OkTrends

This week we will be confronting a fact that, by definition, haunts the average online dater: no matter how much time you spend polishing your profile, honing your IM banter, and perfecting your message introductions, it’s your picture that matters most.

We’re going to look at how your photos affect both the messages you get and how successful your own outgoing messages are. We all know that beautiful people are more successful daters, but let’s quantify by exactly how much.

To illustrate the exact spectrum of looks we’re talking about here, and to put some human faces on our discussion, I want to introduce a few photos of real OkCupid users. Here are two women near the top our range.

And here are two rated in the middle.

As for photos at the bottom of the curve, it didn’t feel right to write someone and say “can I use you to illustrate the concept of ugliness on my blog?” so you’ll just have to extrapolate.

The above featured users have graciously agreed to let me post their pictures, so please don’t make them regret it. Funnily enough, I had to write about a dozen beautiful female users before anyone would even get back to me. Life imitates blog!

Anyhow, I know attractiveness is far from a universal concept, but maybe keep these folks in mind as we go through the data.

We’ll start with a simple line chart. The information I’ll present in this post is not normalized because, as we’ll see, it’s interesting how men and women evaluate looks differently.

Our chart shows how men have rated women, on a scale from 0 to 5. The curve is symmetric and surprisingly charitable: a woman is as likely to be considered extremely ugly as extremely beautiful, and the majority of women have been rated about “medium.” The chart looks normalized, even though it’s just the unfiltered opinions of our male users.

Given the popular wisdom that Hollywood, the Internet, and Photoshop have created unrealistic expectations of how a woman should look, I found the fairness and, well, realism, of this gray arc kind of heartening.

Now let’s superimpose the distribution of actual messages guys have sent:

When it comes down to actually choosing targets, men choose the modelesque. Someone like roomtodance
2/3 of male messages go to the top 1/3 of women.
above gets nearly 5 times as many messages as a typical woman and 28 times as many messages as a woman at the low end of our curve. Site-wide, two-thirds of male messages go to the best-looking third of women. So basically, guys are fighting each other 2-for-1 for the absolute best-rated females, while plenty of potentially charming, even cute, girls go unwritten.

The medical term for this is male pattern madness.

The female equivalent of the above chart shows a different bias:

As you can see from the gray line, women rate an incredible 80% of guys as worse-looking than medium. Very harsh. On the other hand, when it comes to actual messaging, women shift their expectations only just slightly ahead of the curve, which is a healthier pattern than guys’ pursuing the all-but-unattainable. But with the basic ratings so out-of-whack, the two curves together suggest some strange possibilities for the female thought process, the most salient of which is that the average-looking woman has convinced herself that the vast majority of males aren’t good enough for her, but she then goes right out and messages them anyway.

Just to illustrate that women are operating on a very different scale, here are just a few of the many, many guys we here in the office think are totally decent-looking, but that women have rated, in their occult way, as significantly less attractive than so-called “medium”:

Females of OkCupid, we site founders say to you: ouch! Paradoxically, it seems it’s women, not men, who have unrealistic standards for the “average” member of the opposite sex.

Finally, I just want to combine the two charts to emphasize how much fuller the inboxes of good-looking people get. I have scaled this graph to show multiples of messages sent to the lowest-rated people. For instance, the most attractive guys get 11× the messages the lowest-rated do. The medium-rated get about 4×.

This graph also dramatically illustrates just how much more important a woman’s looks are than a guy’s.

Now let’s take a look at how senders’ and recipients’ attractivenesses affect reply rates, not just the number of messages sent.

As you’d expect, more attractive people get more replies. And since they themselves get so many more messages than everyone else, they write back much less frequently. Here’s the graph for female senders, plotted in evenly-spaced “attractiveness groups.”

And here’s the one for male senders.

One interesting thing seems to be going on here: when the best-looking men write the worst-looking women, taste the rainbow,
of self-esteem issues their message success rate takes a big hit. The knee-jerk response would be to somehow chalk it up to hunky spammers, but we very carefully control for that in these articles, and in any event why would better-looking girls be drastically more susceptible to it? It seems to be some kind of self-confidence thing.

As we did before, I’m going to consolidate the line charts to show just how your attractiveness changes how often your messages get responses.

This post has been the preamble to the larger discussion of “what makes a good profile?” We’ve spent a lot of time on OkTrends looking at messages, and since your profile is the other important place you express yourself, we thought it deserved the same treatment.

I wanted to address physical attractiveness right at the start, because obviously it’s a huge factor in how successful your profile is. In the upcoming posts in this series, we’re going to control for attractiveness, so that we can deliver real and useful advice for all the non-models out there.

We’ll look at, among other things: what makes a good picture (is it taken outside? inside? is it full-body? a head-shot? with your pet snake? what?), what kinds of self-presentation will get you the most messages (jokey? flirty? all business?), and how much profile information is too much. Should be good.

Because we run a huge dating site—OkCupid, which is free—and we studied math together, Chris and I like to sit in a circle (a line, really) and look at your data. Normally the data is pertinent and leads to revealing studies of human interaction, like our last post about how race plays a big role in online dating. But this week, just for fun, and to take a break from heavy subjects, we’re going to look at some of the more offbeat numbers we’ve come across—the statistical outtakes, if you will.

All of the following are user-submitted questions from our database. We’ll start with this one:

First of all, it’s nice to see that there are still some users hanging on to “blackguard” as a term of friendly disparagement. Also, we thought it was interesting that gay men lag behind their bi and straight brothers here. I’m sure many of you are already thinking “fafo” to yourselves, so I won’t even make the joke.

We were surprised on a couple fronts here. One, we had no idea that this many straight women were interested in sex with a strap-on. Duly noted, duly feared, ladies. Two, nearly one in seven straight men answered “yes” to this, and even for OkCupid’s sexually adventurous user base, that’s a pretty wild number. My best guess is that since the question doesn’t specify either way, some bros assumed it was asking about strap-on pizzas.

Next we have this incredible table:

Yes, 2 in 5 people (and nearly half of all men!) think they are one in a thousand. You do the math: that’s 100% melted.

Here’s how the U.S. breaks out by state, in one of our color-coded maps. Green means more people than average in a state think they’re geniuses; red means fewer. That bastion of American scholarship, Mississippi, came in green, of course. And apparently almost half of Nevadans are geniuses, which is at once laughable and slightly credible, seeing as how it must take a certain amount of brains to create a hell-on-earth. On the other hand, huzzah to West Virginians for their relative humility.

Finally, we ran a query on suicide for an upcoming article comparing people from Canada to other marginalized Americans. The topic’s certainly less frivolous than the rest of this week’s post—and the Canada/U.S. comparison didn’t turn out to be very interesting—but we felt like we had to publish what we found:

It’s pretty dramatic data. Here’s one way to look at it:

and another:

. . .

Thanks again for reading, everyone. If you’re curious about the dating site we run, or would like to prostrate yourself before a vast pool of online geniuses, check out OkCupid. It’s free and awesome.

Also, we’ve been submitted to the Mashable Open Web Awards, in the Best Corporate Blog category. If you’ve stopped laughing and have the inclination to vote for us, you can do so here.

Welcome back, dorks. We’ve processed the messaging habits of over a million people and are about to basically prove that, despite what you might’ve heard from the Obama campaign and organic cereal commercials, racism is alive and well. It would be awesome if the other major online dating players would go out on a limb and release their own race data, too. I can’t imagine they will: multi-million dollar enterprises rarely like to admit that the people paying them those millions act like turds. But being poor gives us a certain freedom. To alienate all our users. So there.

When I first started looking at first-contact attempts and who was writing who back, it was immediately obvious that the sender’s race was a huge factor. Here are just a handful of the numbers that illustrate that:

The takeaway here is that although race shouldn’t matter in messaging, it does. A lot.

More Compatibility Means More Replies (Normally)

First of all, how do we know that race shouldn’t matter? Are we just making some after-school-special assumption that “true love is colorblind?” No, we’re not: we know race shouldn’t matter to replies because the races all match each other more or less evenly, and reply rate correlates to matching.

On OkCupid you create your own unique matching system, and that means your better matches are people you actually want talk to. Below is a graph showing match percentages vs. reply rates for a random sample of 500,000 people. As you can see, in general, the better you match someone, the more likely you are to reply to a first message from them.

We can see this principle in action when we look at our trusty control, the Zodiac. Here are the match and reply rates side-by-side, with similar rates colored yellow. There’s no real need to inspect the numbers; just observe the similar colors.

• Throughout this post, yellowish colors are short-hand for “neutral” and red and green indicate “strong preference.”

People of the various Zodiac signs match each other all at roughly the average rate, and, as we would expect, they reply to messages similarly. In general, the correlation between match percentage and reply rate means that whenever we compare the match/reply charts for a given breakdown of the population, they should look about the same. However, this, like so many other fine assumptions, totally breaks down when race gets involved:

Again, don’t bother squinting, just check out the colors. We’ll soon look very closely at these tables.

The Race Is On

So here’s last week’s compatibility by race table (I explained how we can confidently measure “compatibility” in that post). This is a blow-up of the leftmost table above:

As you can see, the races all match each other roughly evenly: good news. It means all other things being equal, two people, of whatever race, should have the same chance to have a successful relationshp. But now let’s look at the table of how individuals actually reply to each other’s messages. First we’ll examine messages sent by men to women (I know our gay readers are interested in same-sex versions of these tables, there’s a link to them here and at the end of this post):

The numbers on the perimeter of the table are the weighted average rates for each column/row. Here’s what we can know:

• Black women write back the most. Whether it’s due to talkativeness, loneliness, or a sense of plain decency, black women are by far the most likely to respond to a first contact attempt. In many cases, their response rate is one and a half times the average, and, overall, black women reply about a quarter more often that other women.
• White men get more responses. Whatever it is, white males just get more replies from almost every group. We were careful to preselect our data pool so that physical attractiveness (as measured by our site picture-rating utility) was roughly even across all the race/gender slices. For guys, we did likewise with height.
• White women prefer white men to the exclusion of everyone else—and Asian and Hispanic women prefer them even more exclusively. These three types of women only respond well to white men. More significantly, these groups’ reply rates to non-whites is terrible. Asian women write back non-white males at 21.9%, Hispanic women at 22.9%, and white women at 23.0%. It’s here where things get interesting, for white women in particular. If you look at the match-by-race table before this one, the “should-look-like” one, you see that white women have an above-average compatibility with almost every group. Yet they only reply well to guys who look like them. There’s more data on this towards the end of the post.

Let’s see what happens when it’s the women writing the messages to men.

• Men don’t write black women back. Or rather, they write them back far less often than they should. Black women reply the most, yet get by far the fewest replies. Essentially every race—including other blacks—singles them out for the cold shoulder.
• White guys are shitty, but fairly even-handed about it. The average reply rate of non-white males is 48.1%, while white guys’ is only 40.5%. Basically, they write back about 20% less often. It’s ironic that white guys are worst responders, because as we saw above they get the most replies. That has apparently made them very self-absorbed. It’s interesting that white males do manage to reply to Middle Eastern women. Is there some kind of emergent fetish there? As Middle Easterners are becoming America’s next racial bogeyman, maybe there’s some kind of forbidden fruit thing going on. (Perhaps a reader more up-to-date on his or her Post-Colonial Theory can step in here? Just kidding. Don’t.)

A Last Couple Graphs

These are site-wide answers to a couple user-written match questions. They barely need any explanation: one comments on the other, really. Together they shed more light on the theory/practice schizophrenia of people’s racial attitudes.

It’s Probably Not Just OkCupid That’s Like This

I don’t want anyone walking away from all this thinking that OkCupid users exceptionally horrible mofos. It’s likely that any dating site (and indeed any collection of people) would exhibit messaging biases similar to what I’ve written up. According to our internal metrics, at least, OkCupid’s users are better-educated, younger, and far more progressive than the norm, so I can imagine that many sites would actually have worse race stats. But like I said at the beginning, we’ll probably never know. See you next week.



The same-sex equivalents of this post’s data are here.

This week, we’re going to take a step back from examining messages to your matches and take a look at matching itself. We’ll slice OkCupid’s data on compatibility by religion, race, and other factors, and by the end we’ll have some unsettling conclusions on how people match and interact online. But first, I want to explain something important.

What Does It Mean To “Match” Someone?

All OkCupid users create their own matching algorithms, so when we determine who matches who, we’re just crunching the numbers people give us.
A match percentage between two people is an expression of how well they might get along.
A match percentage between two people is a condensed, yet statistically valid, expression of how well they might get along. 75% is very high, 45% is very low, and 60.2% is the site-wide average. If, for example, a couple match each other 71%, it means they are likely to like each other, based on their own individual definitions of what makes a person cool, sexy, and attractive, not ours. I point this out now so that, below, when we claim that Jewish women are easier to get along with than Christians, you don’t blame us, you blame Jesus.

We discuss matching more at the end of this post, if you’re interested or nerdy. Now let’s move on.

The Zodiac and Other Beliefs

Since he’s a Pisces and I’m a Virgo, Chris and I of course think the Zodiac is total bullshit, and it was very gratifying to have the data bear this out. Here are the grouped match percentages for a random pool of 500,000 users. Astrological sign has no effect whatsoever on how compatible two people are.

We’re showing you this table, as dull as it is, because the uniformity neatly illustrates how beefy our data set is. There are 144 pools considered above, and they all match the mean plus or minus 0.5%. Our next table again aggregates the preferences of those 500,000 random users, but it shows stronger feelings. Red indicates mutual dislike and green, mutual like. For brevity, and because that’s where we have by far the most data, all the tables on this page display data for straight men and women only.

The numbers on the perimeter of the table are the weighted average match percentage, a measure of group likability, for each column or row. Here’s what we see:

• Jews and Agnostics get along better with people. Jewish men, in particular, have an above average match percentage with every religious group. They even match Muslim women better than Muslim men do, which I find both a hilarious irony and a somewhat sad reflection on the fact that Muslim males don’t seem to be doing very well. The data also cast an interesting light on the Jewish people’s history as a persecuted people: the underlying facts indicate an intrinsic mainstream likability, yet Jews have not been, and in some places still aren’t, “liked.” We’ll investigate a similar dichotomy in the second half of this post when we look at matching by race.
• Muslims of both sexes and Hindu men get along worse. Now is a good time to stress that just because a group has low match percentages, even across the board, that does not mean they are bad people. It just means that they’re harder to please. The converse is also true: the above chart is not evidence that Jews or Agnostics are better than the rest of us. Just better liked. In any event, please keep in mind that each individual has designed his own matching criteria, so the poor-matching groups aren’t failing some outsider’s imposed system. Why, for example, Hindu men would match worst with Hindu women is a mystery.
• Catholics are more universally liked than Protestants. While neither Christian group has many extremes of like/dislike, Protestant Christians only truly match well with other Christians. Catholics have above average match percentages with Hindus, Jews, and even Agnostics. Looks like Vatican II is working, guys!

Get Serious, Or Don’t

When we change our question from “What do you believe?” to “How strong do you believe it?” we get a much more orderly color pattern, and we also unlock some of the mysteries of the previous table. Below we plot people by their attitude about religion, as selected on their profile page.

As it turns out, people who hold their beliefs lightly are much better liked, even by people who are themselves serious. Weird huh? While it’s true that the most serious women believers slightly prefer their men to not be “laughing about it”, every other slice of this data indicates that the less serious (or more flexible?) you are about your religious beliefs, the better you get along.
The less serious you are about religion, the better liked you are, even by very religious people.
Please note that when I say “religious beliefs,” I’m talking about the full spectrum of beliefs, from Atheism to Orthodoxy, so don’t take this as anti-god; I also realize that “getting along” is hardly the purpose behind most people’s theological attitudes. Nonetheless, I think it’s interesting that even a man who’s “very serious” about his religion and has presumably designed his matching algorithm around this fact is still more compatible with the women who are laughing about it.

This information goes some way in explaining our first religion table: in our data pool, Muslims and Protestants tend to be more intense about their beliefs than the others, and Jews and Agnostics are by far the least serious. Here’s the first chart, replotted to include overall seriousness in blue.

Race

Ah, race. If religion is a minefield, then race is a field that’s just one giant mine. But luckily, our match-by-race table isn’t nearly as, well, colorful as the religion ones.

As you can see, there are slight matching biases here, but nothing too dramatic. It’s not going to make many people excited to hear that, for example, white people tend to be better liked, (or, if you want to think reciprocally, do more liking) than the other races, or that black and Indian men are less liked/liking, but, still, those differences are small compared to what we saw with religion. In addition, it’s entirely possible that most of the discrepancies might be just reflect different religious attitudes across the races.

More than anything this table shows the overall compatibility of all races—indicating that in a perfect world, yes, we could all just get along. Yet we don’t. And, in this way, it marks the perfect transition point in our discussion. In the real world people largely choose who to get along with, and even who to get to know.As I said in the beginning of this post, match percentage is an excellent predictor of how well two people might get along; however, in the real world people largely choose who to get along with, and even who to get to know. In online dating, we can measure this choice by looking at how often people reply to actual messages from people of the various races, and then contrast that rate with the underlying compatibilities. And that’s exactly what we’ll do in the second half of this post, which will be up next week. Look once more at the match-by-race chart above and then look at the reply-rate-by-race table below.

It’s a glimpse at the jagged terrain where we’ll be going:

Addendum, If You’re Interested: “Match Percentage”

We all know what it feels like to meet someone you really like, but, unfortunately, feelings are something web servers have trouble with. Therefore, our first goal with OkCupid was to quantify this elusive idea of “compatibility” so we could accurately suggest users to each other.

It’s not as simple as saying, Mary really likes hockey and Bob really likes hockey

It’s not as simple as saying, Mary really likes hockey and Bob really likes hockey, therefore they are a good match—which is how many dating sites work. What if instead Mary really likes being dominated during sex? If Bob also needs to be dominated, and good sex is important to them, Bob and Mary are terrible matches. In bed, at least, they both want their opposites.

This, and other thought experiments, eventually led us to a definition of compatibility that’s user-defined. After all:

• You’re great in all kinds of ways we don’t understand.
• You have specific needs we can’t possibly categorize.
• You don’t want our advice, you want to meet people you’ll like.

In short, our method is this: we host an ever-changing database of user-submitted questions, covering every imaginable topic, from spirituality to dental hygiene. To build their own match algorithms, our users answer as many questions as they please (the average is about 230). When answering a question, a user also picks her how her ideal match would answer and how important the question is to her. It’s very simple, and it removes all subjectivity on our part. We simply crunch the numbers.

OkCupid is no more responsible for people’s match percentages than Microsoft Excel is responsible for their net worth.

So, for example, if two people match each other 69%, what it means is that they are very likely to like each other, based on their own definitions of what makes a person attractive, not ours. OkCupid is no more responsible for people’s match percentages than Microsoft Excel is responsible for their net worth. Again, our users write the match questions, choose which ones to answer, and determine how important each answer is. We just do the math. very detailed explanation of exactly what math we do is in our FAAAQ.

We had many requests for the same-sex equivalents of last post’s charts, and we’d like to take a moment to discuss those. We didn’t include them in the original because they would’ve doubled the data we presented, which in our opinion would’ve made for an overwhelming number of charts and figures.

In any event, here they are in detail:

And here they are side-by-side with the straight charts from the first post:

Remarkably, the women-contacting-women curve in the upper left is close in shape to the men-contacting-women one in the upper right. And the two charts in the lower row, showing men-to-men and women-to-men, respectively, also share a similar curve. It seems that women, both gay and straight, respond better as messages get longer, while men, regardless of orientation, get turned-off as messages drone on. Apparently, reading style transcends sexuality. I’d be interested to hear your theories on why this is so.

The overall response rates are much better for same-sex messages (41% vs. 32% for straight messages). I ran these new numbers through last week’s message-efficiency calculations, and as it turns out, because of the low slope (flatness) of the first half of both graphs, the most efficient conversation starters for both kinds of same-sex contact is only 50 characters.