I recently went to the NBER IT & Digitization Summer Institute. I was there to do a poster session on the new ridesharing bias paper I am working on with Jorge Mejia. We also wrote a short LSE Business Review post summarizing the work for those that don’t want to read the entire paper. It’s work I’m pretty proud of and was part of the poster session, but that’s not the main purpose of this post. What I really want to do is share a quick summary of six papers that I thoroughly enjoyed. These are not the only papers I like, but just a few to give you a flavor of the type of work I see and enjoy. Also, not in ranked order but in order of appearance on the conference schedule. Ok, legalese over.
The Gender Earnings Gap in the Gig Economy: Evidence from Over a Million Rideshare Drivers looks at the root causes of the earnings gap. Turns out part of the reason men earn more when driver for Uber is because they drive faster. Anyone want to try teasing out the effect of this on accidents and fatalities?
The Impact of Big Data on Firm Performance: An Empirical Investigation uses Amazon data to find out how scaling to more products or more time periods can lead to improvements in forecasting accuracy. One thing I found particularly interesting is that several thousand products, even if they have reasonably sporadic demand, is enough to back out the seasonal patterns of the category and hence you don’t get much more benefit from adding more and more products.
Death by Pokemon Go: The Economic and Human Cost of Using Apps While Driving uses the introduction of Pokemon Go to find out the cost of using apps while driving. People died as a result of Pokemon Go’s introduction. Wow. The great thing about this paper is the careful attention the authors paid to causal identification. They constructed a difference-in-differences estimator by look at pre/post introduction of Pokemon Go in intersections where there was a PokeStop. Just brilliant work.
The Marginal Congestion of a Taxi in New York City finds that the introduction of a new taxi service in NYC decreased driving speeds in the local area by around 8-9%. The interesting part here is that they have a very clean identification strategy because of limitations on where these particular taxis could and could not pick up rides. They also count taxis in the area using a large number of aerial photographs which sounds like a massive effort to hand code.
Technology, Incentives, and Service Quality: The Case of Taxis and Uber shows that the reduced moral hazard on Uber (relative to taxis) means that Uber drivers takes 7.4% fewer detours. The savings to customers is a cool $40k/day. Not too shabby.
Large-scale Demand Estimation with Search Data tries to tackle a big problem for online retailers: how do you estimate demand when you have so many products? The authors develop a model that uses customer search data to supplement sales data. What I liked the most here was the practical nature of the problem; the authors focused on the computational demands of estimation making this much more likely to be used in the wild. It reminds me of a project I’m working on estimating demand for bundles of products but it’s not ready for the public eye yet so I’ll leave you on the edge of your seat.