Data mining the heart

How do we choose a mate? What scientists are learning from online dating.

To be single these days is to face a sea of advice about how to attract a partner. Men are attracted to youth and beauty; women are attracted to wealth and prestige. Or are they? There’s no shortage of impassioned opinion about what men and women want, yet there is little real evidence to support it. Even though finding love is one of our primary preoccupations, it has always been shrouded in mystery and guesswork. Adages like “opposites attract” feel comforting, but it would be even better to know what qualities actually entice potential partners in the real world.

To really answer the question in a scientific way, we’d need to be able to observe the behavior of thousands of single people and see whom they choose to pursue and whom they pass over. We would need a peephole into the dating world.

As it turns out, for the first time in history such a thing exists: It’s called online dating. Research presented last week at the annual meeting of the American Sociological Association found that 22 percent of heterosexual couples surveyed met online, and researchers believe the Web could soon eclipse friends as the primary means of finding mates. As dating interactions have moved from the privacy of bars and social gatherings to the digital world of websites and e-mails, they are generating an unprecedented trove of data about how the initial phases of romance unfold. Online profiles contain detailed personal and demographic information about website users, and their interactions are indelibly recorded in digital form.

Unlike participants in a dating research study, online daters are behaving candidly, not modifying their behavior for an audience. “It gives us a window into the difference between what people say they want and what they actually do,” says Andrew Fiore, a social psychologist at Michigan State University.

This mountain of information is beginning to yield intriguing findings. The dating website OKCupid has begun publishing statistics about its users’ behavior on its blog, and using the numbers to generate real-world advice. For example: Men get more responses from women if they don’t smile in their profile pictures, and women find most men below average in attractiveness — but write to them anyway. More recently, the site has begun inviting collaboration with academics to do more thorough studies with the data. And in the past few years, several other researchers used data from other online dating and speed dating companies to uncover insights into what makes men and women actually respond to each other. The sheer number of interactions makes it possible for the first time to get a detailed look at how different characteristics — weight, height, race, income, age, appearance, and political leanings, to name a few — influence a person’s ability to get a date. Researchers have found, for example, that a man needs to make several extra tens of thousands of dollars to compensate for being an inch shorter, and that race matters more than people admit.

All of this information promises to give singles advice based on real evidence rather than anecdote. But it also raises questions about how much we can learn about the intricacies of individual relationships by taking a bird’s-eye view of the dating world. Does the opportunity to catalog the flirtations of thousands of daters really tell us what makes two people choose to be together?

The moment of attraction between two people used to be an unpredictable and mysterious event: a mutual spark felt in a glance exchanged at a party, a spontaneous meeting at a coffee shop, or a connection blooming gradually through work or a social network. But the Web has streamlined the dating process, concentrating singles into enormous online pools with a structured framework for searching and interacting. You can now go looking for attraction as easily as you can shop for shoes. The result is a dating scene based more on searchable features than on je ne sais quoi. Before that first glance is ever cast in person, two people often already know each other’s age, height, religion, political leanings, hobbies, and favorite bands. They have usually exchanged e-mails and decided whether or not the other person is worth pursuing. And unlike a private conversation in a bar, all of it has been recorded.

OKCupid has led the way in plundering this information for insight into what makes one human being attracted to another. The free dating and social networking website was founded by a group of former math majors at Harvard who developed a series of personality quizzes to match users based on compatibility. It now publishes a blog, OKTrends, that delves into its database of more than 1 million users to analyze their interactions. (The company has been careful to remove personal identifiers from the data so that individuals are anonymous). Sam Yagan, cofounder and CEO of OKCupid, says that OKTrends grew out of the company’s own interest in statistics and its brand’s emphasis on putting faith in algorithms to find romance.

Recent posts have tackled topics from appearance to race. OKCupid asks users to rate other people’s photos, which gives the company a measure of who’s good-looking and who’s not. The company found that while men rate women’s attractiveness in an even curve — most women being average — two-thirds of men’s messages go to the best-looking third of the women. Women, on the other hand, are more harsh on men, rating the majority as below average, but are more likely than men to send messages to people they don’t find attractive. The blog has also uncovered some intriguing trends about lying: In their online profiles, for instance, all users add an average of two inches to their height and a 20 percent raise in salary.

The data debunk some dating myths. In analyzing 7,000 user photos, the company found that women get more male attention when they flirt into the camera or smile, while men, surprisingly, did better when they looked away from the camera and didn’t smile. Even more surprising, not showing their face in their photos didn’t affect the number of messages users received.

The company can also analyze user communications to see what works and what doesn’t. The word choices most likely to kill your chances of getting a reply are bad grammar and “netspeak,” like saying “ur” for “your.” Physical compliments also appear to be counterproductive: Messages that used words like “fascinating” and “awesome” fare much better than those that used “sexy,” “beautiful,” or “hot.”

Most other companies keep their information secret, but a few academic researchers have been able to study it more deeply. Dan Ariely, a behavioral economist at Duke University and author of “The Upside of Irrationality,” used data from a different website (he promised the site’s owners anonymity) to analyze which attributes determine the success of men and women in getting responses from the opposite sex. For both sexes, the attractiveness of the photo was the most important trait (users’ profile pictures were rated separately to determine how good-looking they were). Beyond that, men’s height was the most important feature to women. In fact, the researchers were able to put the value of height into numbers. By comparing height to salary, they found a man who is 5 feet 9 inches tall needs to make between $35,000 and $40,000 more per year to get as many responses as a man who is 5 feet 10 inches tall.

For men, they found, a woman’s most important feature was body mass index. “It turns out that men like women who are slightly anorexic,” Ariely says. And unlike a man’s height, there’s no amount of money a woman could earn to offset the effect of higher weight. Pursuing degrees doesn’t help either; education beyond a bachelor’s degree for women or a master’s degree for men did nothing to increase desirability. Another surprise: Smoking cigarettes actually increases a woman’s popularity on dating sites, which Ariely speculates may be because men associate smoking with promiscuity.

Fiore and colleagues at University of California at Berkeley looked in detail at the profiles and messaging behaviors of online daters on a major (anonymous) American dating website. They confirmed some conventional gender roles: Men tend to look for younger women, while women look for older men; women were pickier than men about what they were looking for; and over 75 percent of messages were from men to women. They also found that responsiveness matters in messages: The faster someone replies to an initial message, the more likely he or she is to get a follow-up message. Fiore has also been one of the few researchers to study what happens in the weeks after online daters actually meet. He found that daters tend to like one another less after meeting. The judgments they formed about each other before meeting didn’t predict who would continue to date, though they sometimes predicted the quality of the relationship in those who formed one.

Psychologists and social scientists have studied relationships for decades, but their ability to do so has gotten a boost from these new sources of data. Typical relationship studies in the academic world involve recruiting volunteers — usually college students — for an experiment or survey. In contrast to this rather artificial approach, users of online dating sites are acting in their own self-interest, unaware that they’re being evaluated. “The behavior is honest,” Yagan says. That’s especially useful in investigating touchy subjects like race, because “what people say they believe and the way they act are not the same.”

Indeed, some of the strongest findings from multiple dating sites have revealed discrepancies in how we talk and act in relation to race; while most people profess to being colorblind, the numbers show something quite different. Race influences the number of responses aspiring daters get, with white men having the clearest advantage.

Dating companies offer something else that researchers salivate over: large sample sizes. “You can end up with just gorgeous data sets that would be pretty much impossible for a researcher to get on their own,” says Eli Finkel, who heads the Relationships Lab at Northwestern University. He explains that having thousands of dating interactions lets researchers study “low-frequency events,” such as the behaviors of minority groups. He has used data from speed-dating companies to analyze the effects of race and political orientation on daters’ preferences. (He found that among white users, liberals are more likely than conservatives to date a black partner, whereas among black users it’s conservatives who are more likely to date outside their race.)

“Even if you don’t care about dating,” says Neil Malhotra, a political economist at Stanford University who is looking at using OKCupid’s data in his research, “it’s a really nice laboratory for looking at human interactions generally.”

These websites yield beautiful statistics because they let people categorize themselves according to a set of defined attributes. Which points to a potential problem with studying online dating. By setting up romantic interactions as a marketplace, online dating sites may actually fuel the somewhat superficial behavior seen in some of these analyses. In one study, Ariely and his colleagues tried to quantify how valuable people were in the dating market; they then created an algorithm that artificially paired people with others of equal “market value” according to these factors. The real-life behavior of daters corresponded very well with the ideal scenario they created; given a list of characteristics, daters are very good at figuring out where they fit in the social hierarchy and seeking out those who match their own value.

But is that necessarily a good thing? “Within this system — and this system is about searching on these attributes — people function quite well,” Ariely says. “But it doesn’t say they’re really doing the optimal thing. It doesn’t say that they understand who will be their soul mate.” Instead, online dating may create a situation in which people can methodically shop for the best “match,” but miss out on the qualities that lead to good relationships when the interaction moves off-line. Studying these sites may tell us who catches our eye in a profile, but it takes more to tell us who makes us happy in the end.

Ariely’s interest in dating as a market grew out of a personal tragedy. When he was 18 years old, he was badly burned in an explosion. During his long and painful recovery, he realized that his status had dropped; he would no longer be as valuable to potential mates because of his disfigurement. He worried he would have to settle for someone less desirable who would accept him. Not many of us think about dating as a market — a place where buyers and sellers compete for transactions. But, as Ariely says, “The moment you understand that the other people you’re interested in have other options, you understand that you’re in a competitive situation and you’re in a market.”

But while it’s useful for researchers to view dating as a marketplace, Ariely admits that in our own lives, that line of thinking can be harmful. At some point, a relationship isn’t about trading up for the best mate, it’s about falling in love with another person. And that involves seeing qualities in someone beyond the superficial ones that catch people’s eye in the dating market. For all his concerns about his own market value, Ariely ended up happily married — and he didn’t have to settle.

Courtney Humphries is a freelance science writer and the author of ”Superdove: How the Pigeon Took Manhattan…And the World.”  

__________

Full article and photo: http://www.boston.com/bostonglobe/ideas/articles/2010/08/22/data_mining_the_heart/