By James DeanCornell Chronicle

White guests favor Airbnb properties with white hosts, but are more inclined to rent from Black or Asian hosts if they see featured reviews from previous white guests, new research co-authored by Cornell scholars finds.

The research suggests that sharing-economy platforms may be able to algorithmically harness racial bias in responses to recommendations to reduce racial inequality in access to ride-hailing, lodging, dating and other services.

On Airbnb, guests’ preference for hosts of the same race unexpectedly is offset or overcome by the influence of same-race endorsements, the researchers determined.

“Bias in the recommender system can actually attenuate bias in host selection, rather than reinforcing it,” said Michael Macy, the Distinguished Professor of Arts and Sciences in Sociology, director of the Social Dynamics Laboratory, and faculty member of the Department of Information Science in the Cornell Ann S. Bowers College of Computing and Information Science. “It goes against the assumption that different dimensions of bias would be mutually reinforcing.”

The findings are reported in “Fighting Bias with Bias: How Same-race Endorsements Reduce Racial Discrimination on Airbnb,” published Feb. 8 in Science Advances. In addition to Macy, the co-authors are Chao Yu, M.S. ’17, Ph.D. ’21, visiting lecturer in the Department of Communication, in the College of Agriculture and Life Sciences; and Minsu Park, M.S. Information Science ’18, Ph.D. Information Science ’19, assistant professor of social research and public policy at New York University Abu Dhabi.

Recent studies have shown that Black users of ride-hailing services such as Uber and Lyft wait longer for rides and drivers are more likely to cancel their rides. On Airbnb, Black hosts earn 12% less than white hosts of similar properties, and Black guests are 16% less likely to be accepted.

The researchers asked if racial bias in peer recommendations and responses to them exacerbate or lessen such discrimination.

To answer that question, the team analyzed nearly 8,000 Airbnb listings involving more than 7,100 hosts and more than 150,000 guest reviews between 2009 and 2018. Obtained from Inside Airbnb, an independent third party that collects Airbnb data, the listings were all “instant bookable,” meaning reservations could be made without a host’s review and approval. The team targeted those listings to isolate guests’ racial bias in the selection of a host from hosts’ bias in approving guests.

New York City, the researchers said, provided the density and racial diversity needed to analyze many bookings and control for different neighborhood characteristics that might influence the effects of bias. They used facial recognition software to classify the race of a user as perceived by others based on profile photos.

The team compared the chance that white, Black and Asian guests would book stays at properties with hosts of a different race to the expected probability if race were assigned randomly.

Guests of all three racial groups are more likely to choose hosts of the same race, they found, consistent with prior research.

The team focused on choices by white guests, however, because white users are overrepresented on Airbnb, while other groups are underrepresented. White users accounted for nearly 60% of hosts and guests in the data, compared to about 41% of the city’s population. As a result, same-race bias by white users would reinforce inequality, while among Black and Asian users it would increase opportunity for those groups.

The researchers next investigated how guests write and respond to listing reviews, a potential influence on host selection that they said has received comparatively little attention. They found no evidence that reviews – which were overwhelmingly positive – differed significantly in their language or enthusiasm based on a host’s race.

But who wrote the reviews did matter – even more than the host’s race.

The team analyzed recent reviews (up to six) featured on listings’ front pages. White guests’ reluctance to book with hosts of different races decreases, they found, as the number of endorsements by previous white guests increases. That effect was less pronounced among Black and Asian guests.

“White guests largely overcame their racial bias in host selection when hosts were endorsed by previous white guests,” the authors wrote. “The net effect of same-race endorsement is to reduce racial discrimination on Airbnb.”

The findings suggest a new strategy to encourage more equal access to the sharing economy, the researchers said.

Rather than removing profile photos, they said, platforms should ensure that front-page endorsements feature similar racial compositions. That would better tap the potential for same-race reviews to increase white guests’ willingness to choose nonwhite hosts.

“Increasing the exposure of white guests to white-authored endorsements of Black hosts may lead to more white bookings,” they wrote, “thereby making the algorithmic correction less necessary over time.”

Macy acknowledged funding support from the National Science Foundation and Defense Advanced Research Projects Agency while the research was conducted.

This story was originally published in the Cornell Chronicle.