Article Urban Studies 1–20 Ó Urban Studies Journal Limited 2020 Article reuse guidelines: sagepub.com/journals-permissions DOI: 10.1177/0042098020903011 journals.sagepub.com/home/usj Mapping gentrification and displacement pressure: An exploration of four distinct methodologies Benjamin Preis Department of Urban Studies and Planning, Massachusetts Institute of Technology, USA Aarthi Janakiraman Department of Urban Studies and Planning, Massachusetts Institute of Technology, USA Alex Bob Department of Urban Studies and Planning, Massachusetts Institute of Technology, USA Justin Steil Department of Urban Studies and Planning, Massachusetts Institute of Technology, USA Abstract As housing costs continue to increase across many cities in North America and Europe, local gov- ernments face pressure to understand how housing’s rising cost is changing neighbourhoods and to ensure that everyone can access a home they can afford. To confront displacement concerns, cities are adapting models developed within academia to identify neighbourhoods that may be susceptible to gentrification and displacement. We compare four gentrification and displacement risk models developed by and for the US cities of Seattle, Washington; Los Angeles, California; Portland, Oregon; and Philadelphia, Pennsylvania, and apply all four methodologies to one city, Boston. We identify the geographic areas of agreement and disagreement among the methods. The comparison reveals striking differences between the models, both in inputs and outputs. Of the 18 variables considered among the four models, only two variables appear in all four models. In the resulting maps, the four methods identified between 25 and 119 of the 180 Boston census tracts as at risk of gentrification and displacement, or as currently gentrifying. There are only seven tracts that all four models agreed were either gentrifying or at risk of gentrification and dis- placement. The findings indicate a need for cities to consider critically the assumptions of the models that are included in urban policy documents, as indicators and thresholds have major Corresponding author: Benjamin Preis, Department of Urban Studies and Planning, Massachusetts Institute of Technology, 77 Massachusetts Avenue, Room 9-569, Cambridge, MA 02139, USA. Email: [email protected]
20
Embed
Mapping gentrification and displacement pressure: An ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Mapping gentrification anddisplacement pressure: Anexploration of four distinctmethodologies
Benjamin PreisDepartment of Urban Studies and Planning, Massachusetts Institute of Technology, USA
Aarthi JanakiramanDepartment of Urban Studies and Planning, Massachusetts Institute of Technology, USA
Alex BobDepartment of Urban Studies and Planning, Massachusetts Institute of Technology, USA
Justin SteilDepartment of Urban Studies and Planning, Massachusetts Institute of Technology, USA
AbstractAs housing costs continue to increase across many cities in North America and Europe, local gov-ernments face pressure to understand how housing’s rising cost is changing neighbourhoods andto ensure that everyone can access a home they can afford. To confront displacement concerns,cities are adapting models developed within academia to identify neighbourhoods that may besusceptible to gentrification and displacement. We compare four gentrification and displacementrisk models developed by and for the US cities of Seattle, Washington; Los Angeles, California;Portland, Oregon; and Philadelphia, Pennsylvania, and apply all four methodologies to one city,Boston. We identify the geographic areas of agreement and disagreement among the methods.The comparison reveals striking differences between the models, both in inputs and outputs. Ofthe 18 variables considered among the four models, only two variables appear in all four models.In the resulting maps, the four methods identified between 25 and 119 of the 180 Boston censustracts as at risk of gentrification and displacement, or as currently gentrifying. There are onlyseven tracts that all four models agreed were either gentrifying or at risk of gentrification and dis-placement. The findings indicate a need for cities to consider critically the assumptions of themodels that are included in urban policy documents, as indicators and thresholds have major
Corresponding author:
Benjamin Preis, Department of Urban Studies and Planning, Massachusetts Institute of Technology, 77 Massachusetts
impacts on how neighbourhoods in the liminal space of gentrification and displacement are char-acterised. This novel comparison of United States local government analyses of gentrification pro-vides insight as modelling moves from theory to practice.
Rapid neighbourhood socioeconomicchanges are an issue of pressing concern formany urban residents and local governmentsaround the world. As a result, cities areusing spatial analyses to understand wheregentrification-induced displacement is occur-ring or may occur. However, definitions andindicators of gentrification, as well as meth-ods for predicting it, vary widely, and thatvariation has significant implications forurban policy. While scholars have debateddiffering definitions of gentrification used inthe academic literature (e.g. Atkinson, 2003;Barton, 2016; Clark, 2005; Davidson andLees, 2005; Freeman, 2009), there has been
less examination of the ways in which citygovernments themselves are measuring gen-trification. We fill that gap by identifyingfour leading governmental efforts in theUnited States to measure and map gentrifi-cation risk, and then applying them all toone city in order to understand how themeasures differ and what the significance ofthose differences is.
Over the past 50 years, numerous studieshave sought to understand why gentrifica-tion happens, where it is occurring and itseffects (Edlund et al., 2015; Ellen andO’Regan, 2011; Freeman, 2005; Helbrecht,2018; Lopez-Morales, 2011; Smith, 1979;Sykora, 1993; Vigdor et al., 2002; Warde,
18180
25 119
2 Urban Studies 00(0)
1991). Methods used by cities to map gentrifi-cation and displacement risk have operatio-nalised this research in different ways,leading to the emergence of different under-standings of the leading causes and keyoutcomes. Regarding the causes of gentrifica-tion, many have identified the divergencebetween capitalised ground rents (rents cap-tured by the current use) and potentialground rents (maximum rents that could beappropriated with a change of use or user) asa fundamental aspect of gentrification (Clarkand Gullberg, 1997; Lees et al., 2008; Smith,1979; Smith and DeFilippis, 1999). These rentgaps are shaped by legal structures, publicpolicies and the social and political dimen-sions of economic power, all of which havebeen affected by shifting patterns of transna-tional capital investment and urban govern-ance (Lees, 2003; Lees et al., 2008). In theUnited States, these rent gaps and associatedrent seeking are also expressed racially andsocially, as a result of historic and continuingdiscrimination (Clark, 1995: 1496). Salient inrecent decades has been the role of the statein producing gentrification in conscious orunwitting partnership with the private sector,through infrastructure investment, public orsocial housing redevelopment, economicdevelopment policy and marketing (Chappleet al., 2017; Davidson and Lees, 2005; Goetz,2003; Smith, 1996; Vale, 2013, 2019; Wylyand Hammel, 1999). At the same time, eco-nomic restructuring has remade cities fromcentres of manufacturing to centres of busi-ness services and of knowledge and culturalproduction. High-income households areincreasingly opting to live in dense, walkableurban centres (Zukin, 1982) previously hometo low-income residents, a consumptivechoice (Warde, 1991) that may be driven bythe desire for cultural, environmental, trans-portation and recreational amenities(Anguelovski et al., 2018; Ley, 1986; Pratt,2018; Zuk et al., 2018; Zukin, 1987) and maybe related to decreases in leisure time for
high-earning households (Edlund et al.,2015). In recent years, debates about thecauses and direct effects of gentrification haveoften obscured how the phenomenon hasbroadly transformed once accessible urbanneighbourhoods into havens for speculativeprofit seeking, where severe housing cost bur-dens or displacement are the only options formany long-time residents (Marcuse, 1985;Newman and Wyly, 2006; Slater, 2006,2009).
Many have sought to identify where gen-trification is occurring (Atkinson, 2000;Ellen and O’Regan, 2011; Freeman, 2005;Grodach et al., 2018; Holm and Schulz,2018), including those who seek to profitfrom it (Chapple and Zuk, 2016: 125).Adverse impacts, such as displacement oflong-time residents, local businesses and cul-tural amenities; the disappearance of afford-able housing and socioeconomic diversity; aswell as increased real estate speculation andhomelessness (Atkinson and Bridge, 2005)have prompted mapping analyses that aimto identify where to target harm mitigationefforts. These analyses, however, mayobscure disagreements about what gentrifi-cation is and what factors characterise it.
Mapping efforts define gentrification dif-ferently, make varying decisions about howto operationalise components of gentrifica-tion and consequently draw different conclu-sions from results. These operationaldifferences are especially concerning in mapsthat are produced by governments, as theymay then inform public policy throughinclusion in comprehensive plans, such as inSeattle (Seattle Office of Planning &Community Development, 2016) andPortland (Bureau of Planning andSustainability, 2018: GP5-8), or throughconsolidated housing plans, such as inPhiladelphia (Division of Housing andCommunity Development, 2017: 97–98).
In this article, we explore these fault linesby comparing the outcomes of four distinct
Preis et al. 3
efforts to map gentrification from Philadelphia,Pennsylvania; Seattle, Washington; LosAngeles, California; and Portland, Oregon.In Seattle and Los Angeles, city agenciesthemselves mapped gentrification and displa-cement risk for city residents. In Portland,the city commissioned a study, and inPhiladelphia, analysts from the FederalReserve Bank of Philadelphia mapped gen-trification and mobility. While the statedmotivations of these analyses were somewhatdifferent, as discussed below, all claim tomap gentrification in order to identifygentrification-induced displacement.
By applying these four methods to thesame city, we can identify how differentapproaches to operationalising the conceptof gentrification may influence how citiesmobilise resources to address gentrification’snegative effects. This article aims to identifythe relationships between theories of gentrifi-cation, measures of neighbourhood change,mapping methodologies and the neighbour-hoods that are ultimately identified as facingdisplacement pressure. To do so, we firstidentify the tracts pinpointed by eachmethod. We then calculate descriptive statis-tics for the tracts identified by each modeland conduct bivariate analyses to comparethem. Significant disagreement between thefour models points to the importance ofchoosing a model for mapping gentrificationwith awareness of the methodology’sembedded assumptions about what constitu-tes gentrification and how neighbourhoodchange should be measured. Though the ori-gins of this article are based in the UnitedStates, cities globally are confronting gentri-fication and displacement, and this researchilluminates for advocates and policymakersthat there is no ‘one-size-fits-all’ method formapping gentrification and displacementrisk; practitioners must ensure that the meth-odology they use fits the temporal, spatialand socioeconomic context of the city athand.
Mapping gentrification anddisplacement risk
While early studies of gentrification wereoften qualitative analyses of specific neigh-bourhoods (Hammel and Wyly, 1996),recent quantitative measures of gentrifica-tion have relied primarily on census data tomeasure changes in neighbourhood compo-sition by income, race, education, housingvalue and other factors (Clark, 2005;Davidson and Lees, 2005; Ding et al., 2016;Ellen and O’Regan, 2011; Freeman, 2005).Academic debate continues over the appro-priate measures of gentrification (Barton,2016; Bousquet, 2017; Ding et al., 2016;Freeman, 2009), with studies demonstratingthe sensitivity of mapping measures to thevariables included (Galster and Peacock,1986; Mujahid et al., 2019). Previous gentri-fication mapping efforts have included vari-ous population and housing measures,including income, education, race, housingcosts and housing tenure (Bostic andMartin, 2003; Ding et al., 2016; Ellen andO’Regan, 2011; Freeman, 2005; McKinnishet al., 2008).
Beyond these demographic and economicmeasures, other studies also add proxies forpotential causes of gentrification, includingprivate real estate investment, state-led capi-tal investment and the role of creative indus-tries (Davidson, 2007; Grodach et al., 2018;Hamnett, 1991; Newman and Wyly, 2006;Pollack et al., 2010; Smith, 1996; Zuk et al.,2018). Still others add measures of proximityto infrastructure and public amenities(Chapple, 2009), or use novel data analysistechniques, such as machine learning (Readeset al., 2018). Holm and Schulz (2018) devel-oped a methodology for identifying gentrifi-cation that is meant to be transferrable toany city. Easton et al. (2019) reviewed chal-lenges with quantitative assessment of gentri-fication, noting that novel data sources mayease some extant limitations.
4 Urban Studies 00(0)
Decreases in affordable housing, concernsover economic and racial segregation, anxi-ety about the role of public investment inaccelerating gentrification and public outcryover neighbourhood change all make gentri-fication and displacement important to cityadministrators. Identifying neighbourhoodsvulnerable to the phenomenon may helpguide response efforts and future publicinvestments, as a growing body of researchhas found government investment in publicinfrastructure may trigger or exacerbate gen-trification (Chapple, 2009; Chapple et al.,2017; Pollack et al., 2010).
As municipal concern over displacementincreases, efforts to predict gentrificationand displacement multiply. Chapple andZuk (2016) survey the early warning systemsthat non-profit organisations, universitiesand cities are developing, and examine theformat and goals of early warning toolkitsand assess the toolkits’ policy influence.They argue that if the city is the creator orhost of the early warning system, the systemis more likely to have policy influence. Theincreasing number of mapping efforts repre-sent a concerted effort to understand wheregentrification has happened, is happeningand may happen, in order to change internalcity–government dialogue, assist efforts toorganise against gentrification and displace-ment or promote policy changes. These earlywarning systems and models of gentrifica-tion and displacement, however, rely on aplethora of different variables and measures.While locally tailored data and measuresadd value, this lack of consistency also maylead to ‘public confusion about the conceptof gentrification’ (Holm and Schulz, 2018:255).
Data and methods
We began by reviewing the seven leadingmapping efforts in six major cities in the
United States identified by Bousquet (2017),as well as identifying additional efforts con-ducted in the 30 largest cities by populationin the United States.1
Following Chapple and Zuk (2016), wedifferentiate between mapping efforts con-ducted by academic research centres, non-profit organisations, and local governmentsor other public agencies. We limit our studyto mapping efforts conducted by, or onbehalf of, government institutions – therebyfocusing on efforts that likely have the mostdirect policy influence. Governmental map-ping would be expected to be particularlyinfluential, yet these new efforts by munici-palities are understudied.
We select four distinct methods developedby or for four different cities to measuregentrification-related displacement pressure:the Los Angeles Innovation Team Index ofDisplacement Pressure; the PhiladelphiaFederal Reserve Study of Gentrification andResidential Mobility; Seattle’s DisplacementRisk Index from the Seattle 2035Comprehensive Plan; and Portland’sGentrification and Displacement Study.
As neither Seattle nor Los Angeles hadinitially planned their methods for reproduc-tion, we worked with the city staff who hadcreated them to recreate their methodolo-gies, double-checking our methods againsttheirs.2 We were also in contact with boththe Philadelphia and Portland teams, andfollowed their published methodologiesclosely. We then applied all four methods tothe same city, Boston. Variables and mini-mum thresholds used for each method arelisted in Table 1.
Seattle
The Displacement Risk Index was developedby Seattle’s Office of Planning &Community Development for the Seattle2035 Comprehensive Plan Equity Analysis
Preis et al. 5
Tab
le1.
Var
iable
san
dth
resh
old
sin
cluded
inea
chofth
efo
ur
map
pin
gm
ethods.
Indic
ators
Los
Ange
les
Port
land
Seat
tle
Phila
del
phia
Ind
ivid
ual
Perc
ent
(non-)
white
popula
tion
LAIN
C%
chan
gein
white
non-H
ispan
icre
siden
ts
.47%
ofpop.*
.20%
ofpop.
Colle
geed
uca
tion
atta
inm
ent
LAIN
C%
chan
gein
resi
den
tsø
25
with
bac
hel
or’s
deg
ree
.54.8
%ofpop.
no
colle
ge*
.40%
ofpop.
ove
r25
without
bac
hel
or’s
deg
ree
%ch
ange
inco
llege
educa
ted
pop.
.27.2
%*
Ho
use
ho
ldSh
are
ofnon-E
ngl
ish
spea
kers
.15%
ofpop.
Ren
tburd
ened
popula
tion
LAID
Pø
50%
house
hold
inco
me
inre
nt
.10%
ofpop.
under
80%
AM
Iw
ith
cost
burd
enor
seve
reco
stburd
enH
ouse
hold
inco
me
LAIN
C%
chan
gein
med
ian
house
hold
inco
me
LAID
Pø
60%
ofhouse
hold
sea
rnin
gunder
the
med
ian
inco
me
.50%
ofpop.
bel
ow
80%
AM
I*.
25%
ofpop.
under
200%
ofpove
rty
leve
l\
US$
56,3
74.9
9*
House
hold
size
LAIN
C%
chan
gein
house
hold
size
Shar
eofre
nte
rsLA
IDP
%ofre
nter
occ
upie
dunits
.65.7
%ofpop.*
.40%
ofpop.
Ren
talco
stLA
INC
%ch
ange
inm
edia
ngr
oss
rent
\90%
city
-wid
eav
erag
e%
chan
gein
med
ian
gross
rent
.15.6
%*
Housi
ng
pri
ceLA
IDP
Chan
gein
housi
ng
pri
cepro
ject
ions
for
trac
tsw
ith
housi
ng
Cat
egori
sation
of
trac
tsas
App
reci
ated
,Acc
eler
atin
gor
Adj
acen
tbas
edon
ratio
oftr
act
%ch
ange
inm
edia
nhom
eva
lue
.31.7
%*
(con
tinue
d)
6 Urban Studies 00(0)
Tab
le1.C
ontinued
Indic
ators
Los
Ange
les
Port
land
Seat
tle
Phila
del
phia
pri
ces\
80%
med
ian
valu
ean
din
crea
sing
ata
hig
her
rate
than
the
city
-wid
era
te
med
ian
pri
ce/m
edia
nap
pre
ciat
ion
toci
ty-
wid
em
edia
npri
ce/
med
ian
appre
ciat
ion
Neig
hb
ou
rho
od
Rat
ioofpoor/
wea
lthy
house
hold
sLA
INC
%ch
ange
inra
tio
of
low
-inco
me
(<U
S$25,0
00)
tohig
h-
inco
me
(øU
S$75,0
00)
taxpay
ers
Pre
sence
and
expir
atio
nofaf
ford
able
housi
ng
LAID
PN
um
ber
ofunits
wei
ghte
dby
year
of
expir
yPro
xim
ity
toaf
fluen
tnei
ghbourh
oods
LAID
P\
1m
ileto
hig
hly
chan
ged
ZIP
codes
scal
edby
dis
tance
Adja
cent
totr
act
with
housi
ng
pri
ceca
tego
rise
das
eith
erApp
reci
ated
or
Acc
eler
atin
g
Trac
tw
ith
Med
ian
House
hold
Inco
me\
80%
AM
Iad
jace
nt
totr
act
with
Med
ian
House
hold
Inco
me
.120%
AM
IPro
xim
ity
totr
ansi
t–
trai
nLA
IDP
\0.5
0m
iles
tost
atio
nsc
aled
by
dis
tance
\0.5
0m
iles
Pro
xim
ity
totr
ansi
t–
buse
s\
0.2
5m
iles
tost
atio
nPro
xim
ity
tojo
bs
\20
min
ute
sto
job
centr
eA
ttra
ctiv
ebusi
nes
ses
\0.5
mile
sC
ivic
infr
astr
uct
ure
\0.5
mile
sD
evel
opab
lepro
per
ties
Bin
ary
atpar
celle
vel
Not
es:LA
INC
and
LAID
Pre
fer
todiff
eren
tst
eps
inth
eLo
sA
nge
les
met
hod;a
nas
teri
sk(*
)in
dic
ates
that
the
thre
shold
valu
eis
city
-wid
em
edia
n;A
MIin
dic
ates
Are
a
Med
ian
Inco
me.
The
thre
shold
snote
dre
pre
sent
the
min
imum
thre
shold
inea
chgi
ven
met
hod.N
ote
that
this
table
still
under
stat
esso
me
subst
antial
diff
eren
ces
among
the
model
s,es
pec
ially
rega
rdin
gso
urc
esofdat
a.
Preis et al. 7
(Seattle Office of Planning & CommunityDevelopment, 2016: 13–18, 36–51). It is dis-tinct from, but follows, the GentrificationSusceptibility Index developed by Welch(2017: 87–89).
Using raster analysis, each variable (seeTable 1) is given a score, typically rangingbetween 0 and 4. Data sources and defini-tions for the variables are outlined in theSeattle 2035 Comprehensive Plan, and diver-gence for our adaptation is noted in the fol-lowing paragraph. Each of the 14 variablesare weighted approximately equally. At everypoint in the map, the given scores are addedtogether, giving a composite DisplacementRisk score. Thus, every point in Boston couldreceive a score between 0 (if it received a 0for every variable) and 59 (if it received themaximum score for every variable).
Four modifications were necessary to repli-cate the method for Boston. First, while theSeattle methodology used a city-definedmodel of development capacity, no analogousmetric exists for Boston. In consultation withthe Boston Department of NeighborhoodDevelopment, development capacity wasapproximated by parcels’ land-use type.Second, rent data in Boston were taken fromthe website PadMapper (Kaufman, n.d.),which differed from the proprietary data usedin Seattle. Third, King County defines joband manufacturing centres, our modelassesses distance to census tracts with highconcentrations of industrial and office jobs, asdefined by the EPA Smart Location Database(Environmental Protection Agency, 2013).Finally, euclidean, rather than network dis-tance, was used in the case of two variables.In order to compare the final raster analysiswith the methods based on census tractboundaries, zonal statistics were taken inArcGIS to calculate mean scores for eachtract. The categories were then created usingJenks Natural Breaks, a clustering optimisa-tion method, though there were minimal dif-ferences as compared with using quantiles.
Los Angeles
The Los Angeles Index of DisplacementPressure was created by the Los AngelesInnovation Team to reduce displacement,promote revitalisation and inform the priori-tisation of pilot areas for their projects. Itconsists of two steps to arrive at a displace-ment pressure measure: first, the creation ofa Los Angeles Index of NeighborhoodChange (Pudlin, 2016), and second, itsincorporation into the Los Angeles Index ofDisplacement Pressure (Pudlin, 2018).
The Los Angeles Index of NeighborhoodChange (LAINC) incorporates six metrics atthe ZIP code level, as indicated by Table 1.These indicators are normalised, weightedand added to compare the relative level ofneighbourhood change and create a corre-sponding change map, which is then used inthe Los Angeles Index of DisplacementPressure (LAIDP). The Los Angeles Indexof Displacement Pressure includes seven dif-ferent measures, indicated in Table 1. Tractsin which fewer than 40% of householdsearned below the city median income wereexcluded. These values were normalised,weighted and added to measure the displace-ment pressure for each census tract. LosAngeles created categories using quantilesfor the tracts that were neither excluded norhad a negative normalised z-score, assessingrelative risk.
In our replication of the model, we wereable to incorporate much of the same data.We worked with the Boston Department ofNeighborhood Development to acquire adataset for subsidised housing in Boston,but the dataset was incomplete, requiringthe judgement of the authors when incorpor-ating this indicator.
The weighting of the indicators to calcu-late the Displacement Pressure Index wasdeveloped from past analyses and ground-truthing in Los Angeles. Since this was notpossible for this study, we replicated theweights determined for Los Angeles for each
8 Urban Studies 00(0)
variable when applying the method toBoston.
Portland
Portland’s gentrification and displacementrisk assessment was commissioned by theCity of Portland and developed by ProfessorLisa Bates (2013) as a basis for both under-standing gentrification’s impact on Portlandand developing policies to address it. Theassessment identifies risk of gentrificationand displacement by census tract.
Risk typologies are determined by com-bining indicators across three dimensions:vulnerability to housing displacement,demographic changes and housing marketappreciation. Tracts are determined to bevulnerable to housing displacement if threeof four indicators – accounting for race,higher education, rent and income, as indi-cated in Table 1 – are above the city-wideaverage. Tracts are determined to have expe-rienced demographic change indicative ofgentrification if at least three of the vulner-ability indicators have decreased more thanthe city-wide average, or if just the race andhigher education variables decreased morethan city-wide. Finally, housing prices areassessed and tracts are classified accordingto their housing value, level of appreciationor adjacency to high value tracts. Tracts thathad not yet experienced a demographic shiftindicative of gentrification but had popula-tions vulnerable to displacement were classi-fied as either Susceptible or Early: Type 1based on whether their housing market con-ditions were adjacent or accelerating, respec-tively. Tracts with demographic changes andvulnerable populations were classified asEarly: Type 2, Dynamic or Late based onwhether their market conditions were adja-cent, accelerating or appreciated, respec-tively. Tracts not qualifying as vulnerablebut that had increasing portions of white,college educated residents and an
appreciated housing market were classifiedas Continued Loss.
Because all data used in the methodologyare nationally available at the tract level,there were few challenges replicating it forBoston and no modifications were necessary.
Philadelphia
Ding et al. (2016) developed their categoricalneighbourhood gentrification measure foruse in a study of gentrification’s influence onresidential mobility rates. While academi-cally focused, the initial paper was followedby a ‘Practitioner’s Summary’, designed toassist city officials in understanding andaddressing the issues of gentrification anddisplacement.
Following their method, our analysis uti-lised decennial census data from 1980, 1990and 2000, harmonised to 2010 census tractsby the Geolytics Neighborhood ChangeDatabase. Tracts are initially consideredGentrifiable if their household income at thestart of the period of analysis is below thecity-wide median; all others are consideredNot gentrifiable. Gentrifiable tracts wereconsidered to be gentrifying over the periodof analysis if they experienced an above city-wide median rate of increase in their shareof college educated residents and either med-ian gross rent or median home value. Tractsthat did not meet these criteria were cate-gorised as Nongentrifying.
While the main period of analysis was2000 to the present, the methodology alsoassessed whether gentrification occurredfrom 1980 to 2000. If tracts were gentrifyingprior to 2000 and continued to gentrify from2000 to the present, they were categorised asContinued gentrification. Tracts that weregentrifying before 2000 but did not qualifyas gentrifying after were categorised asStalled gentrification. Tracts that only beganto gentrify within the 2000-to-present timeperiod were classified as Weak gentrification,
Preis et al. 9
Moderate gentrification or Intense gentrifica-tion based on their quartile of median grossrent or median home value.
Similar to the Portland methodology, alldata used for the Philadelphia method wereavailable nationally at the tract level.Additionally, procedures for tract classifica-tion assignment allowed for straightforwardreproduction and application to Bostonwithout modification.
Results
When applied to Boston, the four methodol-ogies produce very different maps ofgentrification-related displacement risk, asshown in Figure 1. In order to analyse agree-ment and disagreement between mapping
methodologies, we first had to overcomeinconsistencies between the ways the mapsrepresented gentrification risk. Portland andPhiladelphia both used categorical typolo-gies, while Los Angeles and Seattle usedcontinuous risk scores. To facilitate compar-ison, we converted the continuous scoresinto categorical variables, and furtherreduced them to a binary at risk/not at riskvariable when appropriate.3
Pairwise statistical analysis
Table 2 presents the pairwise comparisonsbetween the four models, showing greatestagreement between Portland and Philadelphia,while Seattle and Philadelphia diverge themost. The Portland and Philadelphia analyses
Figure 1. Maps showing the four methodologies applied to Boston, MA.
10 Urban Studies 00(0)
take restrictive approaches to the census tractsthat they consider to be eligible for gentrifica-tion. On the other hand, the Seattle and, to alesser extent, the Los Angeles methodologiesare more permissive, allowing for more of thecity to be considered vulnerable. The way themethodologies determine tract eligibility forgentrification (see Table 1) and the distinctmethods for determining the severity of the
risk together explain the large difference innumber of tracts identified.
In order to compare agreement betweenmethodologies, we followed Barton (2016)and calculated a chi-squared and Cramer’sphi for each pairwise comparison (Table 2).Cramer’s phi can be read similarly to aPearson’s correlation coefficient for associa-tion between bivariate categorical
Table 2. Pairwise comparisons of different mapping methods.
Philadelphia
At risk Not at risk Chi-squared 12.266***
Portland At risk 12 24 Phi 0.281Not at risk 13 131
Seattle
At risk Not at risk Chi-squared 21.776***
Portland At risk 36 0 Phi 0.362Not at risk 82 62
Los Angeles
At risk Not at risk Chi-squared 2.698
Portland At risk 16 20 Phi 0.137Not at risk 41 103
Los Angeles
At risk Not at risk Chi-squared 9.304**
Philadelphia At risk 15 10 Phi 0.245Not at risk 42 113
Seattle
At risk Not at risk Chi-squared 10.403**
Philadelphia At risk 24 1 Phi 0.257Not at risk 94 61
Los Angeles
At risk Not at risk Chi-squared 29.595***
Seattle At risk 54 64 Phi 0.418Not at risk 3 59
Notes: N = 180; *p \ 0.05; **p \ 0.01; ***p \0.001. NA included in ‘not at risk’.
Preis et al. 11
comparisons. Five of the six pairwise com-parisons are statistically significant with p-values below 0.001. The greatest association(indicated by phi) is between the Seattle andLos Angeles methods, meaning that there isrelatively high correlation between thosetracts considered at risk and not at risk inboth methods.
Map matrix
Figure 2 shows a matrix of the four methods– along the diagonal – as well as the tracts
that are excluded in the method listed alongthe horizontal axis and the tracts that areincluded in the method listed along the verti-cal axis, visualising the findings fromTable 2. The general disagreement betweenthe maps is immediately apparent. ThePortland, Los Angeles and Philadelphiamethods all have at least 40% of their at-risktracts considered not at risk in anothermethod; the average percentage of tractsretained in other models is 63%. This find-ing reveals significant heterogeneity in tractsidentified by the different city models.
Figure 2. Map matrix: Analysis of census tracts identified as ‘At Risk’ versus ‘Not at Risk’ by the fourmethodologies as applied to Boston, MA.Notes: The map displays all Census tracts which are considered ‘Not at Risk’ by the column method, but considered ‘At
Risk’ by the row method. For example, in the bottom row, the left three columns show which Census tracts were
considered ‘At Risk’ by the Los Angeles method (i.e. included), but ‘Not at Risk’ by the Portland, Philadelphia, and Seattle
methods, respectively (i.e. excluded). We used the same ‘Not at Risk’ grouping here with our statistical analysis. Along
the diagonal is the city’s method applied to Boston.
12 Urban Studies 00(0)
Figure 2 is helpful in understanding thesignificance of Table 2. Figure 2 reveals, forexample, one tract in the Philadelphia modelthat is excluded from the other three.Philadelphia’s unique exclusion of most ofDorchester – a neighbourhood with a sub-stantial African American population – is alsoilluminating. Portland’s inclusion of a numberof East Boston tracts – a neighbourhoodwhere over half of residents identify as Latinoand that is undergoing intense residentialdevelopment – is also evident, as is Seattle’suniqueness in identifying as at risk most ofAllston and Brighton, neighbourhoods thatare currently witnessing increased institutionalinvestment with major re-zoning and transitimprovements planned, as well as the expan-sion of the Harvard University campus. Inspite of the fact that anywhere between 25tracts (in the Philadelphia method) and 119tracts (in the Seattle method) are identified asat risk of gentrification and displacement,there are only seven tracts that all four modelsagree are at high risk of gentrification and dis-placement (Figure 3).
Differences in population covered
In order to compare the methods withexpected outcomes, we describe the character-istics of the total population that each modelidentifies as at risk. If gentrification occurs inneighbourhoods with higher proportions oflow-income individuals, people of colour,those with lower educational attainment andrenter households, these populations wouldbe expected to be disproportionately repre-sented in the tracts that the models identify asat risk. We developed Table 3 expecting tofind roughly similar proportions among thefour models, given that they purport to mea-sure the same phenomenon.
Table 3 shows that Seattle, Portland andLos Angeles, as expected, all have a muchgreater share of non-white residents at riskof gentrification or displacement than in thecity overall. Surprisingly, in Philadelphia’smodel, renters and individuals in poverty areidentified as at risk only in equal proportionto the city-wide proportion. The Portlandand Seattle models’ results in relation toBoston’s black population are also striking,
Figure 3. Seven census tracts identified as ‘At Risk’ by all four methodologies.
Preis et al. 13
with Portland counting nearly twice the pro-portion of the black population as being atrisk than the proportion of the city-widepopulation, and Seattle’s at-risk tractsencompassing over 90% of Boston’s blackpopulation. Though no model included evic-tion data in their analysis, every model has agreater portion of the city-wide evictions intheir at-risk tracts than the portion of totalhouseholds in those tracts.4
Discussion and conclusion
The results of mapping all four methodsonto Boston, and our subsequent analysis,show significant differences between the fourmethodologies in the number and locationof tracts identified as vulnerable to gentrifi-cation and gentrification-related displace-ment. While each method aims to identifytracts experiencing, or at risk of, gentrifica-tion and related displacement, there aremajor differences in how each effort opera-tionalises the concept. From different
variable choices to varying risk thresholds,the assumptions embedded within the meth-ods have significant effects on what tractsare identified as vulnerable and, in turn,where city policy responses would be tar-geted if these methods were used.
The seven tracts identified as gentrifyingor at risk by all four methods (Figure 3) areconsistent with anecdotal accounts of neigh-bourhoods experiencing gentrification anddisplacement in Boston. Two of the tracts liein northern Dorchester, two in northernRoxbury, one in Jamaica Plain, one inDowntown/Chinatown and one in EastBoston. These are the five neighbourhoodsmost often discussed as under threat of gen-trification and displacement in Boston (see,e.g., Acolin and Vitiello, 2018). We do notsuggest that identifying the intersection ofall four methods will lead a practitioner tothe ‘true’ at-risk neighbourhoods, but thefindings suggest that the tracts on whichthese methods do agree represent some ofthe highest-risk areas of the city.
Table 3. Differences in population covered by different mapping methods, and subpopulations.
Philadelphia 16 15 17 15Portland 22 30 28 25Los Angeles 32 36 43 39Seattle 73 79 86 80
% Bostontracts at risk
% total households % evictions % city area
Philadelphia 14 13 16 11Portland 20 20 32 16Los Angeles 32 33 43 22Seattle 66 69 88 48
14 Urban Studies 00(0)
Other areas of agreement illustrate howsimilar choices of variables, even with differ-ent weighting (see Table 1), lead to similaroutcomes. For instance, both Los Angeles’and Seattle’s models identify tracts at riskalong the route of the fixed-rail Orange line,because of their inclusion of transit as a pre-dictor of risk. The inclusion of race as anindicator of risk in the Portland, LosAngeles and Seattle models leads all three tohave relatively large coverage of Dorchester,East Boston and Mission Hill, all neighbour-hoods with a large share of residents of col-our. The variables included in or excludedfrom the different methods relate at the mostfundamental level to their authors’ decisionsabout the most salient causes and indicatorsof gentrification, highlighting the degree towhich they understand gentrification to bedriven by private investment, rent gaps,state-led public investment or changing con-sumer preferences towards city living, to givea few examples.
Seattle’s model, with its 14 variablesacross individual, property and neighbour-hood characteristics, includes measures thatreflect multiple theories regarding the causesof gentrification. By comparing rent pricesand development potential, it incorporatesthe rent gap theory directly. The numerousvariables regarding income and race incor-porate attention to social dimensions ofhousehold vulnerability to gentrification anddisplacement. By including amenities such aspublic transit, schools, community centres,restaurants, grocery stores and job centres,the Seattle method also reflects consider-ation of both state-led and consumption-based conceptions of gentrification.
The Los Angeles method similarlybridges the gap between conceptions of gen-trification as driven by rent gaps, attentionto the social-structural dimensions of house-hold vulnerability to gentrification and the-ories emphasising the role of consumptionof public and private goods in
neighbourhood change. The NeighborhoodChange Index within the Displacement RiskIndex uses a base of household-level eco-nomic and social conditions that reflect thesalience of rent gaps and consideration ofracial and other disparities in the ability toresist displacement. The inclusion of accessto public transit as a risk factor and theunique measurement of affordable housingavailability as a mitigating factor reflect theinterplay between state-led andconsumption-based models of gentrification.The inclusion of data on affordable housingilluminates where low-income populationsmay be protected through publicly subsi-dised housing; where low-income popula-tions might be at risk, as affordabilityrestrictions come to an end and subsidisedunits are eligible for conversion to marketrates; and also where cities might considerinvesting in affordable housing preservationor new construction. Local housing authori-ties or other local government agenciescould develop such databases to replicatethis method in other cities.
The Portland method does not includedata on neighbourhood amenities such aspublic transit but focuses on housing tenureas well as housing market spillover effectsfrom nearby neighbourhoods, reflecting anemphasis on private investment-led gentrifi-cation. Portland, like Seattle and LosAngeles, also includes data on racial compo-sition and other neighbourhood demo-graphic characteristics. People of colour andrenter households are more likely thanwhites and homeowners to live in neighbour-hoods where rent gaps (Smith, 1996) exist,as a result of historic discriminatory policiessuch as redlining, and therefore they may bemore vulnerable to gentrification driven byprivate investment today. The method alsoacknowledges that this type of privateinvestment can have spillover effects asinvestors look for nearby areas in which realestate can generate high rates of return.
Preis et al. 15
The Philadelphia method focuses almostexclusively on the rent gap theory of gentrifi-cation, excluding variables on race andhousing tenure that all of the other meth-odologies include. Operationalising gentrifi-cation in this way assumes that tenants andhomeowners as well as people of colour andwhites are equally vulnerable to gentrifica-tion and that increases in income and hous-ing costs alone are the clearest indicators ofgentrification. The method lacks variablesrelated to public investment in neighbour-hoods, which would be reflective of a state-led conception of gentrification, as well asamenities that would measure consumption-based theories of gentrification risk.
Simply stated, the four models representthe operationalisation of multiple theories ofgentrification and the adoption of varyingmethods developed within academia to mapgentrification. Although cities are adaptingthese theories in their own methods, it isunclear the extent to which practitioners areconsciously choosing from this combinationof theories about the causes of gentrificationin order to match their local context. Giventhat Denver has directly adopted Portland’smethod for mapping gentrification (DenverOffice of Economic Development, 2016) andBoston was in the process of adopting theSeattle method when we began our project(Bousquet, 2017), this policy transfer leadsto the inference that cities may not be care-fully tailoring the methodologies to take intoaccount local context when mappinggentrification.
The findings here show that cities thatadopt one of the various different mappingmethods will come to very different conclu-sions about the location and severity of gen-trification based on the method they choose.To the extent that the causes of gentrifica-tion are informed by local historical andcontextual factors, there is reason to ques-tion the wholesale adoption of models devel-oped within academia or the adoption of a
model used in one city for another city, with-out attempting to account for the particula-rities of gentrification experienced in eachindividual city, or at least consciously choos-ing between different methodologies and thetheories that they operationalise.
It would serve practitioners well to con-sider the assumptions of the model they areadapting when they are adopting and modi-fying it. For instance, cities that are experi-encing rapid demographic change andincreasing housing costs may consideraspects of Portland’s approach, while citiesthat are expecting significant public invest-ment in the form of transit and other publicamenities may wish to consider adoptingsome of the data sources used in the LosAngeles or Seattle methods.
The growing availability of novel sourcesof data means that future efforts to map gen-trification may need to evolve beyond earlieracademic models. Given our belief that gen-trification is informed by local context,researchers and practitioners alike may wishto validate the models using ground-truthingof local conditions (see, e.g., Chapple et al.,2017: Appendix J), to see how the outputsrelate to local understandings of gentrifica-tion and its impacts.
Acknowledgements
We would like to acknowledge Sarah Williamsand Eric Huntley for their endless support duringthe process of replicating and verifying the fourmodels.
Declaration of conflicting interests
The author(s) declared no potential conflicts ofinterest with respect to the research, authorship,and/or publication of this article.
Funding
The author(s) received no financial support forthe research, authorship, and/or publication ofthis article.
16 Urban Studies 00(0)
ORCID iD
Benjamin Preis https://orcid.org/0000-0002-0412-310X
Note
1. We limited our search to large cities for threereasons. First, the public and much of theacademic focus on gentrification has been onits effects on inner-city neighbourhoods.Second, most academic studies of gentrifica-tion have focused on major cities. Third,given the data intensive nature of creatingthese maps, we assumed that larger citieswould have the resources and available datato devote to these maps’ creation, whilesmaller cities may not.
2. For specific guidance on the procedures foreach methodology, please contact the corre-sponding author.
3. When creating the binary variable, thefollowing categories were reduced to ‘Not at
Risk’: Portland: Not at Risk, Susceptible;Philadelphia: Not Gentrifiable, Nongentrifying,Stalled Gentrification; Seattle: Very Low,Low Risk; Los Angeles: Over Income, LowRisk. All other categories were considered‘At Risk’.
4. This research uses data from the Eviction Labat Princeton University, a project directed byMatthew Desmond and designed by AshleyGromis, Lavar Edmonds, James Hendrickson,Katie Krywokulski, Lillian Leung and AdamPorton. The Eviction Lab is funded by the JPB,Gates and Ford Foundations as well as theChan Zuckerberg Initiative. More informationis found at evictionlab.org. While this data pro-vides a new source of data on evictions, espe-cially for areas where the data was notpreviously collected, it undercounts the extent ofdisplacement in cities like Boston, as it excludesmore common processes of informal eviction, inwhich rents are raised or a notice to quit is filedand the tenant moves out without the landlordever going to court (Aiello et al., 2018).