Jennifer Stark is a computational journalist at the College of Journalism at the University of Maryland, College Park. Nicholas Diakopoulos is an assistant professor in the College of Journalism at University of Maryland, College Park. This is a guest contribution to Wonkblog.

The goal of Uber’s surge-pricing algorithm is to influence car availability by dynamically adjusting prices. When surge is in effect, and prices are higher, the idea is that the supply of drivers is increased while at the same time demand is decreased. We previously reported that it appears that rather than increase the absolute supply of drivers by getting more cars on the road, existing driver supply is instead redistributed geographically to places with more demand. If drivers are relocating to areas with surge-pricing, those areas will experience reduced wait times for their car, or better service, but the areas the drivers are moving away from will experience longer wait times, or poorer service. So who gains, and who loses? Which neighborhoods get consistently better or worse service?

Our analysis of a month’s worth of Uber data throughout D.C. suggests an answer: The neighborhoods with better service — defined as those places with consistently lower wait times, the pickup ETA as projected by Uber — are more white.

We collected data on wait times — Uber’s estimate for how long you will wait between requesting your car and it arriving — and surge pricing via the Uber API for 276 locations in D.C. every three minutes for four weeks from Feb. 3 to March 2. We didn’t want to miss any surges, so we chose three minutes, knowing that surges in D.C. are no shorter than three minutes. The surge-pricing data was then used to calculate the percentage of time surging. Data were analyzed by census tracts, which are geographic areas used for census tabulations, so that we could test for relationships with demographic information. Only uberX cars were included in our analysis since they are the most common type of car on Uber. (In the interest of making the data analysis transparent, all our code can be viewed  online.)

The map below shows that wait times are, in general, shorter in the center of the District and longer in the periphery.

Census tracts with more people of color (including Black/African American, Asian, Hispanic-Black/African American, and Hispanic/Asian) have longer wait times. In other words, if you’re in a neighborhood where there are more people of color, you’ll wait longer for your uberX.

Some of the tracts most significantly affected by this race-related difference in service are labeled on the map above, including Congress Heights, Bellevue and Washington Highlands, and the southern part of Southwest D.C., where average wait time is almost seven minutes for an uberX. Surge pricing is supposed to stimulate more driver supply and improve service, but given that these neighborhoods don’t surge often, it’s no wonder riders there are waiting longer for a ride.

In contrast, the tracts benefiting from this race-related difference (majority white tracts with shortest wait time) include Dupont Circle, Logan Circle and Georgetown, where average wait time is just over four minutes. These areas surge 43 percent of the time, which makes them attractive to drivers who want to earn more.

The associations between race and wait time as well as race and percentage of time surging are shown in the scatter plot charts below using normalized values (z-scores for the statisticians out there). The line indicates a statistical model that, despite taking into account household income, poverty rate, and population density, still shows the association. In other words, the association between people of color and wait times holds true even when household income, poverty rates, and population density are accounted for. This means that when comparing tracts where income, poverty and density are the same, the tract with more people of color will still experience longer wait times.


In addition to racial demographics relating to service quality in D.C., we found that poverty levels reinforce the higher wait times in areas with more people of color. So if you’re in an area that has a higher percentage of people of color and a higher percentage of poverty, you will wait even longer for your uberX car.

What could possibly be causing these patterns? It’s possible that a longer ETA in some neighborhoods experiencing worse service simply reflects a lack of demand in those neighborhoods. According to its patent application, Uber’s surge pricing algorithm is triggered by the ratio of supply to demand. You might expect that tracts with higher demand would surge more often in order to stimulate more supply. We found that tracts surged for anywhere from 16 percent to 47 percent of the time (see next map), with race also predicting how often tracts surge – even when accounting for differences in income, poverty and population density. Median household income is also related to how often tracts surge in the sense that it softens the influence of race. In other words, the number of people of color has less predictive power for time surging the richer the neighborhood.

Some neighborhoods may just have more “attractions” that make them desirable for Uber drivers to service — universities, bars and restaurants, or even the zoo. Many of the majority white tracts that we examined, with shorter wait times, had some of these elements. The northwest part of D.C. illustrates this well with several colleges in the area, and an average wait time of just 295 seconds. Despite having 75 percent people of color, Edgewood has an average wait time of 292 seconds, which is within the bottom quarter of shortest wait times. Higher demand in that neighborhood (and more surges) may be due to the prevalence of apartments, restaurants and bars, and Trinity Washington and Catholic Universities in the vicinity. In the future, we hope to compare Uber data to D.C. Taxicab Commission data in order to have a baseline on neighborhood-level demand.

But low supply and poor service quality could have a negative feedback loop with demand too: If you often have to wait too long for a car, maybe you stop using the service altogether. Drivers discuss in forums like this one  how they “play the system” in order to optimize for their own convenience or time/money ratio, including actively avoiding non-surge areas, and only going online in areas that typically surge. Some drivers have also admitted to strategically going offline so as to avoid receiving requests in particular areas. This type of driver behavior could be a contributing factor to a dearth of supply in some neighborhoods.

Uber and Lyft hail themselves as a relief from the rampant discrimination found in taxi systems that result from racial screening of pickups, or other screening based on desired destination. When asked for comment on the results of this analysis, an Uber spokesperson said: "City dynamics are complicated, and there is much more research that needs to be done. Uber is working hard to address a transportation status quo that has been unequal for a long time, making it easier and more affordable for everyone to get around their cities."

While any sort of racially biased agenda by Uber is extremely unlikely, our results suggest that race does play a role in predicting the service quality of uberX in different neighborhoods. This raises all kinds of substantive questions. For example, in light of the recent Supreme Court ruling that discriminatory intention is not a precondition to be held accountable for discrimination in the housing market, what are the implications of unintentional discriminatory side-effects being uncovered in other domains?  Should Uber be expected, or required even, to address this type of racial disparity in neighborhood service quality? Or perhaps Uber is a mirror of more general inequality in the city, its data reflecting some truths that we all need to grapple with as we seek to build a more equal society. This could be particularly relevant as Uber begins public-private partnerships with cities, like Boston, to use their data to inform civic transportation planning decisions.