<- All Student Work & Projects
Students at St. Kate’s are engaging with the WTDN in a variety of classes. Here’s how our students are thinking about the issues related to housing inequality and housing segregation. Their work demonstrates “thinking in progress”.
The Effect of Racial Covenants on Owner and Renter Housing Occupancy: Evidence from the Twin Cities
Rin Kilde | Fall 2022
Racially restrictive covenants in housing deeds have lasting effects on the markers of wellbeing such as health and homeownership that we see across neighborhoods today. The more we know about these disparities, the more tools we have to work towards decreasing them. One such marker is population density, which the COVID-19 pandemic has shown to be important for issues of disease transmission. Another marker is home ownership. In this analysis, I use Ordinary Least Squares regression to assess the correlation between racial covenants and the housing trends in Hennepin and Ramsey county, Minnesota, such as owner and renter housing occupancy numbers. I find a statistically significant positive relationship between racial covenants and my largest measure of non-relative housing occupancy, and trends showing increased ownership and decreased rentership in tracts with covenants.
Racial covenants, also called racially-restrictive covenants or simply race covenants, are clauses that were written into housing deeds mandating that Black, Indigenous, and People of Color (BIPOC) were not permitted to live in certain homes. The covenants were predominantly used from the 1910s until they were made unenforceable in the late 1940s, and were widespread in many Northern U.S. cities. They functioned to create exclusively White spaces, and those spaces have remained largely segregated since. Through segregation, these communities were also subjected to disparities in treatment from government and private interests which became institutionalized and remain in place today (Mapping Prejudice n.d.)
Over the past few years, the topic of racial covenants and their impact has begun to be researched and understood in a much broader way than in years prior. Up until recently, the study of racial covenants was largely within the realm of qualitative data. Historians and sociologists like Jones-Correa (2000), Gotham (2000), and Brooks (2011) have used qualitative research methods to uncover the origins, spread, enforcement, and lasting impact of racial covenants. However, in more recent years, the work of the Mapping Prejudice and Welcoming the Dear Neighbor? (WTDN) projects have uncovered an entirely new way to study covenants and their impacts. Through extensive crowd-sourced research, WTDN and Mapping Prejudice have produced comprehensive datasets describing the number and location of racial covenants in Minnesota’s Hennepin and Ramsey counties. These datasets allow for quantitative research to support the prior and ongoing qualitative studies.
Through the use of the Minnesota covenants data, researchers have been able to link historical discriminatory housing practices to a plethora of modern-day disparities. In the article “Long Shadow of Racial Discrimination: Evidence from Housing Covenants of Minneapolis” Sood et al. (2019) focus on the impact of racial covenants on modern-day housing outcomes. They find that historically racially covenanted houses had 4-15% higher prices in today’s market, and that tracts with racial covenants had decreased rates of black occupancy and homeownership. These disparities are only the beginning, however. Other researchers have since found correlations between racial covenants and home foreclosures (Whipple 2021), policing (Delgado-Palma & Boie 2020), and health and economic outcomes (Nitschke Durben, Stefanovich, & Ngangsic-Asongu 2021). My research is driven by two housing-related disparities: disease transmission and generational wealth.
The issues of disease transmission and generational wealth are of great importance today. In the context of the COVID-19 pandemic, many of the disparities enforced through racial covenants and other forms of structural racism were harshly exposed. Sharif and Dey (2021) find population density to be a major factor in the spread of COVID-19. For generational wealth, Pfeffer and Killewald (2018) determine homeownership to be one of the top five ways in which wealth passes from parents to children. Through my analysis, I seek to determine the nature of the relationship between racial covenants and housing occupancy and ownership trends. My key explanatory variable is the share of houses in a tract that are racially covenanted, and my key response variables are a variety of measures of ownership and occupancy. I hypothesize that 1) there will be a negative correlation between the share of covenanted homes and the number of non-relative household occupants and 2) that there will be a larger share of homeowners and a smaller share of renters in areas with more covenants.
There are two theoretical economic models that I am applying to the question of racial covenants and housing. The first is a supply and demand model. In a free market, prices are set by the intersection of supply and demand, where the price at which consumers will buy and the price at which producers will sell are aligned. However, the racially covenanted housing market is not a free market, since the covenants restrict BIPOC from buying certain houses, regardless of the price. Meanwhile, white buyers have a broader range of housing options. Racial covenants were also often a selling point for White buyers as well, so some White buyers were willing to pay more for the same house if doing so excluded BIPOC from the community. In this way, the loss that BIPOC in the housing market experienced and the higher rates that white buyers paid were transformed into profits for white landlords and house sellers.
Fig. 1: Supply and demand graph where E1 is equilibrium without covenants, E2 is equilibrium for BIPOC with covenants, and E3 is equilibrium for White buyers with covenants.
The second model is a budget constraints model. Racial covenants worked in two ways, by restricting housing and by creating geographic lines along which to provide preferential or disfavorable treatment by White businessmen and policy makers. The supply and demand model shows the impact on the housing market, but the budget constraint additionally shows the impacts of disparities along racially covenanted boundaries.
Fig. 2: Budget constraint graph
In order to conduct my research, I retrieved and merged two data sets. The first was publicly available data from the Integrated Public Use Microdata Series (IPUMS), specifically the National Historical Geographic Information System (NHGIS). The NHGIS data set included tract-level information for Hennepin and Ramsey counties of Minnesota with several population demographic variables including individual counts in some cases and household counts in others. I use this data for my outcome variables, measures of unit occupancy and population, and control variables. For the sake of relevance and completeness, I use cross-sectional data from the 2010 census. The second data set I use is from the Mapping Prejudice project, and provides a count of racial covenants by tract for Hennepin and Ramsey counties. The data had been organized by address and was then aggregated into tract-level. I transformed each variable that I used from counts to shares using the variables for total population and number of housing units.
Merging the two data sets may have resulted in some error. Given that many tracts do not have any racial covenants, they appeared as missing data points. I replaced these missing points with the number 0 since they are the observations of tracts without racial covenants. However, if any of the observations had missing data points for other reasons, this may have resulted in an error in the number of racial covenants in that tract. I also dropped two census tracts because they were listed as having no residents in the 2010 census. I also chose to delete one census tract which had a count of 444 historical covenants, but only 180 total housing units today, which resulted in an unusable share value. After cleaning, my data set contained 433 unique observations of tracts. Many of the variables in my data set showed skewed distributions.
Fig. 3: Shares of tracts for racial covenant and demographic data
The nonstandard distributions seen in these graphs are all heavily skewed right, except the variable for population who are White, which is skewed left. The shape of this data indicates that for the presence of racial covenants and for all race categories except White, the majority of tracts have these variables primarily present in small numbers. My measures of non-relatives in households also displayed skewed trends.
Fig 4: Non-relatives in households by share
All of my variables showing nonrelatives in households are skewed right, which illustrates that the majority of tracts have no homes or only a few homes with nonrelative occupants. Although it is still skewed, the data for all nonrelatives shows the highest upper bound percentage, with some tracts showing up to 54.9% of housing units to have nonrelative occupants. Similar trends also appeared in my measures of owner and renter occupancy.
Fig. 5: Owner- and renter-occupied homes in aggregate, for 1-person households, and for households of 7 or more
This data illustrates the distribution of housing trends. Aggregated data for homes that are owner occupied and homes that are renter occupied show mirrored trends. The bulk of tracts have more owned than rented homes, while rented homes are predominantly in the minority. These trends are less strong on the disaggregated 1-person households, and households with 7 or more residents are right skewed for both owners and renters.
For my study, I use Ordinary Least Squares (OLS) regressions, with additional analysis performed on the results thereof. The basic regression I use is:
Housing Occupancy = α + β1SRaceCov
There are many factors which influence individuals’ decisions on whether or not to cohabitate and whether to rent or buy a home. It is impossible to control for every consideration in making housing decisions. However, racial covenants harmed communities by racially segregating them, and then providing unequal amenities and opportunities for those communities or compounding the inequalities experienced by its inhabitants into one place. As such, it is possible to isolate the disparities caused by racial covenants by controlling for race. In doing so, we can see in isolation the factors correlated with being in a racially covenanted tract, rather than the factors correlated with race or racialization. Once the impacts of racial covenants have been isolated, we can then observe that they disproportionately impact communities of color since those communities are more likely to be non-covenanted. As such, I have created a set of regression with controls for race as defined in my data set. I also include a robustness check to obtain heteroskedasticity robust standard errors. The model I use for these regressions is:
Housing Occupancy = α + β1SRaceCov + β2SBlack + β3SAIAN + β4SAAPI + β5SMixed + u
Variable SWhite, the share of residents who are White, is omitted due to perfect collinearity. I use this model to run regressions with twenty measures of housing occupancy as the outcome in order to show differential effects. I have four measures of non-relative occupancy, and two sets of measures for owner and renter occupancy. These variables are defined as follows:
- SNonRelAll: Share of residents in a tract who are additional non-relative occupants in a household.
- SNonRelFam: Share of residents in a tract who are additional non-relative occupants in a family household, specifically.
- SNonRelFamRB: Share of residents in a tract who are additional non-relative occupants in a family household who are listed as a roomer or boarder.
- SNonRelFamRM: Share of residents that are non-relative roommates in a family home.
- SOwnOccAll and SOwnOcc1 to SOwnOcc7up: All owner-occupied housing units and units disaggregated by number of residents from 1 resident to 7 or more residents.
- SRentOccAll and SRentOcc1 to SRentOcc7up: All renter-occupied housing units and units disaggregated by number of residents from 1 resident to 7 or more residents.
Of my twenty regressions, twelve proved statistically significant. The outcome variables which produced statistically significant results out of my measures of non-relative occupancy were SNonRelAll and SNonRelFamRM. From my measures of renter and owner occupancy, the disaggregated forms for one- to four-person households and the aggregated forms were statistically significant. The coefficients, standard errors, and significance levels are displayed in table 1, with a regression without controls (1), with controls for race (2), and controls for race and a heteroskedasticity robustness check (3).
Table 1: Regression with outcome SNonRelAll
The magnitude and sign vary across the regressions. Out of the measures of non-relative occupancy, SNonRelAll was the only variable with a negative sign or a magnitude greater than |0.005|. The coefficient of SNonRelAll shows that for an increase of 1 (or 100%) in racial covenants, there would be a correlated decrease of 3.9% in the number of non-relatives in homes in that tract. There were also differentiations in the results of the race controls within individual regressions. The regression for the outcome variable SNonRelAll demonstrates these differentiations.
Table 2: Regression with outcome SNonRelAll
In this regression, only SAAPI is not significant. The coefficients of controls SBlack and SAAPI are negative, while SAIAN and SMixed are positive. SMixed has the largest magnitude, followed by SAIAN. The R2 shows that addition of controls is important for the robustness of the regression, since regression 1 only explains 0.5% of the findings, while regression 3 explains 16.9% of the findings. Without statistically significant or high-magnitude findings in my other measures of non-relative household residency, I cannot determine more trends. However, the differential coefficients of my other set of variables, those measuring owner and renter occupancy, do show distinct patterns.
When looking at the results of the regressions of racial covenants on owner and renter occupancy measures, there are two noticeable trends. Firstly, owner occupancy has coefficients that are positive while renter occupancy has coefficients that are negative. Secondly, the magnitude of the coefficients decreases as the number of residents in the home increases. For 1-occupant households, an increase of one (or 100%) in racial covenants is correlated with a 12.6% increase owner-occupied homes, but in the case of houses with seven or more occupants, an increase of one in racial covenants is only correlated with an increase of 0.1% in owner-occupied homes. That findings for OwnOcc7up, as well as the increases for five- and six-person homes, is not statistically significant. A mirroring trend is visible in renter-occupied homes. An increase of one in racial covenants is correlated with a decrease of 22.2% in single-resident, renter-occupied homes. For homes with seven or more residents, there is actually a positive correlation, though the coefficient is negligible at 0.002, and the finding is not statistically significant.
To illustrate the change in correlation between households and racial covenants as the number of people in the household goes up, I have plotted the coefficients for each regression against the number of occupants in the home, displaying the relationship on a scatter plot with a line of best fit. For the first set of variables, owner-occupied households by number of residents, we see a strong negative trend.
Fig. 6: Homeowners and household residents
As the number of residents increases, the magnitude of the effect that racial covenants have on occupancy numbers decreases, approaching 0. Low numbers of occupants show the highest positive coefficients.
For the second set of variables, renter-occupied households by number of residents, we see the opposite, a strong positive trend. The coefficients start at well below zero, then diminish in magnitude to statistically insignificant values near zero as the number of occupants increases.
Fig. 6: Renters and household residents
As the number of residents increases, the magnitude of the effect that racial covenants have on occupancy numbers decreases, approaching 0. Low numbers of occupants show the lowest negative coefficients.
The same information is displayed here, showing the disparate relationships with lower numbers of residents converging as the number of residents increases.
Fig. 7: Comparing ownership and rentership
The findings of my analysis do support my hypotheses, albeit with some areas that remain ambiguous. My first hypothesis was that budget constraints caused by the restricted housing market which racial covenants created for BIPOC would lead to an increase in cohabitation. I found that for an increase of one (or 100%) in racial covenants, there is a correlated increase of 3.9% in non-relative household occupants within the tract. This magnitude is visible in only the most generally defined and widespread count of cohabitation, and not in the more specified variables. The findings from my second hypothesis present more distinct trends.
My second hypothesis was that the budget constraints would lead to renting rather than buying of homes. I find this hypothesis to be strongly supported. In line with the findings of Sood et al. (2019), my analysis shows considerable disparities in housing ownership in relation to racial covenants. Specifically, it showed that these disparities have the highest correlation with trends in smaller households. This has implications for long-term housing trends, since smaller “starter homes” are an important step towards widespread increases in housing ownership, and this analysis suggests there are few opportunities to buy such homes in racially covenanted communities. It is difficult to tell how much of the truth is captured in the budget constraints model. While it is possible that renting instead of buying was the result of high budget constraint, it is also quite possible that developers chose to build rental housing in covenanted areas, and it is likely that individuals in covenanted areas have a harder time getting approved for home loans due to redlining practices. It is probable that all of these and many more factors were at work.
- Limitations and Future Research
There are aspects of my data and model that restrict its implications. Using tract level data leads to findings that are inexact. The ideal data set in this case would include household-level observations. Additionally, it would be beneficial to have information on the characteristics of the housing units, such as square footage, lot size, and proximity to community resources. Lacking these, my analysis likely has considerable noise from the natural variation between areas of the Twin Cities. Another way to address this issue would be to use data from a broader sample of cities that used racial covenants. However, a larger sample size cannot correct for all of the issues present.
In addition to the drawbacks of tract-level data, there were also complications due to how race is categorized in the NHGIS data, and how it functions given the small sample size. The variable SMixed, which is the share of individuals who are listed on the census as “two or more races,” behaves within the regressions in a manner which suggests there are aspects of the data I am not accounting for. The variable SAAPI is the share of individuals listed as “Asian American and Pacific Islander,” but was aggregated in the original data set with the category “all other races,” suggesting that it is an imprecise measure. Finally, the variable SAIAN, representing the share of the population who are American Indian or Alaskan Native, is complicated in this context by the fact that Hennepin county is home to the Mille Lacs Ojibwe community, a group which experienced many of the same disparities as communities shaped by racial covenants. However, rather than racial covenants, the tool which was used to segregate the Mille Lacs Ojibwe band was exploitative treaties, and so data analysis based solely on covenant data does not tell the full story.
The findings of this analysis show disparities in housing trends which support a case for affirmative action in housing. Developers, banks, and White homeowners all profited off of the racial covenant system. One policy possibility might be to increase property taxes on covenanted land and business taxes on financial institutions that participated in redlining. In turn, those funds could be used to support new homeowners in areas most negatively impacted by covenants, or the descendants of those neighborhoods’ inhabitants during the time when racial covenants were being enforced. These changes or others like them that address the legacy of racial covenants will likely be a long time in the making.
Empirical research into the quantitative impacts of racial covenants is a field still in its infancy. However, early research on the topic is beginning to display trends in its findings. We see that across the areas the data covers, racial covenants are correlated with disparities beyond those which can be explained by natural trends in community building. My research shows the same thing, pointing to some increases in housing unit occupancy and a considerable double standard in housing ownership. Within these discouraging findings, however, there is an opportunity. When empirical research can put numbers behind the trends of systematic oppression, the need for change becomes statistically definable.
Brooks, R. R. W. (2011). Covenants without Courts: Enforcing Residential Segregation with Legally Unenforceable Agreements. The American Economic Review, 101(3), 360–365. doi.org/10.1257/aer.101.3.360
Delgado Palma, V, & Boie, M. (2020). Policing and Racial Covenants in Minneapolis. Welcoming the Dear Neighbor?, Mapping Prejudice. welcomingthedearneighbor.org/2022/03/10/policing-and-racial-covenants-in-minneapolis
Gotham, K. (2000). Urban space, Restrictive Covenants and the Origins of Racial Residential Segregation in a US City, 1900-50. International Journal of Urban and Regional Research 24 (3), 616–633.
Jones-Correa, M. (2000). The Origins and Diffusion of Racial Restrictive Covenants. Political Science Quarterly, 115(4), 541–568. doi.org/10.2307/2657609
Mapping Prejudice. (n.d.). What is a covenant? What is a Covenant? Retrieved December 15, 2022, from mappingprejudice.umn.edu/racial-covenants/what-is-a-covenant
Nitschke Durben, A., Stefanovich, E., & Ngangsic-Asongu, M. (2021). Mapping Prejudice: The Effect of Racial Covenants on Neighborhoods in Hennepin County. Welcoming the Dear Neighbor?, Mapping Prejudice. welcomingthedearneighbor.org/2022/03/10/mapping-prejudice-the-effect-of-racial-covenants-on-the-neighborhoods-of-hennepin-county/
Pfeffer, F. T., & Killewald, A. (2018). Generations of Advantage. Multigenerational Correlations in Family Wealth. Social Forces, 96(4), 1411–1442. doi.org/10.1093/sf/sox086
Sharif, & Dey, S. K. (2021). Impact of population density and weather on COVID-19 pandemic and SARS-CoV-2 mutation frequency in Bangladesh. Epidemiology and Infection, 149, e16–e16. https://doi.org/10.1017/S0950268821000029
Sood, A., Speagle, W., & Ehrman-Solberg, K. (2019). Long Shadow of Racial Discrimination: Evidence from Housing Covenants of Minneapolis. https://doi.org/10.2139/ssrn.3468520
Whipple, H. R. (2021). The Effect of Racial Covenants on Modern Day Foreclosures: Evidence from Hennepin County. Economics Honors Projects. 109. digitalcommons.macalester.edu/economics_honors_projects/109