airbnb

I am a white Airbnb host. I reviewed 102 guest requests to assess my own racial bias

Peer-to-peer, short-term housing platform Airbnb is probably my favorite consumer web company. A traveling member since 2011, my wife SACM and I have been hosting travelers too for most of the last year. I’m a proud and happy user.

Yet I know that some of the loudest news about Airbnb in its last couple years of mainstream expansion has been controversial: first, about the company navigating municipal hotel taxes and, most recently, its central role in a conversation about racial bias in the sharing economy.

You know, #AirbnbWhileBlack.

So now that my wife and I have been hosting for nearly a year and have received more than 100 requests, I wanted review for our own selection bias.

I believe that racism doesn’t necessarily need to be intentional. Humans are evolutionarily tribal, so even the most aware and best intentioned of us make choices that benefit people who look more like us. It can be unconscious and insidious.

So I recognize I likely contribute to institutional racism (I am a white male host). To counter those biases, we need to be willing and comfortable to self-reflect and self-review. When we talk about unconscious bias and its role in lingering racism, it’s always someone else’s fault. We all need to think about our role. In a small way, that’s what I’m trying to do here.

Now let’s be clear: our total number of guest requests is too small to be as telling as bigger studies (that are showing there is a meaningful problem). This is just meant to be a snapshot and offer advice to other hosts who want to contribute positively to the platform — and to force me to be more conscious of my own actions.

So my first piece of advice to all hosts is to admit selection bias, be conscious of it and then still adopt some kind of defense against that bias.

Here’s what my wife and I do: I manage our Airbnb account and so when a new request from a traveler comes in for dates that we could host, I ask (usually text) SACM if she feels up for it (sometimes we’d rather have a quiet house to ourselves, other times we haven’t cleaned the room since the last guest yet). I do this without offering any name or personal information about the guest. If SACM is up for hosting, I am too.

That means most of the times we turn down a guest without a specific reason, SACM is doing it on our behalf blindly. That surely helps. But there are still decisions I make without the benefit of the blind: like when we get several requests for the same date at similar times, and I just make a choice.

That’s where unconscious bias can lurk.

So to evaluate our track record, I contrasted the racial breakdown (assumed crudely* from profile photos) of everyone who requested with the breakdown of whom we accepted. My assumption here is that the percentage breakdown of accepted and declined guests should look pretty similar to the percentage breakdown of all requests. The more similar, the less bias I show.

So what do the data show?

Spoiler alert: I feel comfortable saying that I do not show meaningful bias. Given the small sample size, what small variations can be seen aren’t damning, yet I still do show some bias. I am white, and these data suggest I am (ever slightly) more likely to host white guests. Still, as you’ll see, the data comes with a complication.

Here’s what I have:

  • From July 2015 to May 2016, we received 102 total requests to stay with us. We accepted 53 of those requests (51 percent), rejected 39 and 10 prospective guests declined themselves after initial contact.
  • Maybe as many as two-thirds of guests come in pairs — often romantic couples — so I’m only calculating based on the person who requested, since it’s her profile photo I saw. Every single person did have a profile photo.
  • We show no real gender bias. 61 percent of requests came from women. That’s right in line with 62 percent of accepted guests being female and 60 percent of those declined being so. Put another way: 59 percent of women who requested were accepted, and 56 percent of men were. That’s a slight pro-female slant but the distinction is marginal.
  • We were more likely to accept requests from foreign travelers than domestic ones. 21 percent of requests were from people who were traveling from outside the country, but 28 percent of those we’ve accepted are international and just 18 percent of those declined were. (This is likely because domestic travelers more often book last-minute, which we are more likely to decline, though I did not evaluate that data).
  • It’s small, but the data suggest I’ve shown a slight preference for white guests (but it doesn’t seem to be at the expense of black travelers). 83 percent of accepted guests were white, though just 78 percent of requests came from white users. Just 5 of 103 total requests came from black travelers (that low number could reflect the overall makeup of Airbnb users, people traveling to my neighborhood or even just of those interested in staying with me, but whatever the case it doesn’t seem statistically valid). Of our black user requests, we hosted two, we declined two (both were for weekends we turned down white guests also) and one cancelled on us. In this data, my white bias seems to come at the expense of Asian travelers: 12 percent of requests came from Asian users, though we accepted just nine percent. Overall, we accepted 59 percent of white guest requests and 56 percent from Asian travelers. That’s five guests we accepted, four we rejected and three declined themselves. The numbers are also small enough to be vulnerable to noise, but there seems to be some bias in there — and it’s particularly unfortunate since Asian hosts, too, are getting shortchanged on the platform.

And maybe that’s exactly why this Airbnb race issue is so thorny. I am a white male Superhost, and the data seem to suggest I’ve shown a slight bias, but the issue gets complicated and personal once you dig in.

Why did I turn down those four Asian guests? One was traveling with a small child (currently, we choose to not host kids). One had to arrive when we couldn’t be home (something we require for first-time guests), and one I declined to instead approve a longer-term, more profitable Australian traveler (a financial bias, not a racial one). The fourth one? She was part of a wave of requests for just this past Memorial Day weekend. I rejected three people, including her (the other two were white), upped our asking price for the night and then got another request, which I accepted. The person I accepted? He was a white dude. Guess what: if the order of those requests were flipped (the Asian woman requested after I upped the price rather than before), then the data wouldn’t show any racial bias at all. The white acceptance percentage would fall and the Asian one would normalize higher with just that one flip.

That’s the trick with small sample sizes. You can pick them all apart. The trouble for Airbnb is that far larger data sets are showing the same thing. So I’m trying my very best to not pick apart my own data because we know there is a problem, and I’d rather be part of the solution than the problem.

The data I reviewed suggest I’ve shown a slight bias toward my in-group of white travelers. So what can I do about it on an individual scale?

  1. SACM and I are going to rely even more heavily on our blind approval process. She says yay or nay without ever seeing a profile pic. But of course, as any student of institutional racism knows, this can’t eradicate unconscious bias for lots of reasons. For example, SACM and I give preferential treatment to requests from people who have already used Airbnb and have positive reviews, so of course a platform that benefits whites will mean whites are more likely to have more positive reviews. Still the blind test is a popular tool to reduce bias.
  2. I’m going to use this data to motivate me to bring my race consciousness into vetting guests to overcome whatever biases I’m showing here. So I simply have to remain as conscious of my potential bias as I can. I hope this can compensate for whatever biases I may retain.

What about Airbnb the organization, the one I admire so much?

How might you solve the potential for racial influence in decision-making, knowing that photos are so vital for the platform to work and so many issues are so tricky? (I won’t touch the thornier problem of whether people should have the right to discriminate on whom they want in their homes: for example, our home has stairs and would be unable to host a person in a wheelchair.) One common suggestion is to require instant booking, which Airbnb is encouraging (though I’d be nervous about it). This means anyone who meets certain criteria can book a room automatically if the date is clear (I’d require existing reviews on the platform and no day-of booking, personally).

But I have another idea. Encourage users to self-identify their race (and other similar subgroups) and give hosts a dashboard showing the racial breakdowns contrasting accepted and requested. If we think much of this is an unconscious bias, data might be a powerful tool.

-30-

*Yes, I’m assuming that people were using photos that accurately depicted them. And even then, of course we know race is a construct. I am using crude assumptions for the purposes of this exercise and don’t mean to suggest that I can be certain how any of these people would self-identify racially (or by gender for that matter). I can’t be. They’re approximations only.

Also, I did track other commonly used racial or Panethnic subgroups — Arabic, Hispanic, Indian and the like but those totals were all one or two people at most so there wasn’t much to tell there. (For example we accepted 100 percent of two Hispanic requests and 0 percent of one Indian request. That would inaccurately suggest I am highly biased toward Hispanic travelers and the most biased against Indian travelers, so I decided not to include them here).

I did not review any variations for longer term rentals or if they had many reviews from past Airbnb trips (we’re more likely to accept both). I also didn’t track or publish here other less identifiable communities of interest, like age (do I have an age bias?), LGBTQ (hard to tell over email) or something like military veterans (we hosted at least one active member of the military).

As always, I hope this can be part of an open dialogue, not a place for hate. Let’s be better versions of ourselves.

Leave a Reply