The Mathematics Of Beauty

January 10th, 2011 by Christian Rudder

This post investigates female attractiveness, but without the usual photo analysis stuff. Instead, we look past a woman's picture, into the reaction she creates in the reptile mind of the human male.

Among the remarkable things we'll show:

Fair warning: we're about to objectify women, big-time. The whole purpose of this blog is to analyze OkCupid's data, and without a little bit of objectification that's impossible. Men will get their turn under the microscope soon enough. As usual, none of this (with the exception of the celebrity examples) is my opinion. All data is collected from actual user activity.

Let's start at the beginning.

All people, but especially guys, spend a disproportionate amount of energy searching for, browsing, and messaging our hottest users. As I've noted before, a hot woman receives roughly the messages an average-looking woman gets, and 25× as many as an ugly one. Getting swamped with messages drives users, especially women, away. So we have to analyze and redirect this tendency, lest OkCupid become sausageparty.com.

Every so often we run diagnostic plots like the one below, showing how many messages a sampling of 5,000 women, sorted by attractiveness, received over the last month.

These graphs are adjusted for race, location, age, profile completeness, login activity, and so on—the only meaningful difference between the people plotted is their looks. After running a bunch of these, we began to ask ourselves: what else accounts for the wide spread of the x's, particularly on the "above-average" half of the graph? Is it just randomness?

What is it about her:

that gets more attention than her:

...even though according to our users, they're both good-looking?

Not all 7s are the same

It turns out that the first step to understanding this phenomenon is to go deeper into the mathematically different ways you can be attractive.

For example, using the classic 10-point 'looks' scale, let's say a person's a 7. It could be that everyone who sees her thinks exactly that: she's pretty cute.

But something extreme like this could just as easily be going on:

If all we know is that she is a 7, there's no way to tell. Maybe for some guys our hypothetical woman is the cat's pajamas and for the rest she's the cat Garfield. Who knows?

As it turns out, this distribution of opinions is very important.

Celebrity photos: to titillate and inform

Let's look at what the ratings distribution might be for a couple famous people. I imagine that for, say, the actress Kristen Bell it would be roughly like this:

Ms. Bell is universally considered good-looking, but it's not like she's a supermodel or anything. She would probably get a few votes in the 'super hot' range, lots around 'very attractive', and almost none at the 'unattractive' end of the graph.

Compare her to Megan Fox, who might rate like this:

On the far right, you have the many dudes who think she's the sexiest thing ever. On the far left, you have the small number of people who have seen her movies.

Unlike Ms. Bell, Ms. Fox produces a strong reaction, even if it's sometimes negative.

Real People

Now let's look back at the two real users from before, this time with their own graphs. OkCupid uses a 1 to 5 star system for rating people, so the rest of our discussion will be in those terms. All the users pictured were generous and confident enough to allow us to dissect their experience on our site, and we appreciate it. Okay, so we have:

As you can see, though the average attractiveness for the two women above is very close, their vote patterns differ. On the left you have consensus, and on the right you have split opinion.

To put a fine point on it:

  • Ms. Left is, in an absolute sense, considered slightly more attractive
  • Ms. Right was also given the lowest rating 142% more often
  • yet Ms. Right gets as many messages

When we began pairing other people of similar looks and profiles, but different message outcomes, this pattern presented itself again and again. The less-messaged woman was usually considered consistently attractive, while the more-messaged woman often created variation in male opinion. Here are a couple more examples:

We felt like were on to something, so, being math nerds, we put on sweatpants. Then we did some work.

Our first result was to compare the standard deviation of a woman's votes to the messages she gets. The more men disagree about a woman's looks, the more they like her. We found that the more men disagree about a woman's looks, the more they like her. I've plotted the deviation vs. messages curve below, again including some examples.

The women along the graph are near the 80th percentile in overall attractiveness. You can click the tiny thumbnails to expand them.

As you can see, a woman gets a better response from men as men become less consistent in their opinions of her.

Our next step was to analyze a woman's actual vote pattern of 1s, 2s, 3s, 4s, and 5s:

If You're Into Algebra

We did a regression on the votes for and messages to a sample of 43,000 women. To keep everything consistent, all the women were straight, between the ages of 20 and 27, and lived in the same city. The formula given in the body of the post was the best-fit we found on our second regression, after dropping the m3 term because its p-value was very near 1.

msgs are the number of messages the woman received during the observation period. The constant k reflects her overall level of site activity. For this equation, R2 = .28, which isn't great in a lab or on a problem set, but is actually very good in a real-world environment.

This required a bit more math and is harder to explain with a simple line-chart. Basically, we derived a formula to predict the amount of attention a woman gets, based on the curve of her votes. With this we can translate what guys think of a woman's looks into how much attention she actually gets.

The equation we arrived at might look opaque, but when we get into it, we'll see it says some funny things about guys and how they decide which women to hit on.

The most important thing to understand is that the ms are the men voting on her looks, making up her graph, like so:

And those ms with positive numbers in front contribute to messaging; the ones with negative numbers subtract from it. Here's what this formula is telling us:

The more men who say you're hot, the more messages you get.

How we know this—the .9 in front of m5 is the biggest positive number, meaning that the guys who think you're amazing (voting you a perfect '5') are the strongest contributors to your messaging income. This is certainly an expected result and gives us some indication our formula is making sense.

Men who think you're cute actually subtract from your message count.

How we know this—because the .1 coefficient in front of m4 is negative. This tells us that guys giving you a '4', who are actually rating you above average-looking, are taking away from the messages you get. Very surprising. In fact, when you combine this with the positive number in front of the m1 term, our formula says that, statistically speaking:

If someone doesn't think you're hot, the next best thing for them to think is that you're ugly.

This is a pretty crazy result, but every time we ran the numbers—changing the constraints, trying different data samples, and so on—it came back to stare us in the face. In plain scientific terms, it was like a baby we were trying to drown had somehow grown gills. (This happens all the time in China.)

What We Think Is Going On

So this is our paradox: when some men think you're ugly, other men are more likely to message you. And when some men think you're cute, other men become less interested. Why would this happen? Perhaps a little game theory can explain:

Suppose you're a man who's really into someone. If you suspect other men are uninterested, it means less competition. You therefore have an added incentive to send a message. You might start thinking: maybe she's lonely. . . maybe she's just waiting to find a guy who appreciates her. . . at least I won't get lost in the crowd. . . maybe these small thoughts, plus the fact that you really think she's hot, prod you to action. You send her the perfectly crafted opening message.

"sup"

On the other hand, a woman with a preponderance of '4' votes, someone conventionally cute, but not totally hot, might appear to be more in-demand than she actually is. To the typical man considering her, she's obviously attractive enough to create the impression that other guys are into her, too. But maybe she's hot enough for him to throw caution (and grammar) to the wind and send her a message. It's the curse of being cute.

The overall picture looks something like this:

Finally: What This Could Mean To You

I don't assume every woman cares if guys notice her or not, but if you do, what does all the above analysis mean in practical terms?

Well, fundamentally, it's hard to change your overall attractiveness (the big single number we were talking about at the beginning). However, the variance you create is under your control, and it's simple to maximize:

Take whatever you think some guys don't like—and play it up.

As you've probably already noticed, women with tattoos and piercings seem to have an intuitive grasp of this principle. They show off what makes them different, and who cares if some people don't like it. And they get lots of attention from men.

But our advice can apply to anyone. Browsing OkCupid, I see so many photos that are clearly designed to minimize some supposedly unattractive trait—the close-cropped picture of a person who's probably overweight is the classic example. We now have mathematical evidence that minimizing your "flaws" is the opposite of what you should do. If you're a little chubby, play it up. If you have a big nose, play it up. If you have a weird snaggletooth, play it up: statistically, the guys who don't like it can only help you, and the ones who do like it will be all the more excited.

Max Shron contributed additional research to this post.
OkCupid is hiring a .

195 Responses to “The Mathematics Of Beauty”

  1. Will says:

    I would be interested in knowing if there is any correlation among the group of men who positively rate the contentious pictures, whether the number of messages those high raters send out is larger than average, and finally if they message their 5 star pick more than others do.

    Obviously this is pure conjecture, but a reasonable alternative hypothesis – which you may have already considered and ruled out – to your (paraphrased) conclusion that “many men behave the same way and seek out the niches in their tastes to reduce competition” could be:

    “There are two subsets of male users who, despite their stated preferences (which you could control for) disagree about the desirability of a certain trait and have different messaging habits”.

    One possible way to imagine this would be if you have one subset looking to ‘hook up’ and another looking for a longer relationship. If the groups have a difference in opinion about a certain feature, such as whether a person looks “likely to hook up”, this could lead to a splitting of the results and the discrepancy in messages. Imagine that the “hook up” group messages the most users in general and prioritizes their messages based on how likely it seems they will score. If the other subset of men dislike that exact feature, but message less on average, then you might see the results found here.

    This is only one of several possible scenarios which could produce the observed correlation and each alternative would have different implications for how women should act depending on their preferences. Overall however, this is a very interesting finding and thanks for sharing such detailed info! It would also be interesting if you would share some of the ‘possible conclusions’ for which you tried but failed to find any correlation.

  2. Jordan says:

    Were these ratings taken from the Quickmatch section or some other rating area? I find myself rating people in Quickmatch based more on their profile than overall attractiveness. Just wondering where the data actually came from. Thanks.

  3. Janik says:

    Oh, poor Annie, being cute and all. … I agree, it might have been more sensible to first objectify the male part of society instead of giving anybody a chance to being a fussy clichée.
    But as a little comfort to you Annie, this might count as a message to you. Already one point up!

    PS. I also generally like the expresion to »choose to only choose to«. Kind of poetic.

  4. Steven says:

    Unfortunately, Annie, the hurtin’ is clouding your reading comprehension. The second paragraph clearly states:

    “Men will get their turn under the microscope soon enough.”

    Which will undoubtedly uncover some interesting facts that might put the final nail in the coffin of my idea that wearing Yu-Gi-Oh! apparel and a hitler ‘stache ironically has certainly upped my sexy quotient by, like, 2 full points. Please OkCupid, let me down gently…

    And don’t worry! There are plenty of people who like people like you, too. :)

  5. Garrett says:

    I’m going this route:

    A guy that sees someone as cute, sees someone as on his level. That is, someone that he could consider himself being in a long term relationship with. However, most online dating sites are used by guys trying to get their rocks off. So they go for the ugly chicks that are ‘desperate’ and the ‘hot’ ones, hoping to get lucky. Uggo and easy or shoot for the stars. That’s my breakdown, take it or leave it.

  6. Tom says:

    @Andrew: I’m fairly certain the article says exactly what you just tried to say.
    @Annie: Care to read the first paragraph? I’m sure Christian is working on the Male version of this as well.

  7. Robzilla1 says:

    I think you may be taking the short answer to what generates messages. I feel there are many other factors, but there are two large items not being taken into account. 1. The males sending the messages. Their favored choices. Are the same males choosing the same female type and send numerous emails. 2. The contents of the profile. Using your example of the ladies with tattoos and piercings. Their profiles may contain items that appeal to a specific sector that would concentrate a segment of the population that may a) be more prolific, or b) be a greater percentage of the online dating population.

    I would agree that people may feel that certain persons will receive less emails due to the perception that “they are out of my league”.

    I do feel that this observation is slightly shall and should be taken deeper before a hypothesis is presented. You should go a few levels deeper.

    Rob

  8. Holly Wood says:

    Eh, I’d be more reserved about applying game theory here. Your data consistently shows men email hot women more frequently. Period. I doubt there’s much thought that goes behind that action besides, “She’s hot! I should message her!”

    Other than that, this is awesome evidence that somewhat validates my previously-held theory that many young men who use OkCupid “on a whim” only use it to try to gain access to what they see as very hot women, ignoring women who they think they could talk to in real life (even if they don’t really). It’s not a critique of your site at all, but something I’ve noticed of the men in my life. They will waffle on a very cute girl’s profile for days, but will instantly message a hot chick with the “What the hell, right?” attitude.

  9. Michael (Tao of Bachelorhood) says:

    Another great piece of research. This pretty much matches up with my observations. I’ll be recommending that my readers message the “cute” girls for less competition (don’t worry girls, I’m teaching them how to dress and behave better too).

    I’ll be looking forward to seeing similar research on men, although I imagine looks won’t carry the same weight.

  10. Adam says:

    So, the ones who I think everyone is all over actually get no messages? Sweet.

  11. jesse says:

    Sometimes I give girls one star because they piss me off in some way. Matter of fact that might be the only reason I ever give a girl one star. In my experience, the more hot a woman is the less polite she is.

  12. Ange says:

    Your trying to evaluate attractiveness in relationship to beauty and its often more about relatedness. How much a person relates to the picture of a possible mate is often measured in how the other person’s personality attracts them. If I look at two pictures of guys I think are equaly attractive in the face but from the neck down show very different life interests, I will judge them differently. Even an ugly faced guy with similar life interests to me interests me more than someone with personal interests not to my liking. Maybe this is just a women thing, i.e. personality over looks.

  13. Gleb says:

    Looking forward to a similar post on guys. Probably better to base it on responses though :)

  14. Joshuah Kuttenkuler says:

    This is all correct but I believe that this theory should be extended to allow individuals
    to receive an attractiveness rating with both horizontal and vertical significance.
    I also believe it extends to males and thusly below is my personal opinion.
    Graph of attractiveness

    Female
    —————
    Cute Hot
    More
    ^
    |
    |
    attractive
    |
    |
    \/
    less
    —————andro fem
    -
    -No gender
    -
    ————– andro male
    Less
    ^
    |
    |
    attractive
    |
    |
    \/
    More
    Cute Masculine

  15. dbt says:

    Annie: “Men will get their turn under the microscope soon enough.”

  16. B says:

    i loved this article. gives me a whole new outlook!

  17. Jake says:

    “OKcupid telling me how being cute is detrimental (when I am often told I’m cute) is not what I need”.

    OK we get it, you’re cute. Let’s move on with our lives please?
    Stop being a sensitive little bitch and appreciate the science behind their analysis, even if it is at the cost of your fragile bitch-self-esteem.

  18. Doop says:

    Very interesting. I think, though, that you should figure out a way to take into account how people actually use the Quick Match section. I simply rate everyone a 2 or 1, depending on the caprice of mood, unless I am interested in them. And my votes almost always take into account their relevance to me as a potential date, and only vary rarely is the star system used to rank based solely on photo alone.

    I mark 4 or 5 simply if I want that other person to get an email that says “someone thought very highly of you, check to see if you can figure out who it is” or whatever that email says. If it’s I rank a 4 and I get no response from that person (via the quickmatch, you both ranked each other highly email) then I move on, never contacting her. If I rank someone a 5 I either vote and then contact her, or I don’t rank her at all and contact her.

    Your analysis is thus suspicious.

  19. Brett says:

    Though the number of messages received does indicate the level of interest from others, that interest isn’t necessarily positive. Could it be that some of the messages are negative messages? Hate messages? Bullying messages?

    If someone thinks a girl is great, they are likely to send a positive message.

    If someone thinks a girl is just okay, they are likely to ignore her.

    If someone thinks a girl is terrible, could they be somewhat likely to send a negative message?

  20. Brian says:

    Could you possibly post a plot of the actual number of messages received vs the number predicted by your equation? I’m interested to see how closely they correlate.

  21. Jason says:

    BRILLIANT!

    Seriously, I’ve had discussions similar to this in real life w/ friends. Nice article.

  22. jason says:

    In a word, it’s about confidence. I think guys can pick up on insecurity and know when you’re trying to downplay parts of yourself you may not like. Confidence is always sexier than insecurity.

  23. pyromosh says:

    I’m curious if you controlled for other factors that are visible to those rating the subject.

    For instance, when I rate someone via a match search, I can see the little thing that says how often the person replies. And it sometimes also says “So and so hasn’t received any messages in a while. You should message her.”

    Could this be skewing your results in the direction you’re seeing?

  24. Katie says:

    I’ve had a lot of guys call me cute, but i don’t see that as something bad. I think I look good, and that’s all that matters. Confidence makes you more beautiful exponentially

  25. Candy says:

    It looks like the women who get the high response rate are the ‘vampish’ ones, who exude more sexuality. But being ‘vampish’ also reduces a woman’s score because men make moral judgements about her.

  26. Another_Phil says:

    Two questions… is the heteroscedasticity anywhere near as pronounced if the data is topped and tailed? It seems that the variance for average looking majority is far more uniform. Also, some of the heteroscedasticity caused by a floor effect in the below average attractiveness message counts. Is there any way of checking for this?

  27. Tyler says:

    There’s one big problem here…

    There is no way to see how many messages a user is receiving, only how often they reply.

    Therefor the entire study is just a nice example of fancy mathematics, because men can’t use information they don’t know to make the decision if they should message a woman or not.

  28. Justin says:

    What happens when a girl gets almost exclusively 1 votes, and does your fancy linear formula account for this?

  29. jrr says:

    Jave you considered trying to counter this trend with the design of your system? It’d be controversial, but perhaps publish users’ relative popularity according to message income, or less directly their overall response rate?

    This way you can get right to the part of the brain that guesses whether other the competition is interested or not,

  30. PE says:

    I think it’s an interesting analysis. I’m not sure the motivations, but I think the end advice is a good one. Stop trying to hide your basic flaws. I mean, don’t you think that guys understands that every pic is cropped a certain way? What’s funny is some of the really bad photos (cell phone shots too) that aren’t even trying to hide stuff.

    I think it does show up one big flaw with all the online dating sites. Men get no responses, women (esp desirable ones) get obliterated. Basic economics. Unless there is a cost to messaging someone (or for women, not), it’s a race to the bottom. Stick that in your game theory.

    Also — the China joke was funny. You just need to get it.

  31. User says:

    Once and for all: OKCUPID USERS DON’T USE THE RATING TOOL MERELY TO RATE SOMEONE’S APPEARANCE!

    For the life of me I can’t figure out what’s driving you to the assumption that all users use the rating tool the same way. Personally I’ve only used it several times, and my vote ALWAYS reflected BOTH LOOKS AND PERSONALITY.
    Thus your research results regarding “prettiness” are absolutely meaningless.

  32. Josh says:

    I like the research, but disagree with the game theory explanation. There are a lot of profiles out there and men actually have to be motivated to send a message. They’ll do it if something stands out about the recipient. What is it about the person that speaks to You? It could be the imperfection itself.

    It’s almost a cliche to note that women latch on to certain “ugly” imperfections that they find attractive in a man. It’s not surprising that men do the same thing with women, but it’s not really talked about.

    I’d like to see pictures of women in the 25 – 50 percentile who get lots of messages. That might be even more instructive.

  33. Josh says:

    I like the research, but disagree with the game theory explanation. There are a lot of profiles out there and men actually have to be motivated to send a message. They’ll do it if something stands out about the recipient. What is it about the person that speaks to You? It could be the imperfection itself.

    It’s almost a cliche to note that women latch on to certain “ugly” imperfections that they find attractive in a man. It’s not surprising that men do the same thing with women, but it’s not really talked about.

    I’d like to see pictures of women in the 25 – 50 percentile who get lots of messages. That might be even more instructive.

    Look at the movie star Christina Ricci – especially before she lost weight and went for a more Hollywood look. Men were crazed for her, not because they thought they could get her, but because there was something about her that spoke to them.

  34. Ali says:

    I’d really like to see one done about gays and lesbians. I get the feeling for gay guys it’s simply the better the looks the higher the messages, but with lesbians it seems to be more complicated. It seems that the less attractive a girl is, the more messages she gets. Perhaps because it’s left up to girls to initiate and girls are generally more insecure; therefore, they are afraid to message the more attractive girls. It might be something to look into.

  35. Jesse Schell says:

    Nowhere does the article directly mention the power of the exotic. Exotic features, or expressions, or accoutrements, or lighting, are all things that excite. But, of course, what is exotic to one person is tedious to another. The reason for someone to connect online is because they see someone very different than the people they meet day to day. So exotic+online is a good formula for getting messages, which I think matches the conclusions of the article?

  36. Charlie says:

    Holly — men do consistently message hot women, but your interpretation oversimplifies OkTrends’ conclusion: men DO NOT AGREE on which women are hot. “I doubt there’s much thought behind it” is an absurd, reductionist thing to say; you seem to be resting on the belief that “men are pigs who’ve been indoctrinated by media images,” and it strikes me as disingenuous. You’re missing the point that men are messaging the women whose beauty is most contested, not most agreed-upon.

    The “game theory” interpretation is convincing to me. However, I wonder if part of this phenomenon can be explained by a girls’ styling and aesthetic choices. That is, I wonder if the polarizing opinions you’re seeing have more to do with perceived subculture than some ugly duckling, eye-of-the-beholder thing. If she’s wearing a Flashdance off-the-shoulder American Apparel sweater and giant librarian frames, she’ll alienate more frat boys and entice more hipsters than a tank top ever would. The male reaction isn’t centered on “she’s ugly!” “no, she’s beautiful!” It’s more, “I hate that hipster crap” vs. “Let’s go back to your loft in Bushwick.”

  37. GuyWhoDoesntLikeSluttyGirls says:

    Am I the only one that noticed that the deviant [sic] woman have a “sluttier” look to them? Even Megan Fox in that picture is practically throwing her breasts at you, while Kristen Bell’s got that “nice girl” look. Just sayin’.

  38. Jodi says:

    This is fabulous! thank you so much for doing that!

    I’m interested to know what the response rate is based on both how attractive the woman or man is, and also based on the length and quality of the emails being received. For example, your perfectly crafted email of ‘sup vrs an email that is more than 10 words and actually relates to something in your profile besides the photo.

    Fantastic job!

  39. Peter Gerdes says:

    Seems to me this analysis misses several key points:

    1) Our estimation of attractiveness is affected by how appealing we find people and a picture can convey a great deal about someone.

    2) People don’t like to be mean. Thus there is a tendency to default to rating someone as ‘above average’ (a four) unless something about them repels you or particularly attracts you.

    Given these points doesn’t the following simple model offer a better explanation of your data:

    A girl’s picture conveys both her physical attractiveness and facts about her personality/subculture. Men’s rating of attractiveness is really just a proxy for their overall level of interest with 1-3 indicating disinterest, 4 ‘could be interesting but I dunno’ and 5 indicating intrest. Obviously more physical attractiveness correlates with men’s interest hence the positive correlation between messages and average attractiveness.

    However, a physically appealing girl who doesn’t reveal any personality features in her photo will almost always get the hedge rating of 4 (maybe I’d be interested). Those who reveal personality features repeal some guys who rate them low and attract others. Guys only tend to message because of active interest not on the possibility of interest.

    In other words girls get low variance because they don’t project any particular personality features.

    However, there are some other hypothesizes I don’t think you’ve ruled out. How about the really simple one that a high percentage of the user base is reluctant to rate girls a 5, i.e., no matter how attractive you are you get at least 50% 4′s or lower so the distribution is highly non-normal and the standard deviation is trivially correlated to average score.

    Also did you do a hypothesis test on the -.1m4 coefficient to check if it could be distinguished from 0?

  40. MK says:

    Great stuff! That the thing about numbers – they don’t lie…

  41. Kim says:

    I would *love* to have my “attractiveness curve” data processed, using your equation. I’d be very interested to see if it agrees with the number of emails I receive. Please feel free to email me, and I’ll be happy to give you my username.

    Great job — I mathboys! ;-)

  42. bahomet says:

    Thanks – but how about a post about The Beauty of Mathematics?

  43. Cory says:

    I know personally the quick match section to be less of an actual rating of attractiveness and more of a do i want this person to receive a email encouraging them to go rate people (aka possibly rating me) and then both know that the other person is interested

    rating is somewhat of an all or nothing for me and a few other users i know

    five stars if you’d like to talk to them and get to know them

    or one star if you simply aren’t interested

    theres no difference in a 1 star and a 3 star or a 4 star and a 5 star 4 and 5 send emails 1-3 means sorry you’re not that interesting

    also the quiver section usually does more of flooding my inbox then presenting people id actually like to talk to so the majority of them get not interested simply based on the fact theres no difference in messaging

    thanks for keeping this site fun though the data analysis blogs are one of the most fun things about okcupid and i hope they continue

    ps permission granted to objectify me for the good of science

  44. hanyouatomsk says:

    I LOL’d at the Splice reference.

  45. drewex says:

    Why dont we see our own ratings. What people think of us. There should be a rating based per picture.