After reading yet another story about how online dating can not predict who will have a romantic relationship I thought I’d say something. Its clear from reading the paper and these posts that Eli and his team do not know what POF is actually doing when it comes to matching. What we are doing is completely different than everyone else.
1. Similarity and other Psychological constructs are not used in our matching system. This is something a Psychologicist would do if they were responsible for making a matching system. In my opinion Psychological approaches to matchmaking are next to useless, they offer relatively small improvements over chance from all the testing we have done. We’ve hired the best Psychologicist’s to try and create a matching system and than validate it using our couples data. In the end Psychological evaluations and matching turned out to be completely useless. So we agree with Eli and team there.
Back in 2007 I went on the today show and revealed that I created a behavioral matchmaking system that helped match people up. Almost a year later Match.com hired a bunch of people to try and copy what i was doing and that seems to be their system today. This is a great system for getting people to send messages and in match’s case optimize revenues but it is a poor system for predicting relationships.
Given that behavioral matchmaking and recommendation systems were not powerful enough I started getting people who left the site to tell me who they entered into a relationship with. Using MILLIONS of people who left the site in a relationship we could very quickly determine what kinds of people got together in a relationship and with who. After about 6 months 45% of the relationships split up giving us a very clear picture of what kinds of traits work in relationships and which ones don’t. Keep in mind we make people answer hundreds of personality questions and we have all the data they entered on their profiles along with other stuff we collect. Using simple predictive models we are able to tell who you are most likely going to start dating with a very high degree of accuracy. If you do enter into a relationship with someone and it is a unstable relationship we could probably tell you the week in which your relationship will end.
Given what i’ve stated above there is no way that Eli and his team can argue that ”a mathematical formula can not identify pairs of singles who are especially likely to have a successful romantic relationship” Predictive modeling works in every other industry imaginable Online dating is no different. In short sites Like Eharmony, PerfectMatch and Chemistry are designed by Psychologists who think the world should work in a certain way and they build a matching system that enforces that worldview. At POF we just find successful couples and lets them be the basis for the matching system.
The following graphs the probability of a relationship breaking up. In order to be better than chance at predicting long term relationships all we have to is make sure we do not generate matches that lead break ups. Last year we could predict which relationships will fail with a greater than 80% accuracy. The relationships shown here are not the result of a matching system but relationships that randomly happened on POF. The sample size is around 21,000 and skewed to heavily to under 35. The way to read the graph is 42% of relationships formed on the week of 2/20/2011 are still in tact on Feb 13th 2012. That is to say neither the male or female came back to the site to signup for a new account.