Authors:
Yanni Loukissas
On November 19, 2020—a peak moment of Covid-induced anxiety and restlessness—the New York Times published an article on "Zillow surfing," an increasingly common form of virtual escapism [1]. Many U.S. readers will be familiar with Zillow, the market's dominant service for selling, buying, renting, and financing residential real estate. In the context of the pandemic, people are finding new and creative uses for the site. "What many are contemplating when they browse Zillow," explains the article, "is not necessarily a purchase, but an alternate life."
Having been largely confined to my own home for the past two years, I am not surprised that people are turning to Zillow to imagine how their living circumstances might be improved. In the first few months of the pandemic, visitors to for-sale listings on the site increased more than fifty percent year over year. For many of these new users, Zillow is simply a convenient and comforting means of temporarily escaping the reality of the pandemic. The article, titled "Zillow Surfing Is the Escape We All Need Right Now," appeared in the Style section of the Times and was meant to be a light diversion from more serious investigative pieces on the political and health impacts of the pandemic. However, Zillow surfing highlights an important phenomenon that predates Covid-19.
Insights
→ People are using data on Zillow to refresh the timeworn dream of owning the right home in the right neighborhood.
→ However, in a society split along racial and class lines, finding your dream home means accepting and navigating many structural inequalities.
→ The market that Zillow aims to optimize is a system by which society segregates itself to preserve social hierarchies.
People have learned to use data to refresh the timeworn dream of owning the right home in the right neighborhood. This new version, which we might call the algorithmic American dream goes something like this: Access to data and algorithmic tools can help us get control of our lives by reducing uncertainties and making our lived realities resemble our filter bubbles.
We create filter bubbles when we use algorithms to personalize our online experiences, by blocking any information that doesn't conform to our existing belief system [2]. The term filter bubble was first used in 2010 to characterize the way that people curate their social media feeds online, effectively excluding perspectives they do not care to see. The same kinds of filters also allow us to address the uncertainties we might face in online dating or in using an app to find a nearby restaurant. Why not use data to seek out a new and more fulfilling home life, if only as a fantasy?
Without the need for a real estate agent, who can only show a couple of listings at a time, visitors to Zillow can filter properties by the home features they care about. As with the news, home searches can be made to conform to almost any preexisting belief system. Unfortunately, this way of using data ensures that we do not see the implications of what we filter out. Rashida Richardson's illuminating and cautionary article, "Racial Segregation and the Data-Driven Society: How Our Failure to Reckon with Root Causes Perpetuates Separate and Unequal Realities" [3], highlighted in this issue of Interactions, suggests one particular concern that might be painful for some to see: How does the history of racial segregation shape the way users of Zillow identify their ideal homes?
In my 2019 book, All Data Are Local: Thinking Critically in a Data-Driven Society, I write about Zillow as a data setting, a context in which property values are meant to be operationally rather than truly understood [4]. Data settings define what we can do with data and what claims about the world we can use them to support. I explain Zillow as part of the "interface economy" [5], a rapidly growing area of business in which companies aggregate data from various sources and provide access as well as interpretive tools. Some of these companies charge a fee. Others use the popularity of their services to draw advertising revenue. Zillow offers a combination of visual, discursive, and algorithmic tools to make sense of data from county records, multiple listing services, and even homeowners themselves. It is best known for generating daily estimates of the current value of nearly every home in the U.S.
The understanding that where you live determines the opportunities available to your children underlies the very notion of segregation.
The widespread fascination with Zillow is not just about appreciating beautiful homes, which is something that I can relate to as a former architect. Rather, the company encourages visitors to think of real estate data as a vehicle for personal fulfillment. "We can help you move forward," reads the messaging on its website. Visitors to Zillow are responding to this sentiment en masse. "There's something therapeutic about searching houses and starting to make plans for something with a positive outcome," explains one Zillow user. "It makes me think there is a light at the end of the tunnel, and someday I'll be at my dream house" [1].
Completing this dream image takes more than data on property values. Zillow offers access to data on other issues of interest to potential homebuyers, such as neighborhood school performance and crime rates. "It's easier to picture your future when you have access to the floor plan of the space or know which school your children would attend if you lived there," says one Zillow user [1]. This is where the true stakes of Zillow become clearer. The understanding that where you live determines the opportunities available to your children underlies the very notion of segregation. In a society split along racial and class lines, finding your dream home means accepting and navigating these structural inequalities.
As Richardson explains, your zip code is a social marker that can be used to discriminate against you. For example, in the 1930s the Home Owners' Loan Corporation created "redlining maps" to claim that there were differential risks of investing in segregated neighborhoods throughout the U.S. Using these maps, federal agencies and private banks alike directed disproportionate financial support to homeowners in predominantly white neighborhoods. Today, the racist effects of redlining can still be felt in cities across the country. Could Zillow have similar cumulative effects, by empowering its users to enact their own beliefs about financial risk through algorithmic filters?
Let me unpack some relevant considerations about how Zillow works as a data setting. The visual interface for the site is a parking-lot-gray map. It is laced with thin white roads and spotted with patches of green and blue to indicate natural areas. Small red dots on the map indicate data points, which are properties available for sale or rent. Underneath each dot is a single value: 425K, 587K, 245K. Zoom in further, and the map subdivides into parcels. Clicking on one brings up more information: an address, the number of bedrooms and bathrooms, the square footage. More-sumptuous details also reveal themselves in this interaction. Artful photographs show off the property, inside and out. Evocative descriptions evoke interior details ("hardwood flooring") and surface materials ("granite countertops"), as well as nearby amenities ("local restaurants and coffee shops").
This context for data exploration is complimented by a discursive framework built around the metaphor of a personal journey. "We can help you find your way home," promises the site. Finally, the entire setting is brought to life by its underlying algorithms. Whether or not they are on the market, each parcel mapped on Zillow comes with a Zestimate: an estimated property value, generated anew each day by the company's proprietary valuation model. According to Zillow, these values are within 4.5 percent of the final sale price of properties fifty percent of the time. Real estate agents are often dismissive of Zestimates. But they can be extremely alluring to buyers and sellers alike. Even if users know that Zillow's numbers are not entirely reliable, Zestimates give them a handhold in an otherwise uncertain and rapidly changing marketplace. What's more, they are free. But there are unseen costs to using Zillow that fall disproportionately on nonusers, particularly those in low-income communities.
In 2019, I wrote about my own experience learning about Zillow as a first-time homebuyer in Atlanta, Georgia:
I found that Zillow supported certain kinds of desires: the right neighborhood, the right price, the right school zone, and the expectation of a stable investment, or even a profit. But it did not support other things my partner and I cared about. We were wary of contributing to Atlanta's latest wave of gentrification. There are few policies in Georgia to protect low-income homeowners and renters from the increasing costs of staying in their own neighborhoods. In fact, we were ambivalent about becoming homeowners at all, if it meant participating in an inherently unjust system [6].
Eventually, I did buy a home. It was a challenge to find a neighborhood in Atlanta that was not either segregated or actively gentrifying. During the civil rights movement, Atlanta became a center for Black cultural life. More recently, middle-class white newcomers, like myself, have either chosen to live apart from long-term Black residents or displaced those who have not had the means to compete in the market. Tragically, gentrification is fundamentally altering some of Atlanta's most historically significant Black neighborhoods, such as the Old Fourth Ward, where Martin Luther King Jr. once lived and is still memorialized today. On Zillow, the neighborhood is represented as no more than a collection of listings for sale. The photos reveal dreamy interiors (remodeled to sell) and their Zestimates are on the rise. The interface economy seems to take all data at face value, without regard for local or contested meanings. In Zillow, more rights are afforded to data than the people or places that they rather crudely represent.
If researchers who contribute to human-computer interaction are to understand the dynamics that Robinson has described, and formulate just responses, we need to raise our own critical consciousness. Looking for an indication of how the field is doing in this regard, I ran a search on "race + segregation" in the digital archives of this publication, Interactions. I also searched the proceedings of three conferences connected with the Association of Computer Machinery (ACM): Computer-Supported Cooperative Work (CSCW), Computer-Human Interaction (CHI), and Designing Interactive Systems (DIS). The past 10 years show a steady increase in the number of articles that mention both terms, which is promising. But 10 years before that, my search returned few or no results.
The study of racial segregation in HCI deserves to be more than a niche topic for a few "social justice" researchers. This cannot happen without the support of the institutions of higher education that host HCI research programs. At the same time, each of us must confront the implications for our own research. Robinson identifies two primary conditions that might be holding individual researchers back from doing so: 1) a lack of understanding about the local histories of segregation and 2) a fear of revealing their own complicity, since most HCI researchers are white and middle class.
This analysis rings true. It also indicates several opportunities to pivot. There are things that HCI researchers can do to raise their own awareness and that of the field overall. They can work to increase the diversity of HCI programs by making those programs more accessible and relevant for BIPOC young people. They can challenge oppressive legislation targeting "critical race theory," which threatens to keep public universities from teaching the truth about race. Through cross-listed courses and collaborative grants, they can learn from and support researchers who study the effects of racial segregation in Black media studies, American studies, urban sociology, and cultural geography.
From my own research, I can offer HCI researchers a few tools for reconnecting data to local histories. I agree with Robinson that racial segregation can only be understood locally, in relation to specific spatial, cultural, and policy conditions. Unfortunately, some researchers in HCI misinterpret local as meaning small scale. They fear that a local perspective will limit the application of their technologies or result in a statistically insignificant study. I contend that local is not a particular scale. Rather, it is a relative designation that is dependent on context. For a nuclear physicist, the local is subatomic. For an astrophysicist, it is the solar system. In HCI, a local perspective is one that acknowledges the seams inherent in computing systems. When HCI researchers dismiss the local, they are assuming that it doesn't matter where, when, or who we are. We are all just users. This is the "myth of digital universalism," writes Anita Chan [7]. Thinking locally can be a form of critical thinking. It is an antidote to digital universalism, which can help us see the human consequences of our filter bubbles.
Richardson's article offers us three important questions for assessing data-driven systems, such as Zillow, from a local perspective. These questions have the power to reveal the underlying connections of data and algorithms to racial segregation: 1) What problem is being solved by the system? 2) What data is being used to train the system? 3) How is the system being evaluated? The creators of these systems usually have ready answers to these questions. But they are rarely local answers, and often overlook important implications. Zillow provides a useful object lesson in this regard. Let's consider each of Richardson's questions in relationship to Zillow.
What problem is being solved by Zillow? According to its own messaging, Zillow aspires to help visitors address a personal question: "How can I find my way home?" Judging by the overwhelming popularity of the site, this question resonates broadly. Users of Zillow may not feel at home in their current life circumstances. These feelings may have been exacerbated by the pandemic, which has forced us to meet many more of our needs at home. Our homes have become our primary workplaces, our schools, and often our only places for respite. It is understandable that people need things their current homes are unable to provide. However, buying or selling a home in a market economy has implications that go beyond personal fulfillment. As I explained above, the sale value of a house feeds back into Zillow as a new data point, with effects on how comparable houses are valued. Depending on the local policy environment, this can lead to increases in rent or property taxes for nearby residents, as well as the overall affordability of a neighborhood. In Atlanta, there are few protections from displacement afforded to low-income residents, most of whom are Black. In this way, finding your dream home can cause others to lose theirs. In a society where wealth is inequitably distributed by race and class, the cumulative effect is segregation.
The entire real estate market that Zillow aims to optimize is a system by which society segregates itself to preserve social hierarchies, such as class and race.
How are Zillow's algorithms being trained? Zillow's algorithms make use of publicly available data to follow the housing market, with the intention of increasing transparency and helping people understand the rapidly changing dynamics of real estate:
Our mission is to empower consumers with information and tools to make smart decisions about homes, real estate, and mortgages. For this reason, we do not remove public record property data from Zillow unless it is shown to be erroneous [8].
The idea of using public data to help people can seem benevolent. After all, this data is created for the public good. Unfortunately, I have found that public data, such as that which is created for tax assessment purposes, can vary widely in its formulation and accuracy [4]. Zillow uses this data because it is better than nothing. Moreover, the company can blame public entitles for any errors, while also putting the responsibility for finding them onto users of the site. Visitors to Zillow may see and correct errors about their own properties, but are unlikely to know about more systemic errors, particularly in data from other places.
Instead, visitors are likely to accept that Zestimates are good enough, and check back regularly to see how their homes, dream or real, are faring. I do this myself, usually with an underlying feeling of anxiety. In this way, Zillow is training not only its algorithms but its visitors as well. Repeat visits to Zillow may be good for the company, but this constant checking in encourages visitors to think about property in terms of fear and financial risk, sentiments that motivated the original redlining maps.
How is Zillow evaluated? Like many companies that operate online, Zillow got its start with an advertising model of revenue. It has long connected users of the site to lenders, real estate agents, inspectors, photographers, designers, contractors, and property managers. These are the original customers of Zillow, and they pay significant fees to be represented on the site. Zillow benefits less from the veracity of its Zestimates than their allure for visitors. More recently, Zillow has begun competing for other kinds of work. It is now a lender, a real estate agent, and a property manager. These roles expand its sources of potential revenue without changing the conditions on which its data and algorithms are evaluated. Data makes Zillow a destination, but the company does not seem to rely on data or algorithms to make money. It is not clear that Zillow holds itself accountable for the quality of its Zestimates or the social implications of generating them on a frenetic daily basis.
I was encouraged to find that Robinson's three questions for data-driven systems connect well to the visual, discursive, and algorithmic framework I have developed for analyzing data settings. Data settings establish what claims can be made visible through data (the problem), what counts as data (the training), and what makes for an acceptable claim (the evaluation). It is important to remember that data settings can be shaped by a variety of interpretive motivations. For example, a data setting might alternatively address the problem of racial segregation, integrate the accounts of those who are most affected, and be evaluated in terms of its ability to raise awareness. Such a setting would not simply establish a more equitable means of shopping for a home. It would reveal the inequitable nature of all property valuation and inspire HCI researchers to help people imagine alternative forms of cooperative inhabitation.
Jean Baudrillard once quipped, "Disneyland is there to conceal the fact that it is the 'real' country, all of 'real' America, which is Disneyland" [9]. He means that we see Disneyland as a distinctive place for fantasy and for being childlike. We imagine that Disney is distinct from the outside world, which is run by serious adults. This distracts us from the reality that most of our own behavior is childish, and that no one is a complete adult. Zillow surfing accomplishes something similar. We tell ourselves it is merely an escape. Who doesn't want the fantasy of a dream home? Meanwhile, we ignore the obvious: The entire real estate market that Zillow aims to optimize is a system by which society segregates itself to preserve social hierarchies, such as class and race. Who in our society is actually entitled to their dreams?
What I have described as a new algorithmic American dream, in which we can use data-driven decision making to reduce the uncertainty in our lives without considering the impacts on others. It has found one of its most insidious manifestations in real estate. Zillow surfing is merely a heightened manifestation of this dream, absurd enough to be entertaining and appropriately distracting for weekend readers of the New York Times.
We do not need to give up on data—provided it is used responsibly and locally—as a means of sharing what we know about the places we live. In All Data Are Local, I suggest approaching unfamiliar data as a starting point, a source of questions, and an opportunity to get closer to the people and places beyond data. "Do not take the availability of data," I caution, "as permission to remain at a distance" [4]. For those who engage in Zillow surfing, adopting this attitude could be transformative. If we give up the fantasy of the right house in the right neighborhood, we can use data to awaken our curiosity about the neighborhoods where we have never considered living and the people we might meet there.
References
1. Lorenz, T. Zillow surfing is the escape we all need right now. The New York Times. Nov. 19, 2020; https://www.nytimes.com/2020/11/19/style/zillow-surfing-home-listings.html
2. Pariser, E. The Filter Bubble: How the New Personalized Web Is Changing What We Read and How We Think. Penguin Books, New York, 2012.
3. Richardson, R. Racial segregation and the data-driven society: How our failure to reckon with root causes perpetuates separate and unequal realities. Berkeley Technology Law Journal 36, 3 (2022); https://ssrn.com/abstract=3850317
4. Loukissas, Y.A. All Data Are Local: Thinking Critically in a Data-Driven Society. The MIT Press, Cambridge, MA, 2019.
5. Finn, E. What Algorithms Want: Imagination in the Age of Computing. The MIT Press, Cambridge, MA, 2018.
6. Loukissas, Y.A. Let's change the way big data present the places we live. Big Data & Society: Essays and Provocations. Aug. 15, 2019; https://bigdatasoc.blogspot.com/p/essays-and-provocations.html
7. Chan, A.S. Networking Peripheries: Technological Futures and the Myth of Digital Universalism. The MIT Press, Cambridge, MA, 2016.
8. How do I remove my home from Zillow? Zillow Help Center; https://zillow.zendesk.com/hc/en-us/articles/360058140754-How-do-I-remove-my-home-from-Zillow-
9. Baudrillard, J. Selected Writings. Stanford Univ. Press, Stanford, 2001.
Author
Yanni Alexander Loukissas is an associate professor of digital media in the School of Literature, Media, and Communication at Georgia Tech. His current research interests include participatory mapping, critical visualization, data studies, and smart cities. He is the author of two books, All Data Are Local: Thinking Critically in a Data-Driven Society (The MIT Press, 2019) and Co-Designers: Cultures of Computer Simulation in Architecture (Routledge, 2012). [email protected]
Copyright held by owner/author
The Digital Library is published by the Association for Computing Machinery. Copyright © 2022 ACM, Inc.
Post Comment
No Comments Found