Discrimination in Real Estate
Discrimination is an important problem facing many countries. Using tools from economics, researchers are able to explicitly quantify it.
Thank you for reading our work! If you haven’t yet subscribed, please subscribe below:
If you would like to support us further, please consider sharing and liking this article!
Welcome to all new subscribers! If you’d like us to cover a topic, feel free to leave a comment!
With the recent Supreme Court decision on affirmative action, discrimination is being widely discussed in the news and social media. Although the topic of discrimination seems like it would be a more ‘descriptive’ issue rather than a quantifiable one (i.e. how much discrimination there is), economists and sociologists have used many methods to establish numerical estimates of the costs and damages of discrimination. We will discuss some of this research, focusing especially on the housing and rental markets.
House Price Differences
Discrimination in the marketplace leads to situations where a minority has access to a smaller set of choices. In the housing market, this means that minorities have fewer homes available to them for purchase or rent. Theoretical work by Becker (1971) argued that any form of discrimination, whether centralized (any explicit restrictions usually via institutions) or decentralized (implicitly done by individuals such landlords and real estate agents), that restricts the set of house choices (or any good or service) for minorities, will result in an increased price for these individuals. This result is intuitive. Non-minorities have access to all houses, while minorities only to a subset of houses. This subset of houses therefore faces larger demand since both minorities and non-minorities can bid for these houses.
Centralized discrimination is rarer, as it is easier to be noticed – explicit rules that discriminate are generally easy to spot. Decentralized discrimination on the other hand is harder to determine. However, economists have found ways to elucidate this form of decentralized discrimination.
“Steering”
Christensen and Timmins (2021) looked at decentralized discrimination in the real estate market. They used a paired-actor audit study conducted by the Urban Institute in conjunction with the Department of Housing and Urban Development in 2012 (Turner et al., 2013). In this paired-actor study, two actors that differ only in one observable characteristic, which in this case was race, were given a particular rental or sale listing advertisement, and then were randomly assigned a real estate agent to whom they reached out about this listing. Each actor in the pair would then record their experiences looking for houses with their real estate agent. This paired-actor design was repeated nation-wide and the results were collated.
The study found that incidence of the most blatant forms of discrimination – such as the denial of appointments or the refusal to show a house – have drastically reduced in the last 50 years. However, there were more subtle forms of discrimination.
Firstly, there was a difference in the number of housing units shown to white and minority actors, as shown below:
On average, minorities were shown about 0.5 fewer homes than white individuals. Beyond discrimination in quantity, Christensen and Timmins also analyzed the quality of housing that was shown to the different actors. To visualize the differences between recommended homes, below is a representation of the recommendations in Chicago:
The black dot is the advertised home with which the tester reached out to the real estate agent. The red dots are homes that were recommended to the white actor, while the blue dots are the homes shown to the asian actor. The green dot is the only home shown to both, even though both actors were identical regarding their stated preferences and incomes.
On average, Christensen and Timmins found that minority applicants were recommended houses in school districts that had considerably lower test scores and lower school ratings. Recommended homes were also more likely to be located in areas with higher poverty, fewer high skilled and college educated workers, and more single-parent households. Additionally, minority applicants are steered towards areas that have higher pollution.
The steering effect was even more pronounced in two scenarios: 1) when minority-testers showed preference towards high income, low poverty, predominantly white neighborhoods; and 2) when minority-testers were told to pretend they had a family with children. In both these instances, steering by real estate agents was stronger – the recommended houses were less reflective of the preferences exhibited by the actors. African-American and Hispanic mothers were shown houses in school districts that had test scores that were 65% to 70% lower than the school districts shown to white mothers. Moreover, the recommended neighborhoods also had much higher incidence of assault (40%) and higher poverty rate (4.7%).
These results are especially damaging, as there is significant evidence that growing up in a neighborhood with low poverty rates, when controlling for all other factors such as income, increases income mobility (Chetty et al. (2018)). Moreover, since the real estate agents steer towards areas with worse schooling outcomes, this further compounds the problem of intergenerational mobility, making it harder for children of minorities to move up the income ladder. Lastly, since the recommended areas also have higher pollution, research has shown that this has significant impacts on child development. Persico et al. (2016) showed by studying Superfund sites (these are areas of high pollution in the US that require long term environmental remediation) that children in areas with higher pollution are more likely to repeat a grade (by 7 percentage points) and be suspended (by 6 percentage points), as well as have lower test-scores.
Cause of the Discrimination
So what pushes these real estate agents to steer towards these neighborhoods. There are two potential causes – ‘true’ bias and ‘statistical’ bias. ‘True’ bias occurs if the person explicitly discriminates – that is they explicitly dislike the minority. In ‘statistical’ bias, an individual attempts to infer something about the person they are interacting with based on their assumptions of the group they perceive the individual belongs to. In the real estate example, the real estate agent may assume that the minority actor may prefer to live in a neighborhood where the minority population is more represented. For African-Americans, Christensen and Timmins found that they were shown houses in neighborhoods that had, on average, a 4 percentage point lower share of white homeowners. Whereas the houses shown to white actors closely match the explicit preferences reflected in the meeting with the real estate agent, minority actors receive options that do not closely reflect their choices. This suggests that real estate agents are guided by statistical bias (or statistical discrimination), as they lean more on the group identity rather than the preferences of the individual.
It is worth noting that the real estate agents are most likely acting ‘rationally’ from their perspective – that is, they are not acting based on ‘true’ discrimination. Real estate agents benefit when they make a sale. They may believe that when recommending houses to minorities, they are more likely to get a sale when they show specific neighborhoods, especially neighborhoods in which the buyer’s minority is more prevalent. This could even come from prior experiences of these real estate agents when making sales to minorities.Thus, from the perspective of the real estate agents and based on their beliefs, they are making the optimal decision for themselves. The issue is that beliefs and statistical inferences of these real estate agents, which are stemming from their experiences, are incorrect.
Who Bought the House
The above research design was an audit study, where no actual transaction occurred – the actors did not buy any of the houses. This is naturally a weakness of these types of studies. Interestingly, Christensen and Timmins decided to follow up and see what happened to the recommended houses. Over half of the recommended houses (around 10,000) ended up being purchased. However, data on the ethnicity of the buyer is not publicly closed. However, the name of the buyer is disclosed. Chrisetensen and Timmins used an algorithm (Imai and Khanna, 2016) “to classify buyer identities on the basis of the probability that a buyer’s first/last name is associated with a given racial/ethnic group”.1 Using this information, Christensen and Timmins found that the homes recommended to Hispanic and Asian actors were disproportionately purchased by buyers of the same race (i.e. Hispanics and Asians).
Price of the House
A common argument put forth against any statistical discrimination concerns is that although the recommended house might have certain worse characteristics (called ‘amenities’) identified above, such as being located in a worse school district or in a more polluted area, the house itself might be cheaper. This would be a compensating differential – a person receives worse amenities, but pays less, and thus from an overall perspective, the minority individual might still be fine. Clearly, a compensating differential can only occur if the house is cheaper since the quality of amenities is worse. Although not a rigorous approach, Christensen and Timmins looked at the sale prices of the recommended houses and found little evidence of a compensating differential. This result may suggest that the Becker mechanism described above (that minorities pay higher prices than they should because they have a limited set of alternatives) does occur.
Renter Markets and Correspondence Studies
Audit studies are not perfect experimental methods, as there will always be certain differences between the actors that cannot be controlled for. Moreover, the actors themselves are aware of what the study is aiming to establish, which could influence their behavior. To deal with some of these issues, correspondence studies are used. In a correspondence study, interactions are not conducted in-person, but mainly via online messaging. Ewens, Tomlin and Wang (2014) conducted a large correspondence study in the US rental market. They sent out rental applications to over 14,000 rental properties, where the only element they would change would be the name of the applicant. Their results suggested that statistical discrimination (rather than explicit discrimination) was present, with African-American sounding names receiving 9.3 percentage points fewer positive responses than applications with white sounding names.
Statistical Discrimination
Statistical discrimination, as described above, is a type of ‘stereotyping’ where people use their beliefs about a group, to inform their decision regarding an individual. Bertrand and Duflo (2016), in their comprehensive summary of the discrimination literature, discuss whether the difference between ‘true’ and statistical discrimination matters. The issue is that statistical discrimination can end up becoming a self-fulfilling prophecy. For example, if individuals assume that a minority worker on average is less productive or worse than a white worker, minority workers may have little incentive to put in extra effort, since it will not be noticed by their managers anyways. A similar pattern can occur regarding education – minority workers may have less incentive to acquire more education, because even if they list their qualifications on the resume, it does not improve their likelihood of getting a job. Additionally, statistical discrimination can impact how much time teachers and managers commit to minority students or employees. Since, they assume minorities are worse, they spend less time with them, resulting in minorities receiving less education or quality experience.
This was shown to happen by Glover, Pallais and Parriente (2017). They found that minority cashiers at a grocery store performed differently when working with a biased versus unbiased manager (bias was determined via an Implicit Association Test). With a biased manager, minority workers were absent more often, spent less time at work and worked slower. One of the reasons for these issues was that the biased manager interacted less with minorities, resulting in them exerting less effort. However, once the minority cashiers were matched with an unbiased manager, they worked better than the majority cashiers. Thus on average, both minority and majority cashiers appear to have the same productivity. But since the minority cashiers worked better than the majority cashiers in an unbiased setting, this implies that minority cashiers were actually higher productivity workers.
This is an example of statistical discrimination. Since the grocery firm only observes the average productivity of workers (they are unaware that some managers are biased), the minority cashiers the grocery stores hire are inherently better than the majority cashiers. The problem for the grocery store is that they have biased managers that reduce the performance of their minority cashiers. With unbiased managers, the grocery firm would perform better.
Additionally, the behavior of biased managers impacts the career trajectory of the minority workers since they do not receive as much on-the-job learning, as biased managers simply interact with them less.
Conclusion
Economics and economic methods have a lot to contribute in the discrimination discussion. Although many of the most explicit forms of discrimination have disappeared, discrimination still exists and has significant long-lasting impacts. Statistical discrimination demonstrates how pervasive and damaging the stereotyping of minorities is. Even though individuals, as in the case of real estate agents, might not explicitly discriminate or even want to discriminate, an individual's subconscious (or conscious) beliefs can still have significant adverse impacts on minorities.
Economists have shown this form of discrimination exists in many fields, of which we mainly discussed housing, and has very high costs to minorities. Via ‘steering’, real estate brokers not only adversely affect the house buyer, but are also adversely impacting the outcome of the children of these buyers. This can perpetuate the negative outcomes faced by minorities over generations. The self-fulfilling aspect of discrimination is a significant threat to the improvement of outcomes for minorities.
Interesting Reads from the Week
Tweet: With the expiration of the Child Tax Credit (which we discussed here), child poverty in the US jumped significantly.
Tweet: Labor participition for Americans with disabilities is significantly above trend – 1.7mln more Americans are working, most likely driven by the ability to work from home.
Note: My commentary on a bad analysis of housing prices conducted by the Financial Times.
Cover photo by David McBee:
If you enjoyed this article, you may also enjoy the following ones from Nominal News:
Community Civilian Interventions (June 12, 2023) – cities have always been interested in improving safety in neighborhoods. Civilian government interventions have been proposed as a solution instead of police. However, civilian interventions work best if local government presence is strong.
Marriage Preferences and Gender Outcomes (March 26, 2023) – how partner preferences impact the glass ceiling.
Government Debt - Should we worry about it? (January 23, 2023) – many are concerned that countries have very high government debt. Why we should focus on debt as a ‘tool’ for conducting inter-generational policy and what are sustainable levels of debt.
The power of this algorithm depends on the differentiability level of the names – that is the difference in probability of a name being associated with the most probable racial/ethnic group and the second most likely racial/ethnic group. For example, if “if the algorithm assigns a predicted probability of 80% Asian, 10% white, and a total of 10% for all remaining groups combined, then the name receives a differentiability score of 80%-10%=70%. If the algorithm assigns a predicted probability of 80% Asian, 18% white and a total of 2% for all other groups, then the name receives a differentiability score of 80%-18%=62%. “The highest differentiability was found for Asian and Hispanic names.