OkCupid Experimented on Me and That’s OK

With over 3.5 million users, OkCupid is a treasure trove of data. If you enjoy human interactions as much as you enjoy studying human interactions, it's not so bad.


Matchmaking social site OkCupid sent me a mysterious message one day claiming their super computer misstated my match percentage with another user. That is, the amount of answers to match questions we shared in common were inexplicably wrong, and they wanted to correct their seemingly anomalous mistake. As it turns out, OkCupid did it intentionally, and they are hardly sorry — in fact, that message was probably one of many sent out to users who were involuntary participants in the social site’s experiments.

As outlined on OkTrends — OkCupid’s data blog that shares statistical information on user behavior — the site, along with many other websites, regularly subject their users to questionable social experiments in order to gain knowledge on the quality of services rendered, either for users or advertisers. For OkCupid, this meant helping users reach their ideal mates — even if it meant lying to them.

Facebook made similar admissions recently by finally sharing its emotional experiments on users. Unfortunately, all social sites are using personal data in order to reevaluate or even test out its user experience as part of its own ongoing “science” experiments. Participating and using sites like OkCupid allows its robots to sift through massive amounts of personal data in order to find something users haven’t found: love.  One can only imagine how terribly wrong that can get, but it didn’t actually alter the end results very dramatically.

For example, in my case, as in many other users, OkCupid’s special match percentage ended up being not as important as one would expect:

The ultimate question at OkCupid is, does this thing even work? By all our internal measures, the “match percentage” we calculate for users is very good at predicting relationships. It correlates with message success, conversation length, whether people actually exchange contact information, and so on. But in the back of our minds, there’s always been the possibility: maybe it works just because we tell people it does. Maybe people just like each other because they think they’re supposed to? Like how Jay-Z still sells albums?
† Once the experiment was concluded, the users were notified of the correct match percentage.
To test this, we took pairs of bad matches (actual 30 percent match) and told them they were exceptionally good for each other (displaying a 90 percent match.)† Not surprisingly, the users sent more first messages when we said they were compatible. After all, that’s what the site teaches you to do.

As it turns out, users were more likely to have the odds of exchanging at least four messages if they were both very compatible and were told so. Ultimately, OkCupid users will find compatibility based on their actual match percentage, but when lied to, they will take an initiative to send the first message to another user because that’s what the site is designed to do. Without enough compatibility, messages and interests tend to wane.


Prior to opening my OkCupid account, I was an avid reader of their data blog, OkTrends. Unlike Facebook, OkCupid’s data practice was semi-transparent and even downright educational: I read subjects that ranged from racial studies (Real Stuff White People Actually Like) to helpful tips for users (Exactly What to Say in a First Message). In fact, it can be downright revelatory:

Whites respond by far the least to anyone. Both white lesbians and white gay men write the fewest replies. In fact, across the two charts, whites respond about 15 percent less often than non-whites, and white gay men show a marked preference for other whites. On the other hand, gay white women don’t have the segregationist tendencies of their straight counterparts; they just dis everyone. Whereas last week we saw that straight white women strongly preferred other whites to the exclusion of other groups, lesbian whites respond to all nine racial groups roughly evenly, and, in general, the lesbian community seems relatively colorblind. Only Indian lesbians receive a response rate far off the average, and as I said above, the sample size there is limited and the results might be skewed by chance.