Sunday, 18 February 2018

Uber as a substitute for ambulances

When a new, cheaper substitute to an existing good or service becomes available, we can expect the demand for the existing good or service to decline. Importantly though, 'cheaper' refers to the full cost of obtaining the good or service, not just the monetary cost. So in the case of ambulance services, the cost may partly be monetary (such as in the U.S.), but also includes the cost of waiting to receive treatment (which is probably a more important component of the cost in New Zealand, although in some cases ambulance services will charge for a call-out. For example, see here).

Are there cheaper substitutes for ambulances? In a country such as the U.S., where the monetary cost of calling an ambulance could easily be in the thousands of dollars, ride-hailing services like Uber or Lyft could easily be a less costly option overall. However in New Zealand, where the monetary cost of calling an ambulance is not as high, the evaluation is not quite as straightforward. If you call an ambulance for something serious, then you benefit from treatment from the ambulance officers as soon as they arrive at your location. But if you call an Uber, the driver probably arrives at your location quicker than an ambulance would (since they have less far to travel), but then you have to wait until you get to the hospital in order to receive treatment. So, for Uber to be cheaper than an ambulance in New Zealand, the monetary cost savings would have to offset the (likely) longer time to receive treatment.

So, we would expect to see Uber having some effect on ambulance services in the U.S., but less so in New Zealand. But how big an effect? In a recent paper, Leon Moskatel (Scripps Mercy Hospital) and David Slusky (University of Kansas) investigate the impact of UberX on ambulance call-outs. Using U.S. data on the timing of UberX's entry into various cities and the number of call-outs in each city, they find that there is: least a 7% decrease in the ambulance rate from Uber entry into a city.
Moskatel and Slusky's paper is short and not particularly detailed. The analysis is fairly straightforward, and perhaps a little too much so. They claim that their analysis "follows a generalized difference-in-differences framework at the city-quarter level", but I don't think it does because they use only a simple dummy variable to test for the effect of Uber.

A 'difference-in-differences' analysis involves computing the difference between two differences (there's no mystery in the naming of this technique). Essentially, in this case you calculate the difference in ambulance call-outs between the period before UberX became available and the period after UberX became available (treatment cities), and then you calculate the difference in ambulance call-outs between the period before and the period after for control cities (where Uber did not become available). The problem with the analysis in this paper is that there are no control cities - all cities in their analysis had UberX become available. This doesn't bias their results, but it does affect how you interpret them, since they are really only testing for a difference in mean ambulance call-outs between cities with and without UberX.

If you have time series data but no control, you could run an interrupted time series analysis instead, which is very similar to difference-in-differences but simply tests whether the time trend in the data changes between the period before UberX became available and the period after. The results that they present in Figure 2 in the paper suggest to me that their analysis is probably under-stating the impact of UberX, since there appears to be a clear break in the time trends between the period before and the period after UberX became available.

Anyway, that is a fairly technical critique of a paper that tells us something interesting. Although, I wouldn't expect there to be as large an effect of Uber on ambulance call-outs in New Zealand.

[HT: Marginal Revolution]

No comments:

Post a Comment