Data Love: A Night With the Man Behind OkTrendsBy Kieran Dahl February 13th, 2015
Two days before Valentine’s Day, I found myself below the streets of Brooklyn listening to a man I’d never met talk about love.
I was in a subway station, but the man in question was not the resident late-night subway-platform philosopher, and no trains had gone through this station in years. I was, in fact, at the New York Transit Museum, housed in a refurbished subway station in the Brooklyn Heights neighborhood. And I had entered these subterranean depths to hear Christian Rudder, the founder of OkCupid and the writer of the wildly popular OkTrends blog, provide insight into love, data, and the creation of a website that now generates 30,000 first dates every day.
I may have gone to the event to try to get dates with women who consider content marketing data analysis to be romantic, but instead I wound up learning some really interesting things about how OkCupid became successful.
OkCupid was founded as a middle ground
“All a dating site is,” said Rudder at the beginning of his talk, “is a way to recommend strangers to one another.” But back in the mid-aughts, Rudder felt unsatisfied with how other dating sites were introducing people to each other.
“There was, on one hand, the kind of Match.com-style idea, which was like, ‘You guys are both 25, and you’re a man and she’s a woman, and you both live in Brooklyn, and there you go,'” he said. The options were too impersonal and offered little guidance.
On the other end of that spectrum was the eHarmony model. eHarmony required—and still requires—you to answer a ludicrously long questionnaire to determine your matching preferences. The problem, beyond its egregious length, was that the questionnaire was designed by the company, not by its users. If you didn’t fit eHarmony’s platonic ideal of a single person, “you were kind of screwed,” said Rudder.
So Rudder decided to merge the two extremes. OkCupid’s questions are now modeled on real-life interactions, in which your answer reflects how you’d want a compatible partner to answer the same question.
User submissions generate lots of garbage
In 2015, many, many people on the Internet say stupid things with relish. If you go to the comment section of even a mildly contentious article, there’s a 100 percent chance you’ll find a firestorm of idiocy and vitriol.
When OkCupid was in its infancy in 2004, Rudder didn’t have the luxury of knowing what crowdsourcing people’s opinions would bring. He thought he might have people submit their own questions. It wasn’t the right move.
“I don’t know if any of you work for websites that let users submit things,” he told the crowd, “but you’re inviting a huge global warming tidal wave of sludge into your office. It’s mostly garbage.”
Rudder needed a system—an algorithm—to triage the submissions. And the first questions you answer on the website needed to be divisive. “Have you ever murdered someone?” was a popular question, and a fair one, but it wasn’t quite right. Questions like “Do you want to have kids soon?” and “Do you believe in God?” had the right mix of divisiveness and a lack of morbidity.
OkCupid at first demonstrated “the focus problem”
After the site was launched, users signed up steadily, but they would often come and go. This was an iteration of what Rudder called “the focus problem,” a result of not controlling for attractiveness.
Initially, men were only clicking on a small subset of very attractive women. The hottest women—“almost always women,” Rudder said—got flooded with messages and left the site out of frustration. In turn, men didn’t get responses, felt unenthused about the efficacy of the service, and either left OkCupid with their tail between their legs or sent a barrage of messages to the second tier of hottest women. The cycle would then repeat itself.
As Rudder put it, “The women were besieged, the men were lonely, and everybody went their separate ways.”
OkCupid’s early rating feature was ripped from Hot or Not
A later feature allowed you to rate people out of 10 stars, with looks and personality accounting for five stars each. The problem was that men often equated looks with personality—but never the reverse. In other words, women in bikinis were getting five-star personalities.
Rudder quickly got rid of that feature.
Men rate women very similarly across dating sites, but women rate men very differently
On OkCupid, Match.com, and Tinder (all owned by the same company), men rate women on what is approximately a bell curve—lots of threes, a few fives, a few ones.
“I guess this contradicts this idea out there that men have unreasonable expectations,” Rudder laughed.
The data curves for women rating men across the three sites, however, look very different. On Tinder, women rarely rate men highly; the vast majority of women swipe “no.” But On OkCupid and Match.com, there’s some overlap with how women rate men; the bell curves come close to lining up.
Men are dogs
Rudder mentioned a researcher who got a good-looking man and a good-looking woman to approach people of the opposite gender, strike up a conversation, and, ultimately, ask a question. When that question was “Do you want to go on a date?” men and women demonstrated about the same amount of selectivity: around 75 percent said yes.
When the question was “Do you want to go back to my apartment?” the responses differed by gender. “The guy goes up to the women,” said Rudder, “and they’re like, ‘No.’ He got something like 5 percent.” The attractive woman, on the other hand, got around the same amount of positive responses as with the first question, “something like 70 percent from the guys, who were presumably like, ‘Yeah, I thought that’s what we were talking about the whole time.’”
And when the question was “Do you want to go have sex with me?” every single woman turned down the attractive guy. The attractive woman, meanwhile, still had a 50 percent yes rate.
Despite OKCupid’s algorithm, randomness is still very important
Rudder’s favorite part of the matching algorithm is a random number. “A random number is sort of the mathematical embodiment of serendipity,” he said. While Google, for example, is deterministic—you search for something and the results will be the same every time—a dating website can’t operate like that.
“We would show you the same fifty attractive people every time and you’d think that’s all the people in New York City,” Rudder joked.
Men say they want women around their age, vote one way, and act another way entirely
When you ask men the ideal age of a female match, they’ll give a number fairly close to their own age. A 39-year-old man will say he wants a 42-year-old woman, for example, or a 33-year-old. (Women are slightly more age-tolerant.)
The votes tell a different story. The SparkNotes (which Rudder also founded) version of that story is that men want 20-year-old women. Forty-year-old men vote highest on women in their early 20s, as do 30-year-old men.
But, perhaps privately knowing that such a desire is frowned upon, men don’t necessarily message the college-aged women they match with. A 40-year-old man will eagerly message a 30-year-old woman—but once he turns 41, he’ll start messaging 35-year-olds.
The Boost feature actually works
“The one thing that actually works is the Boost feature, which costs like two dollars,” Rudder said. “If you pay it, you really do get a lot more attention—basically twenty-four hours of attention in fifteen minutes.”
If data is really surprising, there might be something wrong
When asked if he’d found any surprising dating data, Rudder responded, “Probably not.” He then added, “The more time you spend on this stuff, the more it confirms certain things about people.”
Your gender is more determinant of dating trends than your sexual orientation
The data for lesbian women and gay men is “very similar” to that of straight men and women, said Rudder. “Gay men are very age-motivated, just like straight men, while gay women are more even-handed.” In other words, our OkCupid actions have less to do with your sexual orientation than they do with your gender.
An audience member said, “What I’m taking away from this is that you have to stand in a bikini and change your age.” Rudder evidently had gotten a similar response before.
“It doesn’t behoove you to spoof your profile if you still have to roll your own ass to the date,” he said. “Just be yourself.”Image by g-stockstudio/Shutterstock