REUTERS/Kai Pfaffenbach

At the core of Uber’s wild success and market valuation of over $41 billion is its data and algorithmically fueled approach to matching supply and demand for cars. It’s classic economics, supposedly: “Prices go up to encourage more drivers to go online. The increase in price is proportionate to demand,” says the official Uber video explaining their surge pricing system. It’s an easy line to buy into, but is Uber’s surge pricing algorithm really doing what they claim? Do surge prices really get more cars on the road?

My analysis suggests that rather than motivating a fresh supply of drivers, surge pricing instead re-distributes drivers already on the road.

I collected four weeks worth of Uber’s dynamic pricing information from their own publicly available data for five locations in Washington, DC. Every 15 seconds between March 15 and April 11, I pinged their servers and collected the surge price and estimated waiting time for an UberX car at those locations. Though only a tiny sliver of all of Uber’s data, it provided an initial window into how their algorithms are working.

The data do show that surge pricing works to maintain or improve service quality in the neighborhood of the surge, measured roughly as the estimated waiting time for a car. If the Uber system is working as advertised, we would expect price surges to come into effect to keep estimated waiting times in check as demand increases. The next graph shows how estimated waiting times correlate to changes in prices after some amount of time has passed for drivers to react.

The chart shows that in some locations an increase in price led to decreased estimated waiting times within a minute or two, such as in the Navy Yard, George Washington University, and K St. areas. But on U St. and Adams Morgan, the response was more muted: several minutes after a price spike the wait times were still going up, albeit less severely. The pricing system is working to reduce estimated wait times, but it seems to be working better in some neighborhoods than in others.

But price surging can work in any of three ways: by reducing demand for cars (less people want a car for a higher price), by creating new supply (providing an incentive for new drivers to hit the roads), or by shifting supply (drivers) to areas of higher demand.

The data I collected suggest that surge pricing doesn’t seem to bring more drivers out on the roads, but rather pushes drivers already on the job toward neighborhoods with more demand--and higher surge pricing. As a result, some neighborhoods are left with higher waiting times for a car.

So, why don’t surge prices work to get new drivers on the road? It might simply be that surge prices jump around too much. The graph below shows the surge price variation over the course of 12 hours at the intersection of U St. and 14th St. on March 17. In particular, look at 1:54 p.m. when the price multiplier is a healthy 2.3x. Yet, just five minutes later, the price multiplier is back to a normal 1.0x. In our data we found that Uber prices change every three or five minutes, up to 20 times per hour.

From the rider point of view, the rapid changes in price suggest that if you see a higher surge, you should wait a couple minutes and chances are it will have come down. But this chart also shows why it’s counterintuitive that surge pricing will attract a lot more drivers. When prices are changing so quickly, we can’t really expect drivers to jump in their car and start driving in reaction to maybe earning a bit more from a price surge.

Another reason to think that drivers aren’t jumping in their cars for surge pricing is that prices tend to tick down in bigger steps than they move up. Nine percent of all price changes were less than -0.5 multiplier, but only 5.5 percent were greater than +0.5 multiplier, which again may be great for riders, but for the drivers a higher surge price may be met with a substantial drop off when they show up -- for them, price surges are like an elaborate game of algorithmic whack-a-mole.

A forthcoming study by researchers at Carnegie Mellon University interviewed 21 Uber and Lyft drivers to shed light on actual driver behaviors. The study, led by postdoctoral fellow Min Kyung Lee, concluded that “more than half of the interviewed drivers … were not influenced by surge pricing information.” Drivers were put off by the fleeting nature of a surge price and the fact that being in a surging neighborhood doesn’t even guarantee that you will get pick-up requests from that area. If the surge disappears after three minutes, before the driver even gets to the surging area, that’s not very motivating.

The benefit of surge pricing on overall driver supply instead appears to stem from the long term effects of communicating to drivers when they should in general get on the road for planned events, holidays, or periods of expected high demand like poor weather.

Reached for comment, Uber spokesman Taylor Bennett stressed the long term perspective, “dynamic prices serve not only to allocate the existing set of drivers spatially over a city, but also to signal when the highest value times are for driving.” Such high value times are communicated to drivers via their in-dashboard app which they can learn from, as well as by examining their earnings reports.

But for such a long term outlook, pricing is changing awfully frequently, and is highly localized. The analysis shows that when prices are higher in one neighborhood it saps driver capacity away from an adjacent neighborhood. For instance, if surge pricing is higher on U St. or K St. I found that the estimated waiting time, a proxy for the service quality, suffers in the Navy Yard. As the next chart shows, if the price surge is higher on U St. than in Navy Yard this means a rider in Navy Yard would have to wait longer for a pick-up.

Uber spokesman Taylor Bennett noted that, “[The] analysis of such a small, short-term sample of data cannot capture long-term patterns and does not fully consider the complete set of price elasticities that are present.” That’s true. The data here are limited and estimated waiting times are only a proxy for actual service quality. Only Uber has the capacity to use its internal data to show us to what extent price surging decreases demand, creates supply, or shifts supply. (In the interest of transparency, my analysis is online)

In the meantime, it appears that rather than getting more drivers on the road in the short-term, Uber’s surge pricing instead depletes drivers in adjacent areas. A price hike in one area means drivers move there, but away from another, leaving it underserved. If someone in the newly underserved area now needs a car they wait longer, or perhaps a surge price has to come into effect to get a car over there. At the end of the day the Uber system appears to be more about re-allocation of existing supply. And this raises interesting questions about which neighborhoods end up with worse service quality -- those crosstown from a stadium or bars? Higher prices and better service for some means worse service quality for others.

Nicholas Diakopoulos is an assistant professor in the College of Journalism at the University of Maryland, College Park. His research and teaching are in computational and data journalism with an emphasis on algorithmic accountability, narrative data visualization and social computing in the news.