking male partners, and half were seeking female partners. Then the photos were shown in random order to 90 student volunteers. Photos were displayed for either 33 ms, 50 ms, 100 ms, or 6.5 or 10 seconds. In addition, some of the photos were shown with no time limit at all. Immediately after each photo was shown, a mask of scrambled face parts was shown to clear any afterimages. The students were asked to indicate whether the face they had just seen was likely to be gay or straight. Were they accurate? And if so, how quickly could they do it? Here are the results:


The students responded significantly better than chance for every time period except the 33 millisecond exposure. A chance accuracy rate would be 50 percent, and even after just a 50 millisecond exposure, the students were accurate 57 percent of the time. When the results were corrected using signal detection analysis (to compensate for the fact that fewer than 50 percent of men are gay in real life), accuracy was 62 percent at 50 milliseconds, and as high as 70 percent when self-paced.

One potential problem with this experiment is the source of the photos. Perhaps men on dating sites deliberately present themselves as heterosexual or homosexual to make themselves more attractive to potential mates. To compensate, the researchers found photos of men from Facebook. They carefully chose only photos that were taken and posted by friends or family members, not the men themselves. They used the men's profiles to learn if they were gay or straight, identifying 69 gay men and and 64 straight men. The photos were cropped to show only the faces -- even hairstyles were removed. Then the researchers repeated the original experiment using just a 50 millisecond flash, since that had been the critical interval. Fifteen student volunteers viewed the photos.

Once again, the students were significantly better than chance at identifying gay men, with an accuracy rate of 52 percent, corrected to 54 percent using signal detection analysis. While this means the students were wrong 48 percent of the time, it's nonetheless impressive, especially given the extremely short display time and the fact that hairstyles had been removed from the photos (which was not the case in the first experiment).

In both experiments, there was no different in the results for male and female respondents. Increasing the amount of time beyond 50 milliseconds offered no significant improvement in accuracy.

Why would people be so quick to judge sexual orientation? Rule and Ambady say it might have to do with efficient mate selection. Women need to be able to rule out unsuitable mates, while men need to determine who their potential competition is.

What this study doesn't show, they say, is whether people can successfully conceal their sexual orientation, and how the judgments are being made. Are observers using social cues, or are there visible biological differences between gay and straight men? Either way, it's striking that we can make accurate judgments so quickly, and that the 50-millisecond viewing period is just as accurate as any other.

That said, it doesn't appear that these skills are useful for anything other than improving your odds at detecting sexual orientation. In a situation where it's important to accurately determine whether a man is straight or gay, a 60- to 70-percent accuracy rate still isn't very impressive.

Rule, N., & Ambady, N. (2008). Brief exposures: Male sexual orientation is accurately perceived at 50ms Journal of Experimental Social Psychology, 44 (4), 1100-1105 DOI: 10.1016/j.jesp.2007.12.001

More like this

Family lore has it that my uncle was influential in instituting what is now a fixture in college education: student evaluation of college instructors. He was class president at the University of Washington in the 1960s, when tensions between students and the school administrators were high, and he…
Recently a woman had her sick leave benefits based on a diagnosis of clinical depression terminated because of a few pictures she posted on her Facebook page showing her smiling at a birthday party and enjoying a trip to the beach. Was this a fair assessment of her medical condition? Probably not--…
Farris et al. have a paper coming out in Psychological Science about how men tend to misperceive sexual interest in women. I get the sense that this is a big problem for many women. Any woman who has spent more than 30 seconds in a bar has had at least one random yo-yo hit on them despite what…
Humans are exceptionally good at recognizing faces they've seen before. It doesn't take much study to accurately recall whether or not you've seen a particular face. However, this pattern breaks down when faces come from unfamiliar races. A white person who lives primarily among other whites will…

This study makes me wonder about the standard deviation of an average person's "gaydar". I tend to wonder what could the error bars on those graphs are like. Maybe there are subgroups of people, who perform particularly better or worse than the average at the task? Nevertheless, interesting study.

The error bars (standard errors) are quite small. They'd have to be in order for 52 percent to be statistically different from 50 percent. But remember people are viewing dozens of photos!

I think it might be the opposite: some men are very obviously gay, others are not.

This is a little creepy: "the researchers found photos of men from Facebook. They carefully chose only photos that were taken and posted by friends or family members, not the men themselves. They used the men's profiles to learn if they were gay or straight, identifying 69 gay men and and 64 straight men."

This is also why my profile info is set to "friends only". Although I have so many scientist friends, they could still use my photos for research =P

Just to be clear, the Facebook photos were only shown to 15 research participants, who were never told of the sexual orientation of the people in the pictures.

Also, the photos were from a different university than where the research took place, so the students would be unlikely to know anyone in the photos. I'm not affiliated with a university -- is it possible that if you are, then you can see the Facebook profiles of everyone at the university?

Re comment #2:

Did the study report whether particular photos were statistically more likely to be correctly classified?

would be nice to know if gay people can do it better than straight.

apply the same kind of test using photos from different cultures would be great too.

"is it possible that if you are, then you can see the Facebook profiles of everyone at the university?"

The default privacy setting on Facebook allows anyone in your networks to see your profile in addition to friends, and the default network is usually the university in the email used to set up the account.

By ABradford (not verified) on 08 Dec 2009 #permalink

How much variability was there between individual test subjects? Was everyone about 60-70% accurate, or were some people just very good at judging orientation?

If any of the subjects were gay men themselves, did they have a better success rate?

Well, if the volunteers succeeded in correctly identifying gay men in 54% of the times that they were showed a picture, does it really mean that it is "accurate"? I'm sorry for bothering you with this kind of question since I did not read the paper. Could we know if this difference of 4% statistically significant in such a small sample? I mean, is the observed frequency of correct assignment different from the expected for the sample size? Or is it like flipping a coing to get to know who's gay and who's not?

really, how does a gay person look? this study seems to imply there is a set of characteristics that people can pick out, which i find somewhat offensive. couldn't this instead be interpreted as finding that a certain subset of these men (found on facebook- creepy!) conformed to the stereotype of what a gay man in the US looks like, and in those cases, people were able to pick them out with some amount of accuracy?

Jane, I'm sorry but I find it somewhat offensive that you would suggest that a study should not be conducted because it might find something that you or someone else might find "offensive." I am finding an increase that psychologists (I am one, BTW) are very unwilling to accept and ultra-scrutinizing of findings that don't fit with their own version of PC. I have no issue with being skeptical of results, but we (psychologists) need to ensure that we are skeptical of results that both fit with our views and those that DO NOT fit. While I am not terribly interested in the article nor the topic, why shouldn't people try to identify some physical characteristic associated with sexual orientation if they exist. And, how will we know if they exist unless we study them.

I'm concerned about the ethics of using someone's photograph without their consent, even if that photograph is in the public domain. Would they "reasonably expect" their photograph to be used in such a manner? What are the privacy laws in the US?

The title of this article is misleading. A more accurate title would be "People have a slightly better than average chance of identifying a man's sexual orientation" alternatively you could change "identify" to "judge".

Did participants know that roughly 50 per cent of the photos would be of gay men? Wouldn't better results be obtained by having the proportion of gay men closer to what is found in society? Have the researchers considered that the profiles of men that said they were not gay might be inaccurate?

Were there definitions given for precisely what graduations of orientation and/or behaviour the researchers include in the terms gay/homosexual and straight/heterosexual? And whether/why these were assumed to correspond to subjects' public self-identifications regarding sexuality?

By silverfin (not verified) on 09 Dec 2009 #permalink

There's a recent follow up to this with lesbians:

Female sexual orientation is perceived accurately, rapidly, and automatically from the face and its features.
Rule, Nicholas O.; Ambady, Nalini; Hallett, Katherine C.
Journal of Experimental Social Psychology. Jul 25, 2009. 10.1016/j.jesp.2009.07.010

By lrosielle (not verified) on 09 Dec 2009 #permalink

Hey guys i did a study. Placed 50 red balls, and 50 black balls in a bag. Then I had people guess whether the ball i was holding was red or black.

About 50% of the answers were right. amazing how accurately people could guess. I bet they all have ESP.

Maybe people doing this test were more likely to guess as gay than straight. Maybe because of the very nature of the test, people felt more inclined to say the men were gay.

Is the 54% of correct guesses related to how many gay men were guessed as gay? Does it take into account how many straight men were guessed as gay?

If I'm reading the study correctly, they showed about 15 students about 100 pictures, so N is about 1500. The uncertainty usually goes by the root of N, so an error of about 2%, so the study appears to have a statistically significant result.
FYI this is also why political polls of about 1000 people has an error of about 3%

This is assuming, of course, that all the pictures were of men who were totally honest and comfortable within their sexuality.

A couple of closet cases in those 100 pictures could probably skew the statistics up a bit.

I'm probably just being a little snarky here but I would *love* to know what those people put on their funding forms for this project...

I have a ton of questions about the study that are not adequately answered in the actual paper.

All of these people (mostly women 68 of n = 90 for study 1A and mostly men 9 of n = 15 for study 1B, and all undergraduate students, I might add) were explicitly told they were looking at sexual orientation. Knowing they were being observed for this task would confound the study (the Hawthorne effect).

How were they selected? If they were all psychology majors, they might be more sensitive to visual cues for behavior. Would this result be replicated in a more general population, and not just those in a higher education setting?

Is chance really 0.5? Using self-reported sexual orientation introduces all kinds of confounders. For all we know, 60% of the pictures samples may have been homosexual individuals, and the students really were guessing at 33 ms. Did they do a more conservative two-tailed comparison, or just a one-tailed to make their results look more significant? Did they do a test-retest to validate the reliability of their pictures?

In fact, I'd rather know what percentage guessed the picture correctly at each condition, instead of what percentage a participant got right. A paired test in that case would help compare apples to apples. What was the level of agreement? That they did not present that data seems very suspicious. Also, What was to stop them from creating a set of "neutral" pictures as a real control?

Full of holes. And why it's in a journal with impact factor of only 2.5.

N is not 1500. They were measuring a proportion with each of the 90 students. Either do the power calculation for 90 students or 90 pictures. Not both. Doesn't make any statistical sense to multiple them. 15 students and 160 pictures for 1B.

I agree with Steve P. "The title of this article is misleading. A more accurate title would be "People have a slightly better than average chance of identifying a man's sexual orientation" alternatively you could change "identify" to "judge"."

So women are not people according to the article title. Thanks.

So women are not people according to the article title. Thanks.

Actually "People" are the ones doing the identifying, not the ones being identified. Yes, they do include women. And take a look at the link in #15 for a replication of these results using female faces.

So, firstly, men who are likely to be open about being gay...are also more likely to adopt a stereotypical gay "look" in their choice of personal grooming. This is not exactly news.

And secondly, people living in a culture where there's a lot of openly gay men with that look...can detect it in a few dozen milliseconds. Except a small group got it wrong slightly less than half the time. This is not exactly a result, and doesn't match the headline.

E: are you actually suggesting that they should have created some set of "neutral" photos as a control? How would introducing a homogeneous set of pictures that clearly look different from the other stimuli help?

Hawthorne effect - can you actually come up with any plausible reason that knowing what the study is about would skew the results in a predictable way? The DV is a signal detection measure - the fact that subjects may have been more attuned to the photos because they were trying harder or something does not make this paper less interesting.

And as to the question about what % of the photos were of gay men - that's why they reported a signal detection analyses rather than just analyzing how many gay men were correctly categorized.

Some of these criticisms are really bizarre and unfounded.

I was invited by a friend of mine to join Facebook (not going to happen). I laughed when I saw he identified as being 'straight'. He lives in a very conservative Catholic country where being gay or atheist is akin to being a leper. And that's the first of many problems with this study, unless you use pictures of subjects that have been subject to a peter meter, and their 'true' sexuality is known, the results are suspect immediately. I doubt the men that identified as being gay were lying, but I suspect some of the self identified straight men did lie.

I doubt if the identifying of sexual orientation is actually happening in 50 ms as is suggested. To me, it seems more plausible that 50 ms is enough to load a face into working memory so that analysis can subsequently happen.

Also, I'm wondering how exactly these pictures were collected. Since the researchers wanted 50% gay pictures, maybe they needed to "hunt" for them. Perhaps subconsciously they would be more inclined to check the profiles of men with pictures that they themselves judged as likely to be gay. This might lead to only pictures of (relatively) obviously gay men being used.
Also, self reporting of sexual orientation on an public online profile may not be 100% accurate.

Finally, is 54% classification error something to be genuinely excited about? I understand that it is significant, but it's such a small effect... What does it suggest?