Residential Alienation in Minneapolis: Eviction Filings, Judgments, and Housing
Insecurity
MPP Professional Paper
In Partial Fulfillment of the Master of Public Policy Degree Requirements
The Hubert H. Humphrey School of Public Affairs
The University of Minnesota
Patrick S. Alcorn
05/20/2021
Signature below of Paper Supervisor certifies successful completion of oral presentation and completion
of final written version:
____________________________       ___________________ ___________________
Professor Ed Goetz, Paper Supervisor Date, oral presentation Date, paper completion
_____________________________________ ___________________
Professor Angela Fertig, Second Committee Member Date
Introduction
Landlords file over 3,000 evictions in Minneapolis each year (Minneapolis Innovation
Team, 2016). Rising rents, stagnating wages, the financialization of housing, and predatory
landlord tactics have all contributed to the ongoing crisis. Recent research shows that different
social forces drive evictions depending on characteristics of local housing markets. Two of these
processes can be categorized as evictions that occur in non-gentrifying neighborhoods, often
defined by racial exclusion, poverty, and insecurity, and those that occur in a separate process
that is driven by real estate activity and gentrification (Sims 2021). While distinct from one
another, they are both governed by the broader social forces of the rental market.
Madden and Marcuse develop the term “residential alienation,” an adaptation from
Marx’s alienation of labor, as a unifying concept of the distinct ways that insecurity, extraction,
and displacement affect the housing of working-class people. Rather than results of random
occurrence, residential alienation centers housing insecurity as a predictable and structural
outcome of the capitalist housing system. In this paper, I investigate if either of these forms of
eviction-based alienation are occurring in Minneapolis using data on evictions and rental
property sales.
In the first section, I review the literature on evictions, displacement, and gentrification,
and then situate evictions within the concept of residential alienation. This reveals how evictions
are more than the actions of individual landlords pursuing their own interests; evictions are a
political action with devastating consequences, especially for low-income and BIPOC (Black,
Indigenous, people of color) renters.
In the second section, I use a Spatial Durbin Model to test multiple hypotheses related to
evictions and residential alienation. The Spatial Durbin Model is unique in that it allows both
1
direct effects and indirect (spillover) effects to be modeled. Direct impacts are interpreted as the
impact of a change in an independent variable within a geographic unit on the dependent variable
in the unit, and indirect impacts represent the impact of the change in an independent variable in
a geographic unit on the dependent variable in neighboring units. In addition to modeling a
lagged spatial relationship for independent variables, the Spatial Durbin Model also includes a
lagged dependent variable, in this case the rate of eviction judgments or the rate of eviction
filings.
The three hypotheses being tested in this paper are:
1. Eviction filings and judgments in Minneapolis are driven by a process of racial
exclusion and extraction, mostly occurring in census tracts with a higher-share of
Black residents, cost-burdened renters, and an older housing stock;
2. Gentrification is related to higher rates of eviction-based displacement. A
gentrifying census tract will have a higher eviction judgment rate than a
non-gentrifying census tract;
3. There is a spatial spillover in the rate of eviction filings and eviction judgments.
An increase in the rate of filings or judgments in one census tract is related to an
increase of the filing or judgment rate in neighboring tracts.
I model gentrification as a function of real estate activity in the rental market, using the
percentage of rental units sold in a census tract as the key independent variable. I follow the
work of Raymond and separate eviction judgments from eviction filings in the models. If
gentrification is fueling evictions, I expect a discrepancy between the impact of rental property
sales on eviction filing rates—cases filed with a court—and eviction judgment rates, where an
eviction is enforced and the tenant is removed from the property. Eviction judgments reflect the
2
specific intent of a landlord’s attempt to remove a tenant, while eviction filings are reflective of
multiple different processes, such as intimidation, harassment, and debt collection (Sims and
Iverson 2021; Seymour and Akers 2021; Garboden and Rosen 2019; Raymond, 2021). When a
neighborhood gentrifies and is suddenly able to support higher rents, landlords are incentivized
to remove the current tenants and replace them with more affluent ones. An eviction filing that
results in a settlement or dismissal is qualitatively different in purpose than a filing that ends in
an eviction judgment. Only eviction judgments, eviction filings with the intent to remove a
resident, should be significantly associated with gentrification.
Literature Review
Evictions, Displacement, and Insecurity
Housing insecurity has always been a reality for poor, working-class, and BIPOC
communities in the United States (Madden and Marcuse 2016). Tenement housing, slum
clearance, redlining, urban renewal, the demolition of public housing, the foreclosure crisis, and
gentrification all have the commonality of increasing housing precarity for these groups. Both
the immediate causes of evictions and the populations most impacted by them are fairly well
known (Desmond and Gershenson 2017; Sims and Iverson 2021; Soederberg 2018; Garboden
and Rosen 2019; Seymour and Akers 2021; Leung, Hepburn, and Desmond 2020; Immergluck et
al., n.d.; Hartman and Robinson 2003; Harrison et al. 2020). Not surprisingly, the most common
cause of eviction is the inability to pay rent (Hartman and Robinson 2003). The amount of
cost-burdened low-income households has stayed at extremely high rates in recent years, and
more middle-class families are becoming increasingly cost-burdened. In 2018, over 80% of Twin
Cities renter households making under $15,000 were cost burdened, spending more than 30% of
3
their income towards housing expenses, and over 70% were extremely cost burdened, with over
50% of their income going towards housing costs (Moylan 2020). The percentage of cost
burdened households making between $30,000 and $49,999 has grown from just over 40% in
2006 to over 60% in 2017. An important feature of the post-recession era has been the expansion
of housing insecurity beyond solely extremely low-income families.
Multiple studies have shown that evictions disproportionately impact Latina and Black
women, and that single mothers are especially at risk to face eviction (Desmond and Gershenson
2017). Hartman and Robinson (2003) summarize multiple studies from the 1990s and early
2000s that show poor and BIPOC communities as the primary targets of evictions, even in cities
with relatively strong tenant protections such as Oakland and New York City. Evictions in
Minneapolis are similar to national patterns. According to a study from the City of Minneapolis,
almost 50% of tenants in the zip codes that make up the neighborhoods of North Minneapolis
experienced an eviction between 2013 and 2015 (Minneapolis Innovation Team, 2016). Those
two zip codes account for only 8% of the total rental units in the city, but 35% of all eviction
filings.
Recent scholarship has highlighted multiple purposes behind why landlords file evictions.
Eviction filings can be separated into two broad categories—filings that end with displacement
and filings that do not end with displacement. Non-displacement filings are most often reflective
of adversarial relationships between tenants and landlords (Raymond, 2021). However, eviction
filings are not the only form of involuntary displacement. In fact, formal evictions are a
relatively infrequent method of displacement. Research shows that landlords often utilize
informal evictions to get rid of tenants. These methods include increasing rent to an amount the
tenant cannot afford and/or does not want to pay, or “mutually” agreeing to not renew the lease.
4
In Minneapolis, nonrenewal of lease was cited as one of the common loss-mitigation measures
for landlords (Lewis, 2019).
Recent research shows that displacement is not always the principal reason for an
eviction filing. In fact, the number of evictions carried out is far smaller than the number of
evictions filed. While certainly some of these are simply failed attempts to evict a tenant where
the judge rules against the landlord, recent papers provide strong evidence that eviction filings
are used as a threat to intimidate and harass tenants, and for debt collection (Garboden and Rosen
2019). In this situation, the landlord often has no intention of actually removing tenants from
their property.
Evictions are a costly process. The landlord has to pay legal fees and must take the time
to go to court. Replacement tenants need to be found and each month of vacancy is foregone
revenue. Additionally, repairs, maintenance, and cleaning are likely required before a new tenant
moves in. Most often, it is more cost-effective to recover missing rent payments and late fees
through a payment plan. Garboden and Rosen argue that this process transforms the relationship
between tenant and landlord to debtor and collector. By initiating the legal process, the unpaid
rent becomes a legal debt. This is important not only for the economic implications, but because
the looming threat of evictions enhances the power imbalance between owner and renter.
Gentrification
The causes and impacts of modern gentrification have been subject to immense debate.
Many have theorized it to be a return to the city decades following the mass migration out of
cities during the post-war suburbanization period (Easton et al. 2020; Smith 1979;
Brown-Saracino 2017). This perspective understands gentrification to be mostly a cultural and
5
consumer phenomenon, defined by shifting consumer preferences away from stale cookie-cutter
suburbs towards bustling, vibrant city life (Smith 1979). However, this view obscures the
underlying political and economic forces which drive urban revitalization and gentrification.
While the tastes and preferences of consumers are no doubt important to the shaping of housing
markets, others argue that a narrow focus on consumer-demand is incomplete (Smith 1979). The
centrality of the consumer in the study of gentrification frames it as a naturally occurring process
outside of social and political relationships.
Smith was one of the first to fully formulate a systematic theory of how capitalist
urbanization produces gentrification (Sims 2016). He developed the rent gap theory to argue that
capital flows are a primary driver of gentrification. The rent gap refers to the difference between
a piece of land’s current value and its potential future value. The differential is primarily fueled
by the depreciation of existing land values, often caused by divestment and the mass flow of
capital out of a neighborhood. Once the gap sufficiently widens, real estate actors are able to take
advantage of the cheap acquisition of older, derelict buildings. The gentrification process is
touched by a variety of actors: developers, financial institutions, urban governments, landlords,
investors, and of course consumers—or what Stein calls, the real estate state (Stein 2019).
Smith’s theory highlights the symbiotic relationship between production and consumption,
although production dominates the relationship. In contrast to consumer-preference theories of
gentrification, Smith (1979) argues:
Viewed in this way, gentrification is not a chance occurrence or an inexplicable reversal
of some inevitable filtering process. On the contrary, it is to be expected. The
depreciation of capital in nineteenth century inner-city neighborhoods, together with
continued urban growth during the first half of the twentieth century, have combined to
produce conditions in which profitable reinvestment is possible.
6
The relationship between displacement and gentrification is even more complicated and
has been subject to rigorous debate—especially in regards to quantifying displacement. One
perspective is that race- and class-based displacement is inherent to gentrification. Other scholars
have attempted to frame gentrification either as a politically-neutral or even positive process. The
most well known work from this perspective is from Freeman and Braconi. They argue that
displacement is not necessarily an outcome of gentrification; and furthermore, that residents of
gentrifying neighborhoods appreciate the upgrading of neighborhood amenities (Freeman 2005).
However, these findings have been contested by other researchers, who argue that
residents of gentrifying neighborhoods hold complex attitudes towards investment, amenity
upgrading, and displacement. Residents of the Fort Greene neighborhood in Brooklyn were
reported to appreciate the improved infrastructure and amenities in their community but also
thought their neighborhood had become more expensive. They also identified non-economic
costs such as “the departure of many of their friends and relatives, a decline of cultural and
commercial infrastructure...and the fear that they themselves would one day have to leave a
neighborhood that they view as their home” (Chronopoulos 2016).
This is a risk of a solely empirical approach to gentrification. First, the difficulty of
measuring displacement has been well documented (Chronopoulos 2016; Marcuse 1985;
Davidson 2009; Sims 2016; Smith 1979; Easton et al. 2020). Many studies rely on what Marcuse
called the methodology of last-resident displacement (Marcuse 1985). In quantifying
displacement, this method only takes account of the most recent resident to occupy a unit,
ignoring the possibility of previously displaced households—or chain displacement. He argues
that displacement can not only be viewed as these direct forms. Indirect forms of displacement
also need to be considered, the primary two being exclusionary displacement and displacement
7
pressures. Exclusionary displacement refers to those that would otherwise have moved to a
neighborhood, but are kept out through rising rents or perceived hostility from gentrification. Its
impact is upon the other residents in gentrifying neighborhoods that see their changes take place
within the community, newer affluent residents move in, and feel the disruption and dislocation
of their social network.
The research related to the relationship between gentrification and specifically
eviction-based displacement is underdeveloped but growing. Both Freeman and Braconi (2004)
and Newman and Wyly (2006) exclude evictions from their displacement measures in their
studies. Within the research that studies evictions and gentrification, the results are mixed.
Desmond and Gershenson (2017) find no relationship between gentrification indicators and
evictions. They consider gentrification as a function of increased home values and an increased
percentage of the population with a bachelor’s degree. In Toronto, only neighborhoods in the
early stages of gentrification were associated with an increase in evictions, but neighborhoods
that had gentrified over ten years ago had significantly reduced eviction rates (Chum 2015).
Laniyonu (2019) uses a spatial econometric model in his study of Brooklyn.Looking at
eviction filing rates, he finds mixed results depending on how gentrification is measured. In one
measure based on increased rents in a census tract and the percentage of the population with a
bachelor’s degree, he finds a decrease in eviction filing rates. In a second measure, based on the
relative growth of the number of residents working in “post-industrial” occupations, he finds a
significant increase in the eviction filing rate.
Two recent papers focus on the impact of real estate activity on evictions. Sims (2021)
considers the impact of new multifamily construction on eviction filings. He finds that large
multifamily developments produce an increase in evictions within a small radius, a tenth of a
8
mile. Raymond (2021) measures the impact of investor-purchased apartments on eviction
judgment rates. She finds that investor-purchased multifamily units are associated with eviction
spikes in the eviction-based displacement and a substantial decline in the number of Black
residents in a surrounding neighborhood.
Theoretical Framework: Residential Alienation, Social Reproduction, and
Political Economy
This paper contributes to the research on displacement and insecurity by connecting
social relationships to a broader political economic structure. The immediate causes of evictions
are fairly well understood (Desmond and Gershenson 2017; 2016; Hartman and Robinson 2003).
Lack of employment, stagnating wages, rising housing costs, discrimination, exploitative
landlords and investors, the decimation of the welfare state, and divestment from public housing
all have fueled housing insecurity for working-class people.
Soederberg (2018) argues, however, that these explanations do not reveal the fundamental
nature of why insecurity is critical to the production and reproduction of the capitalist political
economy. She argues that these explanations do not “question the root causes of poverty,
questions of power, and political economy dynamics that reside in evictions.” While critical in
understanding the direct causal mechanisms of evictions, they fail to explain the structural forces
driving housing insecurity and affordability. Here, I explore the structural forces behind housing
insecurity by connecting eviction-based displacement and insecurity in its various forms to
Madden and Marcuse’s use of the Marxist concept of residential alienation. Residential
alienation emerges from the total commodification of housing, or the domination of a property’s
use as a home by its value as a financial asset. Madden and Marcuse, referring to the experience
9
of workers under alienated housing, argue, “their housing is the instrument of someone else’s
profit, and this confirms their lack of social power.” Just as Marx argued alienation is a structural
feature of labor under capitalism, they argue that residential alienation is not an accidental
consequence of the moment but a predictable outcome of the political-economic system.
In order to understand residential alienation, two closely-connected relationships need to
be briefly discussed: 1) production and reproduction and 2) use- and exchange-values. The
production process is the generalized process of commodities being created. While Marx had
industrial capital in mind, production encompasses all forms of abstracted labor. Production
requires more than a one-time buying and selling of a worker’s labor-power. It necessitates an
ongoing relationship between an employer and a worker. The worker requires wages adequate to
meet their basic needs. This is what is called social reproduction—the necessities of life for the
worker to reproduce themselves in order to continue engaging in production. In order for
production to continue, workers require a standard of living that meets their basic needs. Of
course what is included in the bundle of basic needs is dependent upon time, place, the level of
development of a society, and political struggle.
The tension between production and reproduction emerges from capitalism’s drive to
extend market principles into all aspects of society, including land and housing. Forrest and
Williams describe capitalism’s expansionary tendencies as the expansion of capitalist
organization not only to non-capitalist economies, but to items within capitalist economies that
previously were outside of market organization or only partially within it (Forrest and Williams
1984). The production/reproduction tension reveals how market logic has come to fully consume
the sphere of housing.
10
Madden and Marcuse discuss the use- and exchange-values in terms of the
commodification of housing (Madden and Marcuse 2016). Simply put, all commodities have
both a use-value and an exchange-value. The use-value is a commodity’s value in terms of its
utility: its ability to meet a specific practical purpose. The use-value of a home is the provision of
shelter, safety, comfort, etc. The exchange-value is the commodity’s value in terms of its ability
to be exchanged, whether for money or a different commodity. The commodification of housing
is the domination of its economic value over its usefulness as a home (Madden and Marcuse
2016; Forrest and Williams 1984). The landlord’s right to collect rent, an investor’s right to a
return on their investment, the bank’s right to collect mortgage payments all take precedence
over a tenant’s “right” to remain in their home.
The contradictions in the production/reproduction and use-/exchange-value relationships
are critical to understanding alienation. Broadly speaking, alienation is the separation of
something we own from ourselves. When a worker enters into a contract to provide labor to an
employer in exchange for wages, they are alienated from their labor. They engage in
production—whether physical, intellectual, or emotional—for their employer but have no right
to that commodity or the value contained within it. Alienation also removes their autonomy in
the workplace. Somebody else controls their labor for the duration of the time spent laboring.
Marx describes the impact alienation on the worker in the following passage:
“First, the fact that labor is external to the worker, i.e., it does not belong to his intrinsic
nature; that in his work, therefore, he does not affirm himself but denies himself, does not
feel content but unhappy, does not develop freely his physical and mental energy but
mortifies his body and ruins his mind. The worker therefore only feels himself outside his
work, and in his work feels outside himself. He feels at home when he is not working,
and when he is working he does not feel at home. His labor is therefore not voluntary, but
coerced; it is forced labor. It is therefore not the satisfaction of a need; it is merely a
11
means to satisfy needs external to it . . . Lastly, the external character of labor for the
worker appears in the fact that it is not his own, but someone else’s, that it does not
belong to him, that in it he belongs, not to himself, but to another.” (Marx, 1844)
Marx argues that the commodification of labor itself degrades the act of working to a point
where it is fully divorced from the worker. Not only is the product of labor alienated from the
worker, but the act of laboring itself becomes estranged. It no longer is a fulfillment of one’s
humanity but a necessity of survival. However, in this piece Marx limits alienation to the sphere
of production, even making a distinction between the workplace and home. The passage thus
only accounts for one side of the work and home dichotomy.
The subsumption of housing by the market erodes the distinction between the workplace
and the home, and alienation follows the worker back into their dwelling. Madden and Marcuse
outline a definition of residential alienation. They argue that “housing is a universal human
practice” and “an extension and expression of our capacity to create ”(Madden and Marcuse
2016). Alienated housing is a result of commodification, an outcome of the tensions between
production and reproduction and use- and exchange-values. Instead of a place of refuge, comfort,
shelter, and revitalization, the home becomes a site of further extraction. They state, “the
experience of residential alienation in contemporary society, therefore, is precarity, insecurity,
and disempowerment” (Madden and Marcuse 2016).
Eviction-based displacement is a concrete manifestation of alienation. Madden and
Marcuse detail the violence of evictions and their harmful effects on evicted renters. Households
that are evicted face negative physical and mental health consequences, strained social and
professional relationships, further precarity, and harmful impacts for their children (Madden and
Marcuse 2016; Hartman and Robinson 2003).
12
Furthermore, evictions are not only a momentary event, but an ongoing set of social
relationships between tenant and landlord. Take Garboden and Rosen’s serial eviction
framework, for example. They define evictions as, “lasting from the time the landlord informs
the tenant that she is late on her rent until she either pays off her debt or leaves the unit.” Defined
like this, many evictions never actually reach the point of removing a tenant from their home.
Instead, they find that landlords serially file evictions not to remove them, but solely to collect
unpaid rent and often an additional late fee. For many landlords, it is more costly in the long run
to remove a tenant and go through the process of filling a vacancy than it is to receive only a
partial payment from an existing tenant. Garboden and Rosen argue that this transforms the
relationship between landlord and renter into creditor and debtor and works to legitimize “more
direct state intervention” (Garboden and Rosen 2019). This transformation can not only put a
tenant into a perpetual state of arrearage, but also creates the looming threat of eviction. The
serial eviction process shows that residential alienation is not only related to displacement, but
the very structure of the landlord/renter relationship itself.
Gentrification-driven eviction-based insecurity is a particularly clear manifestation of
residential alienation. The nature of the housing market is that the exchange-value of a home can
increase or decrease simply based on its surrounding location. In an appreciating market,
landlords are incentivized to replace current tenants with higher income tenants. Renters in
gentrifying neighborhoods therefore might see rent increases, despite no material improvements
being made to their dwelling. Another example is the landlord who evicts a building of tenants in
order to renovate the building in order to fully capitalize on a hot market. The experience is often
all the same for the renter, the lack of control over their home leads to insecurity or displacement.
Alienation is not only felt at the individual dwelling level, but throughout a neighborhood and
13
beyond. Long term residents of gentrifying neighborhoods experience an influx of newcomers,
cultural loss, estrangement from a community, and often increased policing.
Data
4th District Housing Court Eviction Data
The data on evictions comes from the Housing Court of the Minnesota 4th District Court. The
initial dataset contained 259,120 observations from January 2007 to October 2020, with
observations for each the plaintiff (the property owner) and defendant (the tenant) in the court
cases. Some cases had multiple defendants, where more than one resident of a unit was served an
eviction notice. Because this study is focused on who is being evicted, not those who are filing
the evictions, only the defendants were included in the sample. This left approximately 140,000
observations that were identified as defendants.
The original data contained no reference to the type of property that an eviction filing
occurred, with both residential and commercial evictions taking place. In order to remove as
many commercial properties as possible, any filing where the defendant’s name contained the
following terms was excluded: inc, llc, llp, corp, corporation, company, limited, agency, real
estate, management, asset, and investment. The sample was then cleaned for internal consistency
in regards to the formatting of the address variables. While there were over 140,000 defendants
identified, they were spatially distributed across not only the state of Minnesota, but also
nationally. In order to narrow the observations to only those located in Minneapolis, the sample
was limited to where the city was listed as Minneapolis or the zip code was located within
Minneapolis. Because of data entry errors, a small number of Minneapolis addresses were likely
dropped because neither the city nor the zip code were correctly entered.
14
According to the sample, 44,280 evictions were filed in Minneapolis during this time
period. Addresses were geocoded using ArcMap on the basis of address, city, and zip code. The
geocode successfully matched 96% of addresses in the sample, approximately 42,500 unique
observations. Although some addresses were coded as being in Minneapolis, the address and zip
code combination placed them outside the city. Such observations were also dropped. This left a
total of 42,280 unique observations in Minneapolis from 2007 to 2020. The sample was then
limited to the final data range for the study of January 1, 2015 to December 31, 2019. In the end,
the sample contained 15,532 unique eviction filings. The data was finally aggregated to the
census tract level and combined with data from the American Community Survey 2015-2019
5-year estimates. Table 1 displays summary statistics for the sample.
Minneapolis Rental Licensing and Property Sales
Rental licensing data comes from the City of Minneapolis. All rental units, regardless of size,
must be annually licensed with the City of Minneapolis. Unfortunately, the data set is updated
weekly and is not archived. This means that the sample is a static snapshot of the active rental
licenses in Minneapolis at the time it was downloaded, February 2021. At this time, 99,491 units
were licensed with the city.
The property sales data also comes from the City of Minneapolis, comprising every
property sale from 2015 to 2019. The sample was limited to only residential sales or sales of
vacant land zoned for residential use. Both owner and rental housing are included in the sample.
Sales were classified into five different categories—single family homes, townhomes,
condominiums (including cooperative units), duplexes and triplexes, and multifamily properties,
which are properties with four or more units. The city of Minneapolis classifies any building
15
with four or more units as multi-family housing. From 2015 to 2019 there were 38,982 sales of
residential properties.
Table 1. Summary Statistics by Census Tract 2015-2019 (n=116)
16
Variables & Methodology
Key Dependent Variables
I run two separate Spatial Durbin models. The first model uses the logged eviction filing rate as
the dependent variable. The rate was calculated by taking the number of eviction filings from
2015 to 2019 as a proportion of the licensed rental units in the census tract. Eviction filings are
defined as any case that did not result in a legal ruling. Eviction judgments are excluded from
filings to avoid capturing the impact of filings specifically with the intent of removing a tenant.
The second model uses the logged eviction judgment rate as the dependent variable. Eviction
judgments are defined as any filing that resulted in an enforced eviction against the tenant.
As a note, two census tracts did not have any eviction judgments from 2015 to 2019. In
order to address the issues associated with logging values of zero, I removed these two census
tracts from the eviction judgment analysis, leaving 114 observations. Adding a small constant
term to the eviction judgment rate or using an inverse hyperbolic sine transformation were both
also considered. The exclusion method was used for simplicity and because both the coefficient
estimates and standard errors were similar between that and adding a constant term. The inverse
hyperbolic sine transformation was rejected as it significantly changed the estimates and
interpretation of the values, making comparisons between the eviction filing rate and eviction
judgment rate difficult. Full results of the three alternatives are presented in Appendix A. 1
Independent Variables
The key independent variable is the percentage of rental properties that sold in a census tract
from 2015 to 2019. For this study, a rental property is defined as a multifamily property (four or
1 Bellego and Pape, Dealing with Logs and Zeros in Regression Models.
17
more units), a duplex, or a triplex. I use multifamily and duplex/triplex properties to best capture
the buying and selling of rental properties. One limitation to this approach is that the variable
does not capture rental units in single-family homes. The exclusion of single-family rentals is
especially important in census areas located in areas like North Minneapolis, where a significant
amount of the rental properties are single-family homes. The percentage was calculated by taking
the cumulative number of multifamily, duplex, and triplex properties sold in a census tract as a
proportion of the total number of properties of those types in the census tract. The percentage of
rental properties sold models the volume of real estate activity in a given census tract.
Rental property sales are a reflection of real estate activity in a census tract, and are a
proxy for gentrification in this study. Rental property sales are, of course, not the only driver of
gentrification. However, results from recent papers on gentrification argue for the inclusion of
real estate activity, whether construction of new properties or sales of existing ones, in measures
of gentrification (Raymond, 2021; Sims 2021). Modeling real estate activity also makes sense
from a theoretical standpoint. Within Smith’s rent gap theory, gentrification arises out of a cycle
of divestment and reinvestment. When a new owner purchases a property in a neighborhood with
appreciating land values, they will maximize their investment by raising rents to match the new
increased value. However, it is likely that the existing residents of the property will not be able to
afford the higher rents and are thus displaced. The model presented here addresses whether
displacement is occurring through the eviction filing process.
As stated earlier, Raymond argues for separating eviction judgments from eviction
filings, as the literature shows they serve different purposes (Raymond, 2021). Without
disaggregation, eviction filings capture a variety of different social processes, but eviction
judgments best reflect filings with the intent to remove a tenant. If gentrification-induced real
18
estate activity resulted in an increase in evictions, I expect this to be captured in the eviction
judgment rate, but not necessarily in the eviction filing rate. Eviction filings, regardless of the
strength of the relationship between gentrification and eviction judgments, would not necessarily
increase in a gentrifying tract. Instead, eviction filings are more likely to capture broader
practices of racial exclusion, wealth extraction, and insecurity.
Control Variables
I also include variables to control for demographic and neighborhood characteristics.
Demographic variables included are the percentage of Black residents in the census tract and the
percentage of cost-burdened renters in the census tract. Neighborhood characteristics are
measured by the percentage of the housing supply that was constructed before 1960.
I expect race to be a primary driver of both eviction filings and eviction judgments.
Census tracts with a higher share of Black residents have substantially higher rates of eviction
filings and eviction judgments. Past research also shows that Black renters are disproportionately
impacted by eviction filings and eviction judgments. Cost-burden is defined as a household
paying more than 30% of their income towards housing costs, including utilities. This variable
captures two factors, both income and housing costs. As the percentage of cost-burdened renters
increases, I expect both eviction filings and eviction judgments to increase. Whether it is because
of high housing costs or low-incomes, cost burdened residents are less likely to be able to
consistently make housing payments. As a result, cost-burdened renters are more likely to
experience predatory eviction filings and or be removed from their unit.
Finally, I expect the age of the building stock to be positively associated with eviction
filings and judgments. Older housing stock is more likely to have deteriorated and have lower
rent prices and more likely to be located in a neighborhood that has been underinvested in.
19
Neighborhoods with a greater percentage of older housing are more likely to have poorer
residents, with a lower ability to make rent.
Spatial Econometric Model
Spatial econometric models allow relationships between geographic areas to be analyzed. Spatial
models are well suited to identify relationships that are based on 1) endogenous effects: the
dependent variable in one observation is correlated with the dependent variable of other
observations, 2) exogenous relationships: the dependent variable in one observation is correlated
with the independent variables of other observations, and 3) correlated relationships: the error
terms across multiple observations are related (Elhorst 2010; Manski 1993).
The Manski model is the umbrella equation for other spatial econometric models,
because it includes all three interactions. However, it has several important disadvantages.
Elhorst found the Manski model to create biased estimates for both the endogenous and
exogenous interaction effects, making distinguishing the unique impact of the two relationships
difficult (Elhorst 2010). Because of this, most of the literature argues against specifying the
Manski model. While both the Spatially Lagged X lag includes the exogenous interaction and the
Spatial Error models spatial autocorrelation, neither is able to estimate the endogenous
interaction. The Spatial Durbin Model is unique in that it allows both the endogenous and
exogenous relationships to be modeled, and produces unbiased coefficient estimates. In order to
specify an appropriate model, I follow LeSage by narrowing the potential options to only the
Spatial Durbin Model and the Spatial Durbin Error Model, and their nested formulas—which
include the Spatially Lagged X Model, the Spatially Lagged Y Model, and the Spatial Error
Model (LeSage 2014).
20
There are two main differences between the Spatial Durbin Model and Spatial Durbin
Error Model. First, the Spatial Durbin Model includes the lagged dependent variable, modeling
the endogenous interaction effect. Second, the Spatial Durbin Error Model models spatial
autocorrelation in the error term. The relevant distinction between these two models is that the
Spatial Durbin Error Model models local spillover effects and the Spatial Durbin Model models
both local and global spillover effects. In this case, a local spillover effect would mean that
investment and characteristics in an immediate neighbor census tract has an impact upon a
specified census tract. However, the impacts do not apply to all census tracts across the city, only
immediate neighbors. In the Spatial Durbin Model, the impact is global—meaning that what
happens in one census tract impacts all census tracts in the city.
Following Laniyonu’s work on Brooklyn, I utilize the Spatial Durbin Model. He suggests
that part of the gap in the literature supporting the link between evictions and gentrification is
partially due to not accounting for spatial dependence, or when there is a systematic geographic
relationship between variables of interest (Laniyonu 2019). In this study spatial dependence
would mean that an increase in rental property sales in one census tract is associated with an
increase in the eviction rate in neighboring tracts.
The Spatial Durbin Model also allows a supplementary question to be addressed. In
addition to modeling spillover effects from the independent variables, the Spatial Durbin Model
models spillover impacts from the dependent variable, as well. The lagged dependent variable
can potentially provide insight to landlords’ eviction strategies and actions. A spillover effect
could point to a level of coordination or communication between landlords in local regions
regarding their eviction practices, or at the very least, an awareness of general strategies.
21
The statistical tests used to specify the appropriate spatial model also indicate that the
Spatial Durbin Model best fits the data. The full discussion of the results of the specification
analysis can be found in Appendix C. Results from these tests show that 1) the spatial error term
is not necessary, and thus, the Spatial Durbin Error Model is not the appropriate model. 2) The
Spatial Durbin Model should not be restricted to a reduced form of the equation, either the
Spatially Lagged X Model, Spatially Lagged Y Model, or Spatial Error Model.
The Spatial Durbin Model can be expressed with the following equation (Elhorst 2010):
y = ρWy + αιn + Xβ + WXθ + ε
where y is the dependent variable, the logged the eviction filing rate or the logged eviction
judgment rate; W is the spatial weight matrix; Wy represents the spatially lagged dependent
variable (the endogenous relationship); ρ is the effect of Wy; ιn is a n x 1 vector associated with
the constant term; X is a n x k of k independent variables, related to the parameters β; WX
represents the spatially lagged independent variables (the exogenous relationship), associated
with a θ k x 1 vector of effects; and ε is the error term.
Results
The mean eviction filing rate in Minneapolis by census tract is 23.7%, meaning that there was
approximately one eviction filing for every four licensed rental units from 2015 to 2019. Figure
A shows a map of the eviction filing rate in Minneapolis from 2015 to 2019. Clear spatial
patterns are present in the map, with eviction filings being highly correlated with the race and
class composition of census tracts. Almost every tract with an eviction filing rate higher than the
city average is located in North Minneapolis or in the neighborhoods comprising the
“Southside”, such as Phillips, Corcoran, Central, and Bryant. High-density renter neighborhoods
22
that tend to be more affluent, such as those in Uptown, Downtown, and near the University of
Minnesota, all have eviction filing rates that are less than the city’s overall rate. Figure A also
presents this information in the form of the spread of standard deviations of the filing rate.
Notably, the only census tracts that have a standard deviation more than one standard deviation
away from the average filing rate are located in North Minneapolis.
Figure B presents the same maps but shows the rate of eviction judgments instead of the
filing rate. The mean eviction judgment rate in the city is approximately 9.3%, or approximately
one eviction judgment for every ten units of rental housing. In theory, if gentrification pressures
were significantly impacting the number of eviction judgments, the spatial distribution between
the eviction filing and eviction judgment rate would be different. Filings would likely be
associated with areas with higher poverty levels, and eviction judgments would be associated
both with areas with higher poverty levels and with areas that are experiencing or have already
recently undergone gentrification. However, the spatial distribution of eviction judgments is
quite similar to eviction filings, with only minor variation between the two rates. While not
conclusive, these findings provide initial evidence that gentrification is not a major factor driving
eviction-based displacement.
23
Figure A: Eviction Filing Rate and Standard Deviations in Minneapolis From 2015-2019
Figure A shows the eviction filing rates by census tract in Minneapolis from 2015 to 2019. The eviction filing rate is calculated by dividing the number of eviction
filings in a census tract from 2015 to 2019 by the number of licensed rental units in the census tract.
24
Figure B: Eviction judgment Rate and Standard Deviations in Minneapolis From 2015-2019
Figure B shows the eviction judgment rates by census tract in Minneapolis from 2015 to 2019. The eviction judgment rate is calculated by dividing the number of
eviction judgments in a census tract from 2015 to 2019 by the number of licensed rental units in the census tract
25
Table 2 presents the results from the Spatial Durbin Models. Models 1 and 2 present the
results with the combined “% Rental Properties Sold” variable—the percentage of multifamily,
duplex, and triplex properties sold from 2015 to 2019 compared to the total number of properties
of those types. In a Spatial Durbin Model, coefficients are presented as direct effects, indirect
effects, and total effects.
The direct effect measures the impact of change in a variable within a unit of observation,
in this case the census tract. The indirect effect is used to measure the “spillover” effect—the
impact on a census tract’s eviction rate from a change in an independent variable of a
neighboring census tract. For example, the indirect effect on the percentage of cost burdened
residents variable measures the impact on eviction judgments/filings when the percentage of cost
burdened residents increases in a neighboring census tract. The total effect is a sum of the direct
and indirect impact. Finally, the lagged spatial dependent variable is estimated by the coefficient
rho. This estimates the spatial spillover impact of eviction filings and eviction judgments—what
happens in a neighboring census tract when eviction filings or judgments increase in a census
tract?
The results of Models 1 and 2 show that increases in the percentage of rental properties
sold in a census tract have a significant impact on both eviction filings and eviction judgments.2
An increase in the percentage of rental properties sold in a census tract by one percentage point
is associated, on average, with a 0.93% increase in the eviction judgment rate and a 0.95%
increase in eviction filings. However, only the impact on eviction filings is statistically
significant at the 90% level. The estimated direct effect of rental property sales are similar for
both filings and judgments, but diverge slightly in their indirect impact. When the percentage of
2 Both Model 1 and Model 2 were also run with the raw count of rental properties sold in a census tract.
Neither the direct, indirect, nor total impact were statistically significant for either the eviction filing rate or
the eviction judgment rate.
26
rental properties sold in a census tract increases by one percentage point, the eviction filing rate
increases by approximately 2.90%, on average. However, the eviction judgment rate increases,
on average, by 3.56%. This indicates that not only does an increase in the percentage of rental
properties sold impact the number of evictions filed and executed within a census tract, but also
significantly impacts the tract’s neighbors. Additionally, an increase has a larger impact on
eviction judgments than eviction filings, by 0.66 percentage points. However, when considering
the standard errors and 95% confidence intervals, I am unable to conclude that the estimates are
statistically distinct from one another.
Models 3 and 4 are used as a sensitivity analysis on the results of Models 1 and 2 by
separating out the percentage of rental property sales into multifamily property sales and
duplex/triplex property sales. When separated, the percentage of multifamily property and the
percentage of duplex and triplex sales, neither variable is significantly related to eviction filings
or eviction judgments.
As expected, the demographic and neighborhood characteristics are significantly related
to both eviction filing and eviction judgment rates. Models 1 and 2 show that a one percentage
point increase in the percentage of Black residents in a census tract is associated with a 2.86%
increase in the tract’s eviction judgment rate and a 2.90% increase in the tract’s eviction filing
rate. However, the indirect effect of an increase in the percentage of Black residents is not
statistically significant. This could be because of the geographic clustering of eviction filings and
eviction judgments in census tracts with a higher percentage of Black residents.
The direct impact of the percentage of cost burdened renters in a census tract did not have
a significant impact on the eviction filing rate or the eviction judgment rate. However, both the
indirect and total impact was associated with an increase in both the eviction filing rate and the
27
eviction judgment rate. An increase of one percentage point in the percentage of cost-burdened
renters in a census tract is associated with a 3.44% increase in the eviction judgment rate and a
2.92% increase in the eviction filing rate of the tract’s neighbors. Overall, the impact of a one
percentage point increase in the percentage of cost burdened renters is related to a 3.86%
increase in eviction judgment rates and 3.44% increase in eviction filing rates.
The direct effect of the percentage of buildings constructed prior to 1960 is not
significant for either the filing rate or judgment rate, and the indirect impact is only significant
on the eviction filing rate. An increase of one percentage point in the percentage of buildings
constructed pre-1960 is associated with a 2.04% increase in the eviction filing rate in
neighboring census tracts. However, although neither the direct nor indirect effects are
significant for the eviction judgment rate, the total impact is significant—associated with an
overall increase in the eviction judgment rate of 1.47%.
Finally, the models estimate the impact of spatial spillovers in the filing and judgment
rates. I only found a significant spillover effect in the eviction filing rate, but not the eviction
judgment rate. On average, a 100% increase in the eviction filing rate is associated with an
increase of 25.3% in the filing rate of neighboring census tracts. This provides evidence that
there is significant spatial dependence in eviction filings and that the practices of landlords in
one census tract impact the behavior of landlords in neighboring census tracts.
28
Table 2:
29
Discussion & Conclusion
The model answers three questions related to processes of displacement and residential
alienation. Do eviction filings and judgment reflect patterns of racial exclusion and exploitation?
Is gentrification significantly driving eviction-based displacement in Minneapolis? Is there a
spatial spillover from eviction filings and judgments onto neighboring census tracts?
The results strongly support past work connecting evictions to racial exclusion, poverty,
and neighborhood precarity. Unsurprisingly, race and evictions are highly correlated in
Minneapolis from 2015 to 2019. Additionally, I found that the age of the building stock and the
percentage of cost-burdened renters impact the rates of filings and judgments in neighboring
tracts. The estimated impacts for demographic and neighborhood characteristics are similar
between judgment rates and filing rates.
I also find that the sale of rental properties is significantly related to an increase in the
rate of eviction judgments and eviction filings. The estimated magnitude of the total impact on
the judgment rate is approximately 16% greater than the magnitude of the filing rate. However,
the 95% confidence interval for the impact on the filing rate is 1.1% to 6.6% and 1.2% to 7.2%
for the judgment rate. While I did not run a significance test on the two coefficients, the high
level of overlap in the confidence intervals indicates that the estimates are not statistically
different.
There are two potential conclusions of these results. On one hand, the similarity between
the impact of rental property sales on filing rates and judgment rates could indicate that
gentrification did not significantly impact eviction filings in Minneapolis during this time period.
On the other hand, the results may reflect the weakness of rental property sales as a proxy for
gentrification. Recent work highlights the importance of at least including real estate activity in
30
gentrification models (Raymond, 2021; Sims 2021). However, the model does not account for
major factors often associated with gentrification, such as demographic change, employment
sectoral changes, and rising housing costs.
Additionally, it is further limited by two considerations. First, there is evidence that who
buys a rental property matters. Raymond (2021) found that investor-purchased rental properties
were associated with increases in eviction-based displacement. However, there was no
significant impact when the model was expanded to all rental property sales. It is possible that
when narrowed to investor-purchased properties, a similar pattern exists in Minneapolis. Second,
there is a potential problem of scale. The census tract level might be an inappropriate level of
analysis. Again, recent evidence from Sims suggests that gentrification-led eviction-based
displacement occurs at a much smaller level than the census tract. Aggregating data to the tract
level potentially disguises other patterns that are occurring within neighborhoods.
Despite the limitations model’s limitations, I believe the first option more accurately
reflects the nature of evictions in Minneapolis from 2015 to 2019. Sims concludes that there are
two distinct eviction-based displacement processes—one “based on the (re)production of social
precarity, exclusion at the urban scale, and possibly an ‘eviction economy’ as “a patchwork of
islands of extreme risk and eviction.” The other is from “emergent real estate-led
gentrification-based displacement” (Sims 2021). Different eviction processes are dependent on
the unique characteristics of a regional housing market (Sims and Iverson 2021).
These results do not preclude a connection between gentrification and displacement in
Minneapolis. Rather, it is possible that gentrification-based displacement happens through
different mechanisms. Minneapolis has relatively few protections for renters. There are few
eviction protections, no rent regulations, and tenants do not have a right to an attorney. On one
31
hand, fewer protections means evictions are easier to execute. On the other hand, the lack of
legal protections reduces the need for landlords to go through a legal eviction filing process at
all. Instead, informal “self-help” evictions are often simpler and cheaper than filing an eviction
with the court. A report on evictions in Minneapolis from 2019 found that “mutual termination of
lease by non-renewal” evictions were the most common eviction strategies (Lewis, 2019).
Mutual termination of a lease refers to the practice of non-renewal at the end of a lease. This
practice is especially common with month-to-month leases.
Finally, I find a spatial spillover effect from eviction filings, but not eviction judgments.
The results show that when the filing rate doubles in one census tract, it is associated with a 25%
increase in the eviction filing rate in neighboring tracts. These results press the question of why
there appears to be a spillover effect for filings but not for eviction judgments, especially since
they show similar spatial patterns. Further research should explore this question, and the
possibility of landlord coordination around industry “best practices.”
These results shed light on the various eviction-related processes in Minneapolis,
highlighting the uneven geographies of filings and judgments. Rather than framing eviction
filings as an individual phenomena, the residential alienation framework draws attention to the
broader economic structures that give rise to them. There are certainly crucial intervention points
that policymakers can address within the direct causes of evictions, such as eviction protections,
tenant right to counsel, and pre-filing arbitration. However, the alienation framework highlights
the fundamental power imbalance in the landlord and renter relationship and the incentive
structure that enables these processes in the first place. The results from the two different
processes studied here requires much further exploration, especially from a causal framework.
32
Works Cited
Brown-Saracino, Japonica. “Explicating Divided Approaches to Gentrification and Growing
Income Inequality.” Annual Review of Sociology 43 (May 19, 2017): 515–39.
Chronopoulos, Themis. “African Americans, Gentrification, and Neoliberal Urbanization: The
Case of Fort Greene, Brooklyn.” Journal of African American Studies 20, no. 3
(December 1, 2016): 294–322. https://doi.org/10.1007/s12111-016-9332-6.
Chum, Antony. “The Impact of Gentrification on Residential Evictions.” Urban Geography 36,
no. 7 (October 3, 2015): 1083–98. https://doi.org/10.1080/02723638.2015.1049480.
Davidson, Mark. “Displacement, Space and Dwelling: Placing Gentrification Debate.” Ethics,
Place & Environment 12, no. 2 (June 2009): 219–34.
https://doi.org/10.1080/13668790902863465.
Desmond, Matthew, and Carl Gershenson. “Housing and Employment Insecurity among the
Working Poor.” Social Problems 63, no. 1 (February 2016): 46–67.
https://doi.org/10.1093/socpro/spv025.
———. “Who Gets Evicted? Assessing Individual, Neighborhood, and Network Factors.” Social
Science Research 62 (February 2017): 362–77.
https://doi.org/10.1016/j.ssresearch.2016.08.017.
Easton, Sue, Loretta Lees, Phil Hubbard, and Nicholas Tate. “Measuring and Mapping
Displacement: The Problem of Quantification in the Battle against Gentrification.” Urban
Studies 57, no. 2 (February 1, 2020): 286–306.
https://doi.org/10.1177/0042098019851953.
Elhorst, J. Paul. “Applied Spatial Econometrics: Raising the Bar.” Spatial Economic Analysis 5,
no. 1 (March 2010): 9–28. https://doi.org/10.1080/17421770903541772.
“Estranged Labour, Marx, 1844.” Accessed March 9, 2021.
https://www.marxists.org/archive/marx/works/1844/manuscripts/labour.htm.
Forrest, R, and P Williams. “Commodification and Housing: Emerging Issues and
Contradictions.” Environment and Planning A: Economy and Space 16, no. 9 (September
1, 1984): 1163–80. https://doi.org/10.1068/a161163.
Freeman, Lance. “Displacement or Succession?: Residential Mobility in Gentrifying
Neighborhoods.” Urban Affairs Review 40, no. 4 (March 1, 2005): 463–91.
https://doi.org/10.1177/1078087404273341.
Garboden, Philip ME, and Eva Rosen. “Serial Filing: How Landlords Use the Threat of
Eviction.” City & Community 18, no. 2 (June 1, 2019): 638–61.
https://doi.org/10.1111/cico.12387.
Harrison, Austin, Dan Immergluck, Jeff Ernsthausen, and Stephanie Earl. “Housing Stability,
Evictions, and Subsidized Rental Properties: Evidence From Metro Atlanta, Georgia.”
Housing Policy Debate, September 7, 2020, 1–14.
https://doi.org/10.1080/10511482.2020.1798487.
Hartman, Chester, and David Robinson. “Evictions: The Hidden Housing Problem.” Housing
Policy Debate 14, no. 4 (January 2003): 461–501.
https://doi.org/10.1080/10511482.2003.9521483.
Immergluck, Dan, Jeff Ernsthausen, Stephanie Earl, and Allison Powell. “Evictions, Large
Owners, and Serial Filings: Findings from Atlanta,” n.d., 24.
Laniyonu, Ayobami. “Assessing the Impact of Gentrification on Eviction: A Spatial Modeling
Approach Symposium - Black, Poor, and Gone: Civil Rights Law’s Inner-City Crisis.”
33
Harvard Civil Rights-Civil Liberties Law Review 54, no. 2 (2019): 741–68.
LeSage, James P. “What Regional Scientists Need to Know About Spatial Econometrics.” SSRN
Scholarly Paper. Rochester, NY: Social Science Research Network, January 5, 2014.
https://doi.org/10.2139/ssrn.2420725.
Leung, Lillian, Peter Hepburn, and Matthew Desmond. “Serial Eviction Filing: Civil Courts,
Property Management, and the Threat of Displacement.” Social Forces, September 11,
2020, soaa089. https://doi.org/10.1093/sf/soaa089.
Lewis, Brittany. “The Illusion of Choice: Evictions and Profit in North Minneapolis.” Center for
Urban and Regional Affairs, n.d., 195.
Madden, David, and Peter Marcuse. In Defense of Housing. Brooklyn, NY: Verso, 2016.
Manski, Charles F. “Identification of Endogenous Social Effects: The Reflection Problem.” The
Review of Economic Studies 60, no. 3 (1993): 531–42. https://doi.org/10.2307/2298123.
Marcuse, Peter. “Gentrification, Abandonment, and Displacement: Connections, Causes, and
Policy Responses in New York City.” Washington University Journal of Urban and
Contemporary Law 28 (1985): 195–240.
Minneapolis Innovation Team. “Evictions in Minneapolis,” n.d., 26.
Moylan, Mary. “Report: More Middle-Income Renters Burdened by Housing Costs.” MPR
News. Accessed April 30, 2021.
https://www.mprnews.org/story/2020/01/31/report-more-middleincome-renters-burdened
-by-housing-costs.
Raymond, Elora Lee. “Gentrifying Atlanta: Investor Purchases of Rental Housing, Evictions, and
the Displacement of Black,” n.d., 18.
Seymour, Eric, and Joshua Akers. “Building the Eviction Economy: Speculation, Precarity, and
Eviction in Detroit.” Urban Affairs Review 57, no. 1 (January 1, 2021): 35–69.
https://doi.org/10.1177/1078087419853388.
Sims, J. Revel. “Measuring the Effect of Gentrification on Displacement: Multifamily Housing
and Eviction in Wisconsin’s Madison Urban Region.” Housing Policy Debate, April 15,
2021, 1–26. https://doi.org/10.1080/10511482.2021.1871931.
———. “More than Gentrification: Geographies of Capitalist Displacement in Los Angeles
1994–1999.” Urban Geography 37, no. 1 (January 2, 2016): 26–56.
https://doi.org/10.1080/02723638.2015.1046698.
Sims, J. Revel, and Alicia Adelle Iverson. “Multiple Eviction: An Investigation of Chain
Displacement in Dane County, Wisconsin.” Urban Affairs Review 57, no. 2 (March 1,
2021): 527–51. https://doi.org/10.1177/1078087419886114.
Smith, Neil. “Toward a Theory of Gentrification: A Back to the City Movement by Capital, Not
People.” Journal of the American Planning Association 45, no. 4 (October 1979):
538–48. https://doi.org/10.1080/01944367908977002.
Soederberg, Susanne. “Evictions: A Global Capitalist Phenomenon.” Development and Change
49, no. 2 (2018): 286–301. https://doi.org/10.1111/dech.12383.
Stein, Samuel. Capital City: Gentrification and the Real Estate State. Verso, 2019.
34
Appendix
Appendix A: Different Models to Address Log(0) Values
35
Appendix B: Percentage of Rental Property Sales in Minneapolis from 2015 to 2019
Appendix B shows the percentage of rental properties (classified as multi-family, duplex, or triplex properties) that were sold in Minneapolis from 2015 to 2019.
The percentage was calculated by dividing the number of sales that were coded as rental properties by the number of coded rental properties in the census tract,
multiplied by 100.
36
Appendix C: Model Specification
I conduct a series of statistical tests to determine 1) if the Spatial Durbin Model or Spatial
Durbin Error Model better fits the data, and 2) if the preferred model of the two is a better fit
than the reduced forms nested within them. Following Elhorst’s (2010) approach, I begin by
estimating the model with OLS and testing for spatial autocorrelation, using the Moran I test. For
the test, I assert a null hypothesis that the data is randomly spatially dispersed. If the null
hypothesis can be rejected, there is likely spatial autocorrelation and a spatial model should be
used.
The null hypothesis of the Moran I Test is that there is no spatial autocorrelation in the
residuals. If the null hypothesis is rejected, a spatial model better fits the data. The Moran I test
produced a statistic with a p-value of 0.003, the null hypothesis can be rejected. Next I run four
LaGrange multiplier tests, testing if an OLS model performs better than a spatial lag model and a
spatial error model, one test for each without a robust standard error and one with a robust
standard error. Table __reports the results of the Lagrange Multiplier tests.
Without robust standard errors, both the p-value in both tests is statistically significant,
indicating either a model with lagged independent variables or a lagged error term would fit the
data. However, when adjusting for robust standard errors, only the lagged independent model is
statistically significant. The output of the Lagrange Multiplier tests indicate that a spatial error
term is not necessary in the model.
In order to determine which is best suited for the data, a Likelihood Ratio test can be used
to test whether the Durbin model can be simplified to either a Spatial Error Model or Spatially
Lagged X Model. If both Likelihood Ratio tests can be rejected, the model should not be
37
simplified and the Durbin model is the best fit. However, if one of the tests can not be rejected
then that model likely better fits the data than the Durbin model.
The Spatially Lagged X, Spatially Lagged Y, and Spatial Error model are also all nested within
the Spatial Durbin Model model. To test if the model could be reduced to one of these functional
forms, I ran a series of Likelihood Ratio Tests. The Likelihood Ratio (LR) tests the null
hypothesis that the restricted models better fits the data than the Spatial Durbin Model model. If
the hypothesis can be rejected, the Spatial Durbin Model model should not be reduced to the
nested model being tested against.
Table __ reports the results of the LR test. The p-values for the LR test between the
Spatial Durbin Model and and all of the reduced forms are both statistically significant, meaning
that the model should not be restricted.
38