Monday, June 19, 2017

Michael Luca



                                                                 Photo of Michael Luca 

For this post, I interviewed Professor Michael Luca, Assistant Professor at Harvard Business School.  Professor Luca’s research and teaching focus on the economics of digitization and on using data to improve policy and managerial decisions. In his work he has collaborated with organizations ranging from Yelp to the City of Boston.

Professor Luca wrote “Racial Discrimination in the Sharing Economy: Evidence from a Field Experiment” with Benjamin Edelman and Dan Svirsky, both colleagues at Harvard Business School. I first heard about this study when it was a working paper, and Professor Luca was interviewed on the podcast Hidden Brain. In this study, Professor Luca and his coauthors examined the online marketplace, or what academics call the sharing economy. In particular, they wanted to see how the differing policies of websites play out for the consumers who use them.

Think about the transactions you carry out on eBay or Amazon. They are pretty much anonymous: you usually know nothing about the people you are doing business with. But, what about Airbnb? If you’ve used it, you know that this site requires users to share personal information in order to participate. The requirements have changed over time, and they are different for hosts vs. guests. You must share at least your name, and hosts must post their pictures. Looking at a random sample of guests, Professor Luca and his coauthors found that 44% had posted profile pictures. The stated reason is that sharing this personal information builds trust between host and guest. But, as heard on the Hidden Brain podcast, this sharing of personal information has led to the unintended consequence of racial discrimination, against African American guests. In addition to the anecdotes you hear on the podcast and elsewhere (search #airbnbwhileblack on Twitter to see more), this paper provides evidence that discrimination on the platform is widespread.

How did the authors find evidence of racial discrimination on Airbnb? They exploited the fact that Airbnb users must share personal information, in particular their names. They created 20 Airbnb accounts, 10 whose names sounded distinctively African American and 10 whose names sounded distinctively white. Half of each were female, and half were male. Examples of female African American sounding names included Tanisha Jackson and Latoya Williams and examples of male white sounding names included Brent Baker and Brad Walsh. In total, they sent about 6,400 messages to hosts in 5 cities requesting bookings. It turned out that guests with white sounding names were accepted about 50 percent of the time while those with African American sounding names were accepted only 42 percent of the time. 

But, you might wonder, who is carrying out this discrimination? The authors found that even hosts who they thought might not discriminate were: hosts with multiple listings (many of whom are subject to and violating existing discrimination laws), hosts in diverse neighborhoods, hosts with more experience, and hosts with lower priced listings all discriminated. The only hosts who didn’t discriminate were the ones who had hosted an African American guest in the past. (The authors were able to gather this last bit of information from pictures of previous guests of the hosts.)   

Airbnb is aware of these findings, and has begun to take action. A 32 page report responding to the mounting evidence is posted on their website. A team of data scientists now has the responsibility of  evaluating this issue internally. Professor Luca has met with members of the staff, and shared possible solutions. One would be to eliminate completely the sharing of personal information. This, however, is not something that Airbnb is willing to do – Airbnb clearly wants to maintain a culture of sharing personal details, in contrast with Priceline and other short-term rental platforms. They do have an option called “Instant Book”, which to be clear, existed before, but now the goal is to get 1/3 of properties booked through it. From the Airbnb website: “Instant Book listings don't require approval from the host before they can be booked. Instead, guests can just choose their travel dates, book, and discuss check-in plans with the host.” So, it makes Airbnb more like your standard hotel site. Professor Luca argues that the effectiveness of instant booking in curbing discrimination depends largely on which hosts are using it: if those who were discriminating start to adopt it, it will make more of a difference, but if it is mainly adopted by people who were already less likely to discriminate, then the impact will be more limited. Airbnb also has added to its terms of service, which now includes an anti-discrimination policy, which every host must read and sign.

But, what if the hosts don’t even realize that they are discriminating? Professor Luca explained to me that this could be part of what is happening here:  it is what psychologists call unconscious bias. Hosts make decisions about who to accept, and don’t notice that they are systematically turning down African American guests.  Maybe these new policies will make them more aware. Hopefully they will now at least stop and think about the possibility that they are discriminating, and that they have the ability to take action to prevent it. (As a teacher, this reminded me of the existing evidence that all teachers, even female teachers, call on boys in the classroom more than they do on girls. I think this must be unconscious bias as well.)

Beyond his research on discrimination, Professor Luca has other work about information in online platforms. For example, in his working paper, “Survival of the Fittest: The Impact of the Minimum Wage on Firm Exit,” with Dara Lee Luca, he uses data from Yelp to study how increases in the minimum wage affect restaurants. The authors find that the least expensive restaurants are forced to close when the minimum wage increases.  

Let’s talk! I would love to know what you think about this example of unintended consequences. Please submit comments and questions.