Okay, Cupid! How to Bring More Diversity to Online Dating

We carry our biases online. But we can also, new research says, overcome them.


In 2002, Wired made a prediction: "Twenty years from now, the idea that someone looking for love won't look for it online will be silly, akin to skipping the card catalog to instead wander the stacks because the right books are found only by accident."

As more and more people look to algorithms to play the matchmaking roles traditionally filled by family and friends, Wired's looking more and more prescient. There's OkCupid, the free dating site with over 7 million active users that's striving to be, in various ways, the Google of online dating. And there's Match.com. And eHarmony. And all the other sites, from the mass to the veryvery niche, that promise to connect people online in a much more efficient way than they could ever be connected by the vagaries of IRL circumstance. Which is a good thing (arguably) not only for the increasing number of people who are meeting each other ... but also for the academics who study their behavior.

"We have a remarkably impoverished understanding of what people care about in mate selection," says Kevin Lewis, a sociologist at Harvard, largely because the only large data sets previously available for analysis -- public marriage records -- don't actually contain much data. Marriage records note racial backgrounds and religion, Lewis notes, but not much more than that -- and they definitely lack information about the personal qualities that create that notoriously unquantifiable thing we call "chemistry."

For his dissertation research, Lewis got ahold of a large selection of OkCupid's trove of data, which contains information not only about user demographics, but also about user behavior. The (anonymized) info allows for analysis, Lewis told me, of contacts made from one user to another -- and of contacts not made (and, ostensibly, decided against). It highlights dating preferences expressed not against the constraints of real-world social structures, but against the expansiveness of potential partners online. With the data set, Lewis has been able to do what's been so hard for sociologists to do previously: to disentangle preference from circumstance.

One of Lewis's most intriguing findings has to do with what his (as yet unpublished) paper calls "boundary crossing and reciprocity" -- that is, the initial message from one user to another, and the reciprocation (or lack thereof) of that message. There's a big difference, Lewis found, between contacting someone on a dating site ... and replying to someone who has contacted you. It turns out, first of all, that many of the biases we have in the real world replicate themselves online. Homophily -- the old "birds of a feather" phenomenon that finds people seeking out those who are similar to them -- is alive and well in the online dating world, particularly when it comes to race.

But: There's an exception. While homophily is a big factor in terms of determining whether a user sends that initial message -- you're much more likely to reach out to someone of your own racial background than you are to reach out to someone of a different race -- similarity can actually hurt your chances of receiving a reply. And diversity, for its part, can help those chances. Here's how Lewis's paper puts it:

Online dating site users tend to display a preference for similarity in their initial contact e-mails but a preference for dissimilarity in their replies. And in fact, the reciprocity coefficients are indeed significant in precisely those cases where the boundary for an initial contact message is the strongest: While any two users of the same racial background are significantly likely to contact one another, reciprocated ties are significantly unlikely between two users who are black (p<.01), two users who are Indian (p<.01), two users who are Hispanic (p<.05), and two users who are white (p<.05)--and so by extension, reciprocated ties among two users from different racial backgrounds are comparatively more common.

This is fascinating, and not only as a data point -- one that, Lewis points out, deserves much more research in future work -- but also as a kind of morality play in miniature. We may, yes, carry our biases with us to the digital space; but there's an easy way to overcome them, it seems. And it starts with a simple hello.

Image: Shutterstock/Robert Zywucki.