Yesterday, the wife wrote a nice blog post on the power imbalance that mutual-like based matchmaking services like Tinder create – since they notify the mutual like immediately after the second counterparty has liked, it is evident to both who liked whom first. And that creates an inherent power imbalance in the relationship, even before it has begun.
She goes on to say that it’s better if these services waited a random amount of time after the second like in order to notify the mutual like. That way, it would be unclear who liked whom first, and that can lead to a more balanced relationship. Here I intend to put on my quant hat and quantify how this random wait period should be determined.
First of all, it must be quite clear why this wait period needs to be random – if it is deterministic, users can be able to back-calculate (after they’ve gotten a few data points themselves) and determine who liked first. Randomness eliminates any such possibility of back-calculation.
Now that this is established, what should the random wait period be? Should it be a minute? Or an hour? Or a day? And what kind of distribution should the random number be drawn from?
The thing with apps such as Tinder is that they match people at different time scales. The intent is for instant match – so that two people who are checking out each other at a bar or a wedding hall can know that mutual like exists, and do something about it while they’re still in close proximity.
In a market such as India, however, where Tinder has still not built up much liquidity, some matches are “offline” – I check you out and “like” you today. You come across my profile tomorrow and like me back. Notice that this time scale is an order of magnitude higher than the bar/wedding timescale. So the random number generator needs to take care of this.
If the random wait period is a lot number, it might work well in the bar/wedding case, but not in the offline like case – if i’ve liked you yesterday and get notified today, I know I’ve liked you first. A high random wait period, on the other hand, can kill the bar/wedding market – there is no use that there was a mutual match at the wedding you attended yesterday, for there’s nothing you can do about it now.
So the solution is rather simple – the random wait period should of the same order of magnitude of the time difference between the two likes. If the two likes have happened minutes apart, the wait should also be minutes apart. If hours, then hours, and so on. The best way to achieve this is to use the time difference between the two likes as a parameter in the random number generation.
And what kind of a distribution should be used? I argue that the distribution should be something closer to an exponential distribution than something closer to a normal distribution. This is because the latter has a long “left tail” as well, and that can be used by users to double guess the minimum possible time before the notification that the last like happened, and know the power balance. Exponential distribution, on the other hand, doesn’t have much of a left tail, and can fuck better with the minds of anyone who tries to double guess.
Moreover, the exponential distribution can be described with one parameter. And we want the wait period to be a function of the time difference between the two mutual likes. It should be easy to build this in without further assumptions!