Your study found that landlords use two sets of tools to screen tenants, referred to as “algorithmic proxies” and “gut proxies.” Can you describe them, explain the differences, and comment on which kinds of landlords tend to use which?

Philip: Most landlords want to screen tenants in a way that identifies the most profitable tenant: Who’s going to stay in the unit the longest, pay rent reliably, or cause minimal wear and tear to the space? We found a strong difference between large, professionally managed properties and landlords with small portfolios. They operate their screening approaches differently in ways that have a complicated and interesting set of outcomes for tenants. 

Large professional property managers who manage onsite, for example, a 200-unit building, are structured and algorithmic about how they screen tenants. They put all the pieces of information about a tenant, such as credit score or income, into a computer that identifies which tenant is eligible or not. The property manager, in that case, doesn’t have a lot of authority to supersede that because the computer makes the determination. Indeed, if the manager did supersede the computer, they could be vulnerable to fair housing lawsuits. Large professional management companies do get evaluated for fair housing issues frequently, so they institute systems that are designed to prevent that. That’s the “algorithmic proxy.” You’re using these measurable and observable features of a tenant’s characteristics to try to predict their behavior in the future. 

Landlords with small portfolios operate differently. They feel that these observable measures, like credit score and whatnot, are useless. They don’t really differentiate, especially in the low-income market, between tenant behaviors, and they trust their own “gut” much more. So, they meet the tenants face-to-face. They have a conversation with them, look at how the tenant is dressed, how they comport themselves, how rapidly they return various application materials, or how they parent their children. These landlords use these hard-to-quantify approaches to determine whether they think the tenant is a reliable person. So, that’s what we call the “gut proxy.” It’s using subjective measures to determine the same outcome: Is this going to be a tenant that’s profitable to house down the line for that landlord?

Your study also found that both kinds of screening methods result in racial discrimination. Can you expand on this?

Philip: I think there are interesting ways in which both screening methods reproduce patterns of racial inequality in the country. Sometimes, it depends on how narrowly you define discrimination. If it’s based on this idea of, “I don’t like this particular racial group and so I’m not going to rent to them,” then that’s different than how we define it, which is: using approaches that exacerbate structural inequalities within society. Landlords are using observable features which are all predicated on a tenant’s lived experience. Tenants facing challenges with their housing also face challenges in the labor market. They’re less likely to have good credit, high levels of income, and so on. So, despite these objective measures and these technical tools—where the race of the tenant is blinded to these algorithms—the algorithms are more likely to reward tenants who have rarely experienced discrimination, hardship, and structural marginalization.

I think it’s easier to see why the “gut proxy” results in discrimination. Landlords have pre-existing beliefs about low-income minority and female tenants. Often, their gut is checking to what degree the tenant aligns with those stereotypes or how much they’re able to “perform respectability” beyond what that landlord sees. Again, tenants who can make the effort to perform characteristics that defy a landlord’s stereotypes are more likely to be housed. The amount of scrutiny that’s placed on women and tenants of color is much higher than the scrutiny of more advantaged demographics.

It seems that some landlords use algorithms to reduce bias and avoid violating fair housing laws. Were some aware that these methods can still be biased?

Eva: Maybe. Maybe not. I think avoiding bias is not the priority for most landlords. Their top concern for their business is paying their mortgages. 

We met some landlords who talked about wanting to reduce bias. It’s hard to know exactly how much of that is related to a true commitment and how much it is just a part of their business spiel, but that’s a different study to figure out exactly how deeply those beliefs run.

We talked to a few landlords who use specialized algorithms. When it came to the Housing Choice Voucher program, these landlords were creating algorithms, specifically to account for the fact that rent would be paid reliably and directly to the landlord. In those cases, we did find that reliable payment could reduce bias on the basis of race and gender, as these payments helped landlords to overcome some of the biases that they might otherwise have. The value of reliable rent coming from the voucher program was baked into that algorithm. But I think the landlords’ main concern here is getting their rent paid so that they can pay their mortgages.

Some landlords who use gut proxies to find a good tenant conduct home visits with potential renters. How prevalent is this across the country?

Philip: That’s a great question and, obviously, our data cannot make any claims of prevalence across the country. Our paper quantifies home visits within our random sample, but that’s the best we can do. This practice was much more common among small amateur landlords compared to property managers of professionally managed complexes, where they are less likely to implement home visits because of potential fair housing violations. Also, it’s a logistical burden to go out to do these visits.

The fact that landlords would go through and make the effort to make these home visits in places like Baltimore and Cleveland did surprise us. It was a sizable number of our sample. A key thing we learned here in our landlord research is that landlords, while they all share similar goals and utilize similar tools to achieve those goals, are going to be responding to market features and factors. When you look at housing and landlords in California versus Baltimore versus Chicago versus Washington D.C., we see so much heterogeneity. It did surprise us that home visits would be less common in other contexts.

Your study draws on interviews with landlords from cities with varied housing markets. How do these practices change in strong versus soft markets, or areas with varying degrees of racial segregation?

Eva: It’s challenging, in a qualitative research study like this, to tease out these different market factors. The four cities vary on some of these characteristics. Broadly speaking, we can say that in tighter housing markets, landlords have more choices about who they rent to and tenants have less choice. There are more opportunities for landlords in tighter housing markets to discriminate and to wait for the next application in their pile. Some of that might be based on bias, which is more prevalent among the smaller landlords than the larger landlords. The latter are much more likely to be submitted to a fair housing audit. 

We found that there were pockets and neighborhoods in each city that were more like neighborhoods in other cities than other neighborhoods in their own cities. For instance, the rental market in South Dallas, which is a neighborhood that has a high population of African American renters, is much like much of the city of Baltimore, which looked a lot like neighborhoods east of the river in D.C. There were other neighborhoods in D.C. with a lot of students and a lot of corporate housing that looked very much like other parts of Dallas. 

One of the things we ask is how markets vary across cities and how there are micro-markets within cities where landlords often specialize within certain types of neighborhoods. When you have neighborhoods that are already highly racially segregated and already have high poverty rates, landlords have a different set of strategies than they might be using in a more heterogeneous or more affluent neighborhood.