On Racial Bias and the Sharing Economy

When I first started with UberX, I was getting a lot of negative ratings. I asked the other drivers, “Guys, what am I doing wrong?” And they said, “Listen, you got to put some candy and water in the back.” I did, but my rating didn’t move. So then I cut my name in half, to Americanize it, and after two months my rating had jumped up. – Hass, NY Magazine

According to Susie Cagle, “the sharing economy doesn’t build trust—it trades on cultural homogeneity and established social networks both online and in real life.  Where it builds new connections, it often replicates old patterns of privileged access for some, and denial for others.”  Take, for example, the widely-cited Harvard Business School study by Edelman and colleagues on Airbnb that found that, controlling for other factors, “non-black hosts earn roughly 12% more for a similar apartment and similar ratings and photos relative to black hosts.”  Or check out the authors’ more recent study, finding that Airbnb guests with distinctly African-American names are 16% less likely to be accepted than identical guests with distinctively white names, even though engaging in this discrimination was “costly” for Airbnb hosts (Airbnb has sent a statement to Bloomberg contending that they have a “zero-tolerance” policy for discrimination).  Uber has faced allegations (another thoughtful piece here) of retail redlining, just like pizza chains and other business decades ago.  And, this “new” economy has done nothing to solve old concerns highlighted by Yale Law Professor Ian Ayres and his colleagues, concerning racial disparities in tips for drivers, which may also affect drivers’ ratings and thus their ability to remain employed.  As the Anti-Eviction Mapping Project puts it, “as long as you don’t think too deeply about the implications for labor law … the Sharing Economy sounds marvelous.”

Has the “New” Sharing Economy Actually Taken Us Back in Time?

Hotel owners do not have the ability to examine the names of potential guests and reject them on the basis of race.  But the HBS study shows that home owners regularly discriminate on the basis of their guests’ race on Airbnb.  A similar story exists on regarding consumers: given that hotels do not have a single owner in the same sense that Airbnb rentals do, consumers are unable to discriminate on the basis of the owner.  On the other hand, the HBS study illustrates that Airbnb guests are discriminating against African-American home owners.  And this racial discrimination is becoming even more prevalent given that Airbnb continues to comprise a larger share of the hotel market.

Similarly, like other businesses, taxis cannot engage in redlining.  Legally, they cannot refuse to pick up or drop off a passenger because of the passenger’s geographic location (although, of course, this does occur in practice).  Because Uber drivers can choose where to work, they can engage in de facto redlining and avoid to picking-up in certain neighborhoods.  There is insufficient data to conclude whether Uber engages in redlining or not, but, unlike taxis, Uber’s business model coupled with a dearth of regulation allows for this possibility to occur without any legal repercussions.

A Change in Design

Although most of us have racial biases, the Sharing Economy can be devised to reduce the impact of our biases.  For example, changing the design of the Sharing Economy such that we are not exposed to irrelevant information that can lead us to make undesirable choices (i.e., racial discrimination) can help eliminate bias. Indeed, the information that you don’t have when evaluating someone can be just as important as the information you possess—“confusing and complicated and ultimately irrelevant pieces of information [] can serve to screw up your judgment.”  In Blinding as a Solution to Bias, Professors Robertson and Kesselheim put forth the example of a fingerprint examiner determining whether certain fingerprints match the suspect of a crime.  If the examiner overhears an eyewitness asserting that the suspect is indeed the perpetrator, the professors aver, he will probably be more likely to make a faulty determination (i.e., the fingerprints match the suspect when they in fact do not) than he otherwise would have.  This is because the eyewitness testimony may unconsciously influence the way the examiner looks at the prints.  In short, if information serves no useful purpose yet leads us to engage in discrimination, it should be left out.  Understandably, this proposal may be hard for many of us to swallow given our propensity to view access to more information as beneficial.

Edelman and colleagues seem to agree, as they too contend that changing the design of online marketplaces may reduce discrimination.  Indeed, they gives multiple examples of how we can use Gladwell’s assertion to reduce bias in the Sharing Economy.  They state that “there is no fundamental reason why a guest needs to see a host’s picture in advance of making a booking—nor does the guest even need to know a host’s name (from which race may be inferred).”  Applying this to Airbnb, they suggest that the company can reduce racial discrimination by expanding its “Instant Book” option, in which hosts, like a hotel, accept guests without screening them first.  In fact, this would be both more convenient for guests, and more profitable for owners, as the authors show that engaging in racial discrimination is financially harmful for hosts.  Further, Airbnb can conceal guest names, just as they do email addresses and phone numbers, or, like eBay, use pseudonyms.  The same can be said about driving platforms such as Uber and Lyft to eliminate the concern that drivers may refuse passengers due to their race.

Limitations

Noticeably, the above-mentioned design changes would only reduce racial bias during the initial interaction between the service provider and customer.  Reviews and ratings—incredibly crucial components of the Sharing Economy—are provided at the end of their interactions.  This is significant because, as Professors Linda Hamilton and Susan Fiske point out in Behavioral Realism in Employment Discrimination Law, our preferences and behaviors are not static; rather, they change based on the situations in which we find ourselves.  In other words, just because one’s bias does not result in discriminatory behavior in an initial encounter, it does not mean his bias will not result in discriminatory behavior in a subsequent encounter.  Hamilton and Fiske’s assertions have far-reaching implications for the Sharing Economy: all else being equal, service providers (Uber drivers or Airbnb hosts) may give customers (Uber passengers or Airbnb guests) different reviews or ratings based on their race, and vice versa, making people of color economically worse off than their white counterparts.  In addition to the negative effects on black Airbnb users in the studies above, negative ratings may also result in, for example, black Uber drivers being more likely to be terminated or black passengers less likely to be picked up.

Conclusion: Can We Address these Limitations?

There is no easy fix for the racial bias that plagues the Sharing Economy’s consumer ratings systems.  One possible approach would be to get rid of ratings altogether.  We might also adjust ratings to take racial bias into account.  For example, Uber has enough data to ascertain how much a driver’s (or passenger’s) rating is negatively effected by his or her race, all else being equal.  Uber might thus adjust users’ ratings to compensate for this negative effect.  Or, companies could use techniques to “debias” users, perhaps by alerting users about the effects of racial bias when they are reviewing each other, or asking users questions and framing them in such a way as to reduce the effect of bias.  Finally, if users are required to wait a couple of weeks before they rate their hosts or their drivers, they might remember their experience but not the (already biased) rankings that could perpetuate more bias.

The Sharing Economy is not all bad from the perspective of racial bias.  Services such as Uber and Lyft deserve credit for mitigating the problem of “hailing while black” and, perhaps, better serving low-income communities than do traditional providers.  But we ought to hold the Sharing Economy accountable for the problems it creates, exacerbates, or perpetuates, and encourage it to find solutions to eliminate these concerns.