In my last blog post I claimed that taxi dispatchers faced a fundamental problem: how much to favor drivers versus passengers. The specific example that I gave was that drivers do not want to give rides to far off destinations because they would have to waste gas to get back to the city center and because it would take them longer to find the next ride. I claimed that Uber, the mobile app for hailing cabs, favored its passengers because it did not ask them were they were going before they requested a ride.
Today I had the pleasure of meeting the Uber data team and being proven wrong. It turns out that giving rides to far off destinations tends not to hurt drivers but to actually help them. Uber compensates drivers extra for far away rides and a long ride is likely to actually minimize the downtime of a driver because long rides are continuous drives. Furthermore, some rides to uncommon locations actually help the Uber ecosystem because they allow for quick pick ups from those places which typically do not have a lot of demand.
The mistake I made in the previous post was that I theorized without data. I made some implicit assumptions about how Uber prices rides and the geospatial distribution of demand, which led to the wrong conclusion. I worry that what I did, theorizing without data, is done too frequently, especially in economics (i.e. the Journal of Economic Theory)1. For people with analytic ability, theorizing comes naturally. On the other hand, finding pertinent data, doing research and talking to experts is hard no matter who you are.
The upside of being empirical is that you learn about new, interesting problems that you would have never thought of daydreaming. For example, Uber has a related problem that I will be thinking about in the future: sometimes all rides on Uber move in the same direction and cause a mass exodus of drivers away from the points of initial demand. What is the best way for Uber to balance the market in those situations?
I think that the enterprise of economic theory is very worthwhile, especially if it is well motivated by existing empirical facts. ↩