Applied Geography 40 (2013) 61e70

Contents lists available at SciVerse ScienceDirect

Applied Geography
journal homepage: www.elsevier.com/locate/apgeog

Spatial analysis of urban decline: The geography of blight

R.C. Weaver a, *, Sharmistha Bagchi-Sen b
a
University of Redlands, MS GIS Program, 1200 E. Colton Ave, Box 3080, Redlands, CA 92373, USA
b
Department of Geography, University at Buffalo, SUNY, 105 Wilkeson Quad, Buffalo, NY 14261, USA

a b s t r a c t
Keywords: Urban studies literature suggests that anti-blight resources are frequently deployed in arbitrary fashions
Urban for short-term political objectives, rather than in concentrated, empirically-driven ways intended to
Blight
manage complex urban problems. This creates an ambiguous and subjective conceptualization of blight
Decline
Policy
in practice, which often leads to mismatches between actual urban conditions and codified public policy
targets. Therefore, this research points to the practical need for an operational definition of blight. It is
reasonable to assume that focusing anti-blight efforts in spaces identified using empirical data will in-
crease the efficiency of a city’s policy efforts. To that end this paper quantifies blight with an approach
that is replicable by virtually any city in the United States. We then examine blight patterns for a selected
city using spatial clustering methods that highlight areas where policy intervention might be warranted.
The findings demonstrate how spatial analysis combined with contextual urban geographic information
can assist local policymakers in identifying and understanding the geographies of blight in their
municipalities.
Ó 2013 Elsevier Ltd. All rights reserved.

Introduction reiteratively exhibit “piecemeal and inventive patterns” that vary

markedly within and between cities and across municipal admin-
Despite the decades-long crusade against “blight” initiated by istrations (Gordon, 2008). Gordon (2004) notes that politicians and
the urban renewal programs of the mid-20th Century, neither governing boards have declared blight in everything from shopping
policymakers nor social scientists have settled on a definition of the malls without desired anchor stores to entire cities; not because the
phenomenon (e.g., Brown, 2004; Robick, 2011; Shlay & Whitman, malls or cities were in deplorable conditions, but so that bigger
2006; Weber, 2002). Moreover, there is no collective appetence malls could be built and better tax breaks could be offered.
among either of these groups to ascribe a standardized meaning to These examples illustrate how anti-blight resources are
the concept. Urban scholars tend to describe blight as an “idea” held frequently deployed in arbitrary fashions for short-term political
in the minds of area stakeholders (Breger, 1967; Shlay & Whitman, ends, rather than in concentrated, empirically-informed ways
2006). In that sense the term is not associated with a single intended to manage a complex urban social and economic problem
denotation, since no two stakeholders are likely to have the same (Gordon, 2004; Weber, 2002). Put differently, an ambiguous
“idea” of what blight is (Shlay & Whitman, 2006). Accordingly, the conceptualization of blight can lead to mismatches between actual
theoretical notion of blight involves dynamic social negotiations urban conditions and codified public policy targets. Even the most
amongst actors with different preferences (Breger, 1967; Shlay & well-intentioned of blight policies, including those that aim to fix
Whitman, 2006). A similar issue arises in policy settings, insofar the problem via trickle-down economic effects, insubstantially
as extant anti-blight efforts in the U.S. often take the form of eco- reduce blight because they fail to understand its spatial and dy-
nomic redevelopment projects, funded in large part by federal namic properties (e.g., Swyngedouw et al., 2002). Consequently,
programs, in areas subjectively declared to be “blighted” by poli- this research points to the practical need for an operational defi-
cymaking officials (e.g., von Hoffman, 2000; Somin, 2006; Weber, nition of blight. It is reasonable to assume that focusing anti-blight
2002). Under these circumstances, local blight declarations efforts in spaces identified using empirical data will increase the
efficiency of a city’s policy efforts. To that end this paper quantifies
blight with an approach that is replicable by virtually any city in the
* Corresponding author. Tel.: þ1 9097488525.
United States. We then examine blight patterns for a selected city
E-mail addresses: [email protected] (R.C. Weaver), geosbs@ using spatial clustering methods that highlight areas where policy
buffalo.edu (S. Bagchi-Sen). intervention might be warranted. The findings demonstrate how

0143-6228/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.apgeog.2013.01.011
62 R.C. Weaver, S. Bagchi-Sen / Applied Geography 40 (2013) 61e70

spatial analysis combined with contextual urban geographic in- a commission of the fallacy of composition, that is, attributing a
formation can assist local policymakers in identifying and under- behavior (blight-generation) to a group because it is believed to
standing the geographies of blight in their municipalities. apply to a subset of group members. And while there is intuitive
appeal to equating the conditions of poverty and vacancy with
Background and context blight, it is somewhat more likely that these phenomena are
causally linked or correlated rather one and the same (Glaeser &
A lengthy body of interdisciplinary literature discusses the pit- Gyourko, 2005). Finally, the index does not explicitly capture
falls associated with treating blight subjectively in public policy what is typically agreed to be a key component of blight: qualitative
(Adams, 2005; Brown, 2004; Gordon, 2004, 2008; Somin, 2006, decline in real property conditions (Breger, 1967; Brueckner &
2008; Weber, 2002). Legal scholars in particular observe that under Helsley, 2009).
existing institutions the term “blight” functions as a powerful The critical role played by real property deterioration in un-
policy and rhetorical device that legally authorizes local govern- derstanding blight is best articulated by Breger (1967), who iden-
ments to fund or subsidize private economic development projects tifies “three elements that unify the blight concept”: (1)
simply by declaring properties to be “blighted” (e.g., Brown, 2004; nonacceptance; (2) real property; and (3) depreciation. More than
Eagle, 2007; Pritchett, 2003). It is this fact that places the urban four decades after Breger’s publication, these components continue
blight problem in desperate need of the type of precise, empirical to permeate modern blight theory (Gordon, 2004; Hartshorn, 1971;
redefinition proposed here. That a blight label enables the “at will” Krumm & Vaughan, 1976; Shlay & Whitman, 2006) and practice
acquisition and expenditure of real property and public dollars for (Brown, 2004; Eagle, 2007; Pritchett, 2003; Robinson & Cole, 2007;
subjective interests reasonably generates a demand for a problem Somin, 2006, 2008). Briefly, real property is viewed as that which
definition that discourages or minimizes rent-seeking behavior becomes blighted through a process of depreciation, during which
among the city officials who manage such resources (Brown, 2004). it transitions from a state of “acceptance” to one of “nonacceptance”
Nevertheless, there are decidedly few published endeavors to (Breger, 1967). Blight therefore involves qualitative decline in real
meet this demand, to quantify the phenomenon, or, by extension, to property conditions (Brueckner & Helsley, 2009; Spivack, 1991),
apply empirical data to blight research (Adams, 2005). In one of the such that declining conditions generate negative externalities
few attempts to do so, Krumm and Vaughan (1976) represent blight which impose costs on community members to whom they are
in hedonic housing price equations by using discrete scores for unacceptable (Bourne, 1981). Hence, the reason that blight is an
seven categories related to a parcel’s physical appearance, level of ambiguous concept is because “acceptance” thresholds for property
maintenance, and overall noisiness. Although the authors find conditions vary considerably across individuals (Shlay & Whitman,
significant relationships between these attributes and housing 2006).
prices, their factor selection and scoring techniques require several As suggested earlier, the academic literature generally consents
layers of value judgments. Accordingly, the procedure is difficult to to this viewpoint (Shlay & Whitman, 2006). Namely, blight means
replicate by individuals interested in operationalizing blight for different things to different stakeholders, and its social and physical
broader policymaking and analytical purposes. Similar “eyeball” representations change across space and time (Breger, 1967; Shlay
methods have been applied to interesting cases elsewhere in the & Whitman, 2006). However, as mentioned above, the interplay
literature, but with analogous implications for replicability (Bales, between this view and its interpretation in political environments
1985; Seo & von Rabenau, 2011). tends to promote policies that favor short-term, subjective interests
More recently, Miekley (2008) quantifies the phenomenon as a over long-term public welfare (Gordon, 2004; Weber, 2002). This
function of five neighborhood factorsdabandoned buildings, un- tension establishes the need for an operational definition of blight
kempt property, vacant lots, graffiti, and litterdas well as local that can facilitate targeted policies and inform strategic de-
crime rates and property values (viz., low relative property values ployments of public resources to spaces identified with empirical
are called a condition of blight). Here the approach is similar to data (e.g., Haining, 2003).
Krumm and Vaughan (1976), except that it (i) adds crime and Toward that end, this paper argues that nonacceptance can be
property values into the calculus, and (ii) establishes what are substituted with noncompliance to eliminate subjectivity in an
intended to be objective thresholds for scoring each blight factor. operational definition of blight. A basic feature of any polity is that
Nevertheless, value judgments and benchmarks are still required what is prohibited under municipal law constitutes the minimum
for distinguishing between “signs of abandonment”, “some litter” set of unacceptable actions in a given community (e.g., Stone,
versus “significant litter”, and so forth (Miekley, 2008). What is 2002). Thus real property conditions that are noncompliant with
more, it is potentially inappropriate to equate the concepts of blight municipal housing and building codes are therefore unacceptable
and crime, as the two are frequently viewed as separate elements to members of the community in which such conditions exist.
within a (multi-)causal relationship (e.g., Sampson & Raudenbush, Following this logic, one way to quantify blight is to use violations
2004; Skogan, 1990; Wilson & Kelling, 1982). of real property codes, where such codes are essentially universal
A slightly more systematic approach by Adams (2005) abandons attributes of American cities (Ross, 1996).
the subjective factor scoring method altogether and offers a “blight One critique that can be applied to any attempt to quantify a
index” based on observable census data. In particular, the pro- complicated social problem in this way is that quantification
portions of block group (i) renter-occupied units, (ii) vacant units, necessarily reduces real world complexity (e.g., Wilson, 2011). For
(iii) units built before 1939, (iv) people with income below the this reason, representing blight, which is generally seen as a
poverty level, (v) people 25 years or older without a high school multifaceted problem that both reflects and is a reflection of larger
education, (vi) unemployed labor force members, and (vii) female- urban structural issues (Shlay & Whitman, 2006), as a single,
headed households, are transformed into standard scores and discrete event might be difficult to accept at first. However, to
summed to create a composite value. Although readily replicable paraphrase Wilson (2011), one cannot understand what one cannot
and not reliant on subjective rankings, the method has a few measure. Indeed, our suggested use of housing code violations as a
practical issues. First, it is somewhat unfair to include classes of proxy for blight is not unlike Wilson’s (2011) own use of holiday
citizensdrenters, the impoverished, the unemployed, the less decorations to proxy for prosocialitydit is founded on a theoretical
educated, and female heads of householddas blighting factors. Not understanding of the problem, and it facilitates the creation of
only does this imply that people can constitute blight; but it is also spatial knowledge about the phenomenon of interest.
 R.C. Weaver, S. Bagchi-Sen / Applied Geography 40 (2013) 61e70 63

While we recognize that there is certainly more complexity in Esri’s ArcGIS 10. Only properties having match likelihood scores of
blighted areas than unkempt or substandard property, when 87.5% or higher were retained in the sample. In all, 20,082 (95.5%)
grounded in the contextual urban geography of a particular city, of the property violations fit into this category. Point locations
substandard property has significant utility as a surrogate for blight were then aggregated to the census tract level to obtain values for
in applied geographical and policy analysis. Influential urban (1) total open code violations accumulated over the entire range
housing studies discuss blight as one of several interacting urban of dates, and (2) violations observed during each quarter-year
problems present within a multi-causal process of “urban decline”, interval of the time series (JaneMar, ApreJun, JuleSep, Octe
characterized by positive feedbacks, whereby increases in one Dec). For each of these quantities, the number of violations in a
element cause increases in another and vice versa (Glaeser & census tract is weighted by the reciprocal of the given tract’s
Gyourko, 2005; Grigsby et al., 1987; Rothenberg et al., 1991). The estimated stock of housing units at the end of the time series.
complementary problems associated with blight (and one another) Because real property is effectively the “at risk” population for
in these discourses are vacancy, low human capital, unemploy- blight (Breger, 1967), this procedure scales violation counts so
ment, and poverty (Glaeser & Gyourko, 2005). To the extent that that clusters are not mistakenly identified in areas with larger at
these phenomena correlate with substandard property, code vio- risk populations. The resulting measure can be thought of as a
lations can reasonably be interpreted as blight in the sense that census tract’s “blight per unit” of housing (BPU). Finally, each tract
they are physical manifestations of urban declineda common view is merged with relevant socioeconomic status (SES) indicators
in the blight literature (Breger, 1967). taken from the 5-Year U.S. American Community Survey (ACS) for
This research applies the above approach to an empirical 2005e2009 (Table 1).
investigation of the geographic patterns of blight (property code Table 2 affirms the commonly held view that blight tends to be
violations) in Buffalo, NY. Buffalo is selected for its rapid and associated with other factors of “urban decline”, such as vacancy,
consistent levels of depopulation and decline since the deindus- low human capital (measured as the percentage of the adult pop-
trialization of urban America in the mid-20th Century. Explicitly, ulation without a high school diploma), poverty, and unemploy-
Buffalo, Detroit, St. Louis, and Cleveland have uniquely experienced ment (Glaeser & Gyourko, 2005). Moreover, the right side of Table 2
such continuous and consequential shrinkage since 1950 that “vast advises that blight is not only associated with the levels of these
stretches of their urban lands [have been] virtually deserted” complementary urban problems within a given geographic unit;
(Rappaport, 2003). In addition, the New York State Legislature but it is also positively correlated with such problems in sur-
indirectly linked blight in Buffalo to housing code violations when rounding geographic units. Thus if policymakers are able to identify
it established a dedicated city housing court in the 1970s to blight with empirical data, deploy resources to targeted spaces, and
“address a crisis of deteriorating housing stock” (Buffalo Housing reduce blight in those areasdthen it is entirely plausible that such
Court, 2004). focused interventions can aid in stopping the spread of blight and
Thus while the methods employed herein can be applied equally urban decline both within the targeted geographies and into
to any blighted (or non-blighted, for that matter) city having nearby neighborhoods. It is this possibility that motivates our
codified real property regulations, Buffalo is of particular interest operational definition and empirical investigation of blight, which
for its historical urban decline and judicial acknowledgments of the we carry out with the following methods.
city’s substandard housing issues. Spatial and aspatial data for the
city show strictly positive and statistically significant spatial auto- Tests for the detection of clusters
and cross-correlations within and between blight (code violations),
vacancy, low human capital, unemployment, and poverty inside the Tests for the detection of clusters involve surveilling all the sub-
study area (Data and methods section). Hence we again stress the geographies of an entire study area with the purpose of discovering
importance of quantifying blight so as to identify strategic locations “hot spots” of an event of interest for which the researcher does not
for policy interventiondif arbitrary blight declarations result in pre-specify focal locations for analysis (Besag & Newell, 1991).
suboptimal resource deployments, then poor blight management Hence the objective is to identify “areas that merit further inves-
can contribute to increases in a host of other socioeconomic tigation” (Besag & Newell, 1991). Accordingly, such tests can be
problems related to urban decline (e.g., Brueckner & Helsley, 2009; highly effective at targeting spaces for policy intervention.
Glaeser & Gyourko, 2005). The overall goal of this research is thus to These tests can be broadly subdivided into two classes by their
demonstrate how an operational definition of blight, contextual temporal scope of analysis. Retrospective tests are static assess-
information about a city’s urban geography, and applied spatial ments of cluster locations based on data collected for a specific time
analysis can aid planners and policymakers in their collective ef- (Rogerson & Yamada, 2009). Such tests operate by either applying
forts to improve conditions of blight in urban areas. local statistics (e.g., Anselin, 1995; Getis & Ord, 1992) to all subareas
in a study region, or examining the maximum of applied local
Data and methods statistics across subareas (e.g., Rogerson, 2001). Prospective tests,
on the other hand, involve computing test statistics immediately as
The data for this study come from an information request
made to the City of Buffalo Department of Management Infor-
Table 1
mation Systems in early 2010. The query was for open property Descriptive statistics.
code violations from the time the Department’s real property
Variable Mean Std. dev. Minimum Maximum
database was brought online through the end of 2009, and it
yielded 21,032 open violations. Each violation in the resulting Property violations 234.3 168.9 3 634
Housing units 1672.1 830.8 104 3816
dataset featured an associated address and the date it was entered a
BPU (blight per unit of housing) 0.149 0.087 0.003 0.365
into the system. The dates range from October 1, 1999 through %Pop 25 þ years, no HS diploma 0.210 0.099 0.037 0.512
June 30, 2009.1 All property addresses were batch geocoded in %Families below poverty level 0.245 0.174 0.0 1.0
Unemployment rate 0.124 0.076 0.0 0.412
Vacancy rate 0.200 0.110 0.022 0.561
1 N ¼ # of census tracts 91
Violations for the final two quarters of 2009 were not fully merged into the
a
database as of the request date. BPU ¼ # code violations/total housing units.
64 R.C. Weaver, S. Bagchi-Sen / Applied Geography 40 (2013) 61e70

Table 2
Blight and urban decline, aspatial and spatial relationships.

Pearson correlations Spatial auto- and cross-correlations

Let: row ¼ spatial lag of variable (j) and column ¼ original variable (i)

BPU %Family Unemp. rate %Pop 25 þ no HS BPU %Family Unemp. %Pop 25 þ no HS Vacancy
poverty poverty rate rate
BPU 0.315 0.133 0.133 0.278 0.281
%Family poverty 0.365 0.148 0.214
Unemp. rate 0.347 0.353 0.121 0.116
%Pop 25 þ no HS 0.514 0.563 0.325 0.277 0.332
Vacancy rate 0.643 0.290 0.401 0.434 0.267 0.371

All Pearson correlations significant at p < 0.01; all spatial auto- and cross-correlations significant at p < 0.05 BPU ¼ blight per unit of housing (see Table 1).

new data become available, thereby allowing researchers to iden- subareas yields the global Moran’s I plus the sum of the spatial
tify emergent clusters within the study area (Yamada et al., 2009). proximities (Rogerson & Yamada, 2009), thus making it a LISA
Rogerson (1997) and Rogerson and Yamada (2004) have applied (Anselin, 1995):
prospective tests in practice using both Shewart and cumulative XX X
sum (cusum) charts from quality control analysis. I ¼ wij Ii (3)
In this study we utilize both retrospectivedAnselin’s (1995) i j i
Local Moran LISA and Rogerson’s (2001) spatial Mdand
prospectivedmultiple univariate cusum (Rogerson, 1997)dcluster Spatial M test
detection tests to create a holistic picture of blight in Buffalo. The Rogerson’s (2001) spatial M test requires estimating values of
comparative approach of this paper offers a way to survey both interest with kernel-based smoothing and assessing the maximum
present conditions and their path-dependencies to identify of a set of local statistics. While the method has been used in
currently blighted areas and [perhaps] anticipate to what locations practice for, among other things, geovisualization and exploratory
blight is likely to spread in the absence of targeted intervention. analysis (see Rogerson & Yamada, 2009 at Sec. 5.10), the primary
concern here is with information that the test reveals about
Local Moran LISA neighborhood bandwidth. To carry out the test, we assign weight to
Local Indicators of Spatial Autocorrelation (LISA) are a class of the spatial relationship between two subareas on the basis of a
local statistics that atomize global statistics (e.g., Moran’s I) so that Gaussian kernel, given by:
(1) the type and significance of spatial clustering of an event of
2 3
interest is determinable at each subarea in a study region, and (2)
1 6 d2ij 7
the sum of all subarea LISAs in a given study region is proportional wij ¼ pffiffiffiffiffiffiexp4   . 5 (4)
to the relevant global statistic (Anselin, 1995). There are several ps 2s2 m A
varieties of LISAs (Anselin, 1995). A very common LISA, the local
Moran’s I (Anselin, 1995), has been used in urban studies to detect
where there are m subareas in the study region of area of A, dij is
clusters of such phenomena as depopulation (Wang & Meng, 1999),
the distance between subareas i and j, and s is the bandwidth
public services (Hewko et al., 2002), GDP per capita (Yu & Wei,
that defines the neighborhood around subarea i (Yamada et al.,
2003), poverty (Amarasinghe et al., 2005; Orford, 2004), unem-
2009).
ployment (Patacchini & Zenou, 2007), and housing prices
In practice, for cluster detection tests founded on kernel-based
(Baumont, 2009). Following these examples, we use the local
smoothing, such as the prospective tests discussed in the next
Moran LISA to test observed (cumulated) patterns of blight against
section, the value of s should be pre-specified as the size of the
the null hypothesis of spatial randomness.
anticipated cluster (Yamada et al., 2009). Clearly, however, for
Briefly, the local Moran LISA derives from the global Moran’s I,
research engaging in exploratory analysis with no preconceptions
which is a measure of spatial autocorrelation that ranges from 1
about the sizes or locations of clusters, this task is difficult at best.
to þ1. These boundary values represent high negative and positive
One of the advantages of the spatial M test is that it can be carried
spatial association for a quantity of interest, respectively (Moran,
out for many possible bandwidths, allowing one to identify the
1948). The global statistic is computed as:
value of s associated with the maximum local test statistic.
P Pm   Rogerson and Yamada’s (2009) GeoSurveillance software package
m mi j wij ðyi  yÞ yj  y automates this process. For our purposes, the spatial M test is
I ¼ P P P (1)
m m m 2 conducted in GeoSurveillance to inform bandwidth selection for a
i j wij i ðyi  yÞ
series of prospective cluster detection tests.

where a study region consists of m subareas, yi is the observed

Multiple univariate cusum charts
value of the quantity of interest, y is the observed mean value of the
To add dynamic insight to our study of blight in Buffalo, we
quantity of interest, and wij represents the spatial proximity be-
simultaneously monitor cumulative sum (cusum) charts (e.g.,
tween subareas i and j. In this study we define spatial proximity or
Montgomery, 1996) for each census tract over 39 quarter-year pe-
neighborhoods using queen contiguity (e.g., Anselin, 2005).
riods from October 1999 through June 2009. Such spatial cusum
The local Moran’s I for subregion i is therefore calculated as:
applications have been used by Rogerson (1997) and Rogerson and
  Yamada (2004), and suggest the following approach.
mðy  yÞ X m
Ii ¼ Pm i 2
wij yj  y (2) Let Oi;t equal the observed number of property code violations in
i ðyi  yÞ j census tract i during time t. Denote the expected number of vio-
lations for the relevant spaceetime combination Ei;t , which repre-
where all variables are defined as they were in Equation (1). It is sents the expectation under the null hypothesis of no raised blight
fairly straightforward to see that summing Equation (2) over all m incidence. In this sense, Ei;t can be computed as the product of the
 R.C. Weaver, S. Bagchi-Sen / Applied Geography 40 (2013) 61e70 65

Fig. 1. Blight clusters in Buffalo, NY detected with local Moran LISA for time period October 1999 through June 2009.

citywide level of “blight per unit” (BPU) of housing in Buffalo and used by other studies in the urban geography literature (e.g.,
the number of housing units in census tract i (refer to Background Baumont, 2009), a Bonferroni adjustment is applied.2
and context section and Table 1). The variable of interest (viola-
tions) can then be transformed into a z-score, such that the uni- Results and discussion
variate cusum in census tract i at time t is equal to:
To carry out the analyses described above, three of Buffalo’s 94
  census tracts (2000 decennial geographies) were dropped for
Si;t ¼ max Si;t1 þ zi;t  k; 0 containing zero housing units. Two of the three omitted tracts
(5)
Si;0 ¼ 0 wholly correspond to two university campuses, which expectedly
contained no property code violations. The third omitted tract
where k is a parameter set to 0.5, since it is chosen to minimize the corresponds to the city’s central business district (CBD), which
time to detect an increase of 2k standard deviations in the mean of features government buildings including City Hall and court houses
zi;t (Yamada et al., 2009). (City of Buffalo, 2002). Here, 19 violations were in fact geocoded to
Stated more simply, the cusum for tract i at time t is the accu- the CBD; however, these observations account for less than 0.1% of
mulated deviation in the mean value of violations for that tract. the overall violations from the dataset. Hence, it is assumed that all
Clusters are detected when the value of a cusum exceeds a pre- three tracts can be dropped without affecting the analytical results.
defined threshold. This threshold, h, is determined in conjunction
with a pre-specified “false alarm rate”, which operates similarly to
Local Moran LISA
type I error (Rogerson & Yamada, 2009; Yamada et al., 2009). The
false alarm rate corresponds to the average number of time in-
Definition of a spatial structure is a precondition for tests of
tervals between “false alarms” under the null hypothesis of no
spatial dependency. Hence, to utilize local Moran LISAs for blight
significant change to the mean number of violations in a given
cluster detection it is first necessary to define wij from Equation (2).
tract. Call this rate ARL0 (e.g., Rogerson & Yamada, 2009). A com-
It is this quantity which conveys information related to how sub-
mon value of ARL0 is 100, such that a “false alarm” occurs every 100
areas are spatially linked to one another. For this analysis, spatial
periods (e.g., Yamada et al., 2009). For this study, because viola-
structure is defined with a binary queen contiguity matrix, such
tions are observed during quarter-year intervals, ARL0 is set to
that wij is set equal to 1 if tracts i and j are contiguous in any di-
400dthat is, a “false alarm” will occur every 100 years by chance
rection, and 0 otherwise (e.g., Anselin, 2005). From this, Equation
alone.
(2) can be applied to all 91 census tracts contained in the analysis.
Given the desired false alarm rate, the threshold (h) can be
The results are pictured in Fig. 1, where panel (a) shows the loca-
approximated as in Rogerson (2006), though for convenience h is
tions and types of blight clusters detected, and panel (b) pictures
automatically calculated by GeoSurveillance once ARL0 and other
their associated pseudo-significance based on 10,000 random
parameters are specified (Yamada et al., 2009). Here “other param-
permutations.3 These patterns are considered in more detail below.
eters” include the (1) number of subareas, (2) bandwidth, and (3)
type of multiple testing correction method desired. Insofar as cusum
charts are applied to 91 census tracts in Buffalo simultaneously, one 2
A Bonferroni adjustment involves dividing the desired level of significance, a,
must account for the likelihood that typical significance levels are by the number of subareas in the study region (Rogerson & Yamada, 2009).
too likely to produce “signals” by chance alone (e.g., Rogerson & 3
This technique is more appropriate than actual significance given that the
Yamada, 2009). Following the multiple testing correction method global Moran’s I suggests statistically significant global clustering (Anselin, 1995).
66 R.C. Weaver, S. Bagchi-Sen / Applied Geography 40 (2013) 61e70

Table 3 Table 3 shows, however, two values satisfy this criterion (flagged
Spatial M test results. with an ‘X’ in the final column of the table). To break the tie, the
Bandwidth (s) Test statistic (M) Critical value (M*) (M  M*) bandwidth associated with the higher global score statistic is
0.0 42.2 3.3 38.9 chosen (Yamada et al., 2009). The global score in this case is 70,171.8
0.1 42.2 3.3 38.9 for a bandwidth of 1.1, and 70,810.4 when s is equal to 1.2.
0.2 42.3 3.3 39 Accordingly, the latter is selected.
0.3 45.3 3.3 42
0.4 50.2 3.3 46.9
0.5 55.0 3.2 51.8 Prospective cusum tests
0.6 60.3 3.2 57.1
0.7 67.4 3.2 64.2 For a bandwidth of 1.2, a false alarm rate of 400, a k parameter of
0.8 73.8 3.2 70.6
0.5, and a Bonferroni correction for multiple testing over 91 census
0.9 78.3 3.2 75.1
1.0 81.2 3.1 78.1 tracts (Data and methods section), the “cusum tool” in Geo-
1.1 82.6 3.1 79.5 X Surveillance calculates the threshold value (h) for analysis to be
1.2 82.6 3.1 79.5 X 8.644. That is, any census tract for which the cusum defined by
1.3 81.6 3.1 78.5 Equation (5) during period t exceeds 8.644 is signaled as the site of
1.4 79.8 3.0 76.8
1.5 77.4 3.0 74.4
an emergent blight cluster for that quarter. Based on this, the
1.6 74.5 3.0 71.5 cusum for each spatio-temporal combination of census tracts and
1.7 71.4 3.0 68.4 time periods in the dataset is computed in GeoSurveillance. “Sig-
1.8 68.2 2.9 65.3 nals” and near signals are mapped for selected quarter-year periods
1.9 65.0 2.9 62.1
in Fig. 2, where panels (a) through (d) depict the emergent clusters
2.0 61.8 2.9 58.9
detected for the first, first quartile (11th), third quartile (30th), and
X ¼ maximum difference between the spatial M test statistic and corresponding
final (39th) periods of observed data. Note the differences between
critical value.
the dynamically detected blight clusters here and the static clusters
detected in Fig. 1.
Spatial M test
Comparisons, locations, and practical interpretations of the clusters
It was explained earlier that the primary motivation for per-
forming a spatial M test here is to gain knowledge about the Note that from Equation (5), the cusum methods used here and
bandwidth that defines neighborhoods in the study region. That implemented in GeoSurveillance are one way. Put differently, for the
being said, we call on GeoSurveillance software (Yamada et al., dynamic component of the analysis the methods detect clusters of
2009) to carry out the test for a range of bandwidths from 0.0 to blight exclusively, and not clusters of “non-blight” analogous to the
2.0 in increments of 0.1. In order to proceed with this operation, two “LoweLow” areas identified by the local Moran LISAs in Fig. 1. Thus
values are needed: the observed (Oi;t ) and expected (Ei;t ) number of while it is possible to discuss where each method detects signifi-
violations in each tract. cant blight clusters, a reciprocal comparison of significant
Yamada et al. (2009) demonstrate this exercise for a base period unblighted areas is not available.
of interest. For the Buffalo data, recall that the “at risk” population With that it is convenient to begin by comparing the static
for a given census tract is taken to be its number of housing units clusters detected by the local Moran LISAs in Fig. 1 to those
estimated by the 5-Year U.S. Census American Community Survey dynamically detected for the final period of the cusum analysis in
(ACS) for 2005e2009. Because the ACS reports period estimates the lower right panel of Fig. 2. In effect, both figures map clusters of
covering a five-year stretch, the most compatible “base period” of blight based on the end-period, cumulated stock of property vio-
interest should describe the annual average number of violations lations per housing unit in each census tract. To aid in this com-
observed in each census tract from 2005 to 2009. This value is parison, Fig. 3 represents each census tract as part of a “planning
calculated in the following manner, adjusting for the two missing community” and smaller “planning neighborhood” classified by the
quarter-year periods at the end of 2009: City of Buffalo Office of Strategic Planning (2002).
The most obvious difference between Fig. 1 and the bottom right
P panel of Fig. 2 relates to the extent of the largest cluster detected by
4 2009:Q 2
t¼2005:Q 1 Violationst the two methods. Namely, while both methods reveal sizeable,
Oi;20052009 ¼ (6)
18 contiguous blight clusters in the area of the “East Side”, “East
Delavan”, and “Masten” planning communities,4 the dynamic
where multiplying the numerator by 4 transforms the observed analysis estimates the area to be much larger than what is detected
values from quarter-year quantities to full year quantities, and 18 is by the local Moran LISAs. Specifically, in the final period of the time
the number of periods from the first quarter in 2005 through the series, the cusum methods signal the entire East Delavan, East Side,
last observed quarter (Q2) in 2009. Equation (6) therefore provides Masten, and Ellicott planning communities, as well as nearly all of
a tract-level period average of violations that is comparable with the Buffalo River community. The resultant cluster encompasses 32
the period average of each tract’s housing unit stock reported by the census tracts, compared to 10 in its static analog.
ACS. That the dynamic tests identify a larger eastern cluster relative
Next, a Poisson-based transformation is performed on the to the LISA method is partially due to the temporal information
observed values after Equation (6) is applied to all census tracts contained in the former. Recall that the cusum analyses signal
(Rogerson & Yamada, 2009; Yamada et al., 2009). Subsequently, census tracts when their observed deviations from the expected
GeoSurveillance performs the spatial M test over the desired range amount of blight per unit (BPU) of housing exceed the pre-specified
of bandwidths. The results appear in Table 3. threshold. Signals therefore indicate that BPU in a given census
Yamada et al. (2009) point out that a reasonable bandwidth
selection is the value which maximizes the difference between the
spatial M test statistic and its corresponding critical value. We 4
Henceforth, refer to Fig. 3 for all mentions of “planning communities” or
follow this approach for its convenience and intuitive appeal. As “planning neighborhoods” by name.
 R.C. Weaver, S. Bagchi-Sen / Applied Geography 40 (2013) 61e70 67

Fig. 2. Emergent blight clusters in Buffalo, NY detected with cusum methods during four selected quarter-year time periods.

tract has stabilized at a new, higher level relative to its past levels. aggregate BPU for all observed violations in the subregions over the
To the extent that expected BPU in a given quarter year is calculated entire time series. For this reason, it does not convey how a given
based on the citywide rate of blight for that time period, such tract’s relative level of blight changes with the citywide level over
signals capture information both about fluctuations in overall time. When such information is taken into consideration, the
violation trends as well as past intra-census tract blight conditions. dynamically detected cusum eastern cluster illustrates that BPU has
By contrast, the LISA method detects clusters based on the mean elevated to new, higher levels in over three times the subregions
68 R.C. Weaver, S. Bagchi-Sen / Applied Geography 40 (2013) 61e70

Fig. 3. City of Buffalo, NY planning communities and neighborhoods.

signaled by the static cluster. This is also true of a smaller cluster in data, then they are unlikely to fundamentally address its inherent
the northwest section of the citydin the “Riverside” and “West spatialities and inter- and path-dependencies.
Side” planning communitiesdwhich is not detected by the LISAs. In As noted above, a second key difference between Fig. 1 and the
this context, it can be argued that static snapshots of blight patterns bottom right panel of Fig. 2 is the former’s depiction of “LoweLow”
at single points in time might fail to yield key information about the clusters, or areas in which the observed BPU in two or more
true geographical extent of the problem. The implication is that if contiguous subregions is at least two standard deviations below the
policymakers govern complex blight problems with purely static expected value. These clusters occur in the “North Park”, “Park
 R.C. Weaver, S. Bagchi-Sen / Applied Geography 40 (2013) 61e70 69

Meadow”, and “Parkside” planning neighborhoods in the north, the In this sense, measurement is not useful merely for understanding,
“Bryant” and “Grant Ferry” neighborhoods in the west, and the but for policymaking as well.
“Abbott McKinley” and “South Park” neighborhoods in the South. Using Buffalo, NY as a study area, tests for the detection of
Not coincidentally, these spaces house some of the city’s most clusters reveal that static snapshots of blight patterns at single
notable cultural resources. For example, the northern “non-blight” points in time might fail to yield key information about the true
cluster features Delaware Park, the largest of the city’s Olmsted geographical reach and complexity of the problem. Spatial
Parks and a focal feature in a historic neighborhood containing methods of cluster detection contextualized with descriptive sta-
early 20th Century mansions and other cultural attractions (City of tistics about Buffalo’s urban geography clearly demonstrate that
Buffalo, 2006). The western cluster represents a portion of Buffalo’s blight and urban decline are positively related in time and space.
Elmwood Village, a mixed-use strip of shops, restaurants, art gal- Expressly, dynamic cumulative sum analysis suggests that blight is
leries, and housing that has been named one of the country’s “10 an inherently spatial phenomenon that spreads from its initial
Best Neighborhoods” by the American Planning Association (2007). locations to contiguous territories over time. Our analysis for
Lastly, the southernmost cluster houses an “Irish Heritage District” Buffalo illustrates how planners and policymakers can use
featuring tree-lined parkways, high homeownership rates, and geographic concepts and spatial analysis to gain an understanding
adjacency to the city’s second largest Olmsted Park and botanical of potential issues before the problems become widespread. The
gardens complex, which sits next to a historic and elegant cathedral theoretical contribution of this study is therefore that it provides a
(City of Buffalo, 2006). In this sense, a second, slightly looser dynamic view of urban change. Continuous monitoring of blight
interpretation of Fig. 2 is that “HigheHigh” and “LoweLow” clus- within a Geographical Information Systems (GIS) framework has
ters illustrate where investment in real property is likely to be low policy implications in terms of managing housing stock, neigh-
and high, respectively. borhood quality, land use, and ultimately population-based and
These findings have direct policy implications for blight in the economic shrinkage.
city of Buffalo. As Fig. 2 illustrates, in the final period of the cusum As the data for Buffalo show, there exist direct and statistically
analysis, census tracts on the edge of “LoweLow” cluster planning significant associations between blight in one neighborhood in a
neighborhoods in “Elmwood”, “South Buffalo” and “Northeast” are city and poor socioeconomic conditions in adjacent neighborhoods.
approaching signal levels. In other words, blight from the identified Because these phenomena are interconnected within a complex
clusters is spilling over neighborhood boundaries and threatening process of urban decline characterized by positive feedbacks
stable areas. Moreover, there are no signs that high levels of blight (Glaeser & Gyourko, 2005), failing to control blight in one location
within the clusters are being successfully mitigated, as all but five is therefore tantamount to insufficiently addressing issues of
signaled tracts from period 30 are signaled again in the final period. poverty, unemployment, low human capital, and vacancy both in
Compounding this, one non-signaled tract from period 30 in the given area and its surroundings. Alternatively, when increases
“Riverside” is signaled in the final period, and three additional non- in these conditions are observed in nearby territories, it is likely
signals from period 30 approach signal level in period 39. Thus that conditions of blight and substandard housing are also on the
there is evidence to suggest that Buffalo’s blight policies are rise in a given neighborhood. Consequently, understanding the
underperforming expectations, and that anti-blight resource de- patterns of blight in a city requires knowledge of its relationships
ployments are not having significant blight-reduction effects. with socio-demographic and economic conditions. Future research
The signals from Fig. 1 and the lower right panel of Fig. 2 point to will extend the understanding of blight spillover from city to sub-
areas that are currently under threat of becoming blighted, and, as a urbs by expanding on city-level applications, such as the one pre-
result, potentially suggest clustered spaces to which anti-blight sented here, to include the metropolitan and regional levels.
resources can be redirected in the short- and long-term.
Combining this empirically-derived spatial knowledge with
References
descriptive statistics and contextual information about Buffalo’s
urban geography, such as neighborhood vacancy and poverty Adams, H. D. (2005). Toward determining patterns of residential blight and blight
levels, planners and policymakers gain the ability to effectively mitigation in St. Louis County, Missouri. Unpublished Ph.D. thesis, Saint Louis
University.
delimit the boundaries of blight and decline within the city. That is,
Amarasinghe, A., Samad, M., & Anputhas, M. (2005). Spatial clustering of rural
in contrast to arbitrarily making spatially-decentralized “blight poverty and food insecurity in Sri Lanka. Food Policy, 30, 493e509.
declarations” that serve subjective interests (e.g., Brown, 2004), American Planning Association. (October 2, 2007). Elmwood village named one of 10
Buffalo decision-makers can concentrate blight resources on great neighborhoods in America. Buffalo, NY: American Planning Association.
http://www.planning.org/newsreleases/2007/oct02-4.htm Accessed 01.07.12.
existing and emergent clusters of substandard housing. This, in Anselin, L. (1995). Local indicators of spatial association e LISA. Geographical
turn, transforms the unit of analysis in urban blight policy from the Analysis, 27(2), 93e115.
unscientifically-selected and scattershot parcel or property, to the Anselin, L. (2005). Exploring spatial data with GeoDa: A workbook. Urbana, IL: Center
for Spatially Integrated Social Science.
empirically-determined neighborhood or census tract. Strategically Bales, K. (1985). Determinants in the perceptions of visual blight. Human Ecology,
focusing resources in this way is likely to increase a city’s capacity 13(3), 371e378.
to successfully mitigate urban policy problems (Haining, 2003). Baumont, C. (2009). Spatial effects of urban public policies on housing values. Pa-
pers in Regional Science, 88(2), 301e326.
Besag, J., & Newell, J. (1991). The detection of clusters in rare diseases. Journal of the
Conclusions Royal Statistical Society, Series A (Statistics in Society), 154(1), 143e155.
Bourne, L. S. (1981). The geography of housing. Toronto: V.H. Winston & Sons.
Breger, G. E. (1967). The concept and causes of urban blight. Land Economics, 43(4),
This paper puts forward an operational definition of urban 369e376.
blight based on the assertion that the phenomenon involves Brown, C. S. (2004). Blinded by blight: a search for a workable definition of “blight”
deteriorating real property conditions in a given urban area (Breger, in Ohio. University of Cincinnati Law Review, 73, 207e233.
Brueckner, J. K., & Helsley, R. W. (2009). Sprawl and blight. In 46th Annual AREUEA
1967; Robinson & Cole, 2007; Spivack, 1991). Specifically, “blight” is
conference paper.
quantified as noncompliance with real property regulations. Given Buffalo Housing Court. (2004). About the court. 8th Judicial District, Erie County, NY:
that quantification tends to be one of the prerequisites for under- New York Courts. http://www.nycourts.gov/courts/8jd/erie/bcchousing.shtml
standing a social problem (Wilson, 2011), our replicable definition Accessed 01.07.12.
City of Buffalo, NY Office of Strategic Planning. (2002). City of Buffalo planning
affords planners and policymakers a way to consistently oper- neighborhoods and census tracts by planning community. http://www.ci.
ationalize blight and to analyze its spatial and dynamic properties. buffalo.ny.us/files/1_2_1/MapDesc/dd_panels_comunity.pdf Accessed 01.09.09.
70 R.C. Weaver, S. Bagchi-Sen / Applied Geography 40 (2013) 61e70

City of Buffalo, NY Office of Strategic Planning. (2006). Queen city in the 21st Rogerson, P. A. (2006). Formulas for the design of CUSUM quality control charts.
century: Buffalo’s comprehensive plan. http://urbandesignproject.ap.buffalo. Communications in Statistics e Theory and Methods, 35, 373e383.
edu/pub/pdf/BCP10-06.pdf Accessed 01.09.09. Rogerson, P., & Yamada, I. (2004). Monitoring change in spatial patterns of disease:
Eagle, S. J. (2007). Does blight really justify condemnation? Urban Lawyer, 4. comparing univariate and multivariate cumulative sum approaches. Statistics in
Getis, A., & Ord, J. (1992). The analysis of spatial association by use of distance Medicine, 23, 2195e2214.
statistics. Geographical Analysis, 24, 189e206. Rogerson, P., & Yamada, I. (2009). Statistical detection and surveillance of geographic
Glaeser, E. L., & Gyourko, J. (April 1 2005). Urban decline and durable housing. Uni- clusters. New York, NY: CRC Press.
versity of Pennsylvania Institute for Urban Research Publications. http:// Ross, H. L. (1996). Housing code enforcement and urban decline. Journal of Afford-
repository.upenn.edu/penniur_papers/10 Accessed 10.11.08. able Housing & Community Development, 6(1), 29e46.
Gordon, C. (2004). Blighting the way: urban renewal, economic development, and Rothenberg, J., Galster, G. C., Butler, R. V., & Pitkin, J. R. (1991). The maze of urban
the elusive definition of blight. Fordham Urban Law Journal, 31, 305e337. housing markets: Theory, evidence, and policy. Chicago, IL: The University of
Gordon, C. (2008). Mapping decline: St. Louis and the fate of the American city. Chicago Press.
Philadelphia, PA: University of Pennsylvania Press. Sampson, R. J., & Raudenbush, S. W. (2004). Seeing disorder: neighborhood stigma
Grigsby, W. G., Baratz, M., & Maclennan, D. (1987). The dynamics of neighborhood and the social construction of ‘Broken Windows’. Social Psychology Quarterly,
change and decline. Oxford: Pergamon. 67(4), 319e342.
Haining, R. (2003). Spatial data analysis: Theory and practice. New York, NY: Cam- Seo, W., & von Rabenau, B. (September 2011). Spatial impacts of microneighborhood
bridge University Press. physical disorder on property resale values in Columbus, Ohio. Journal of Urban
Hartshorn, T. A. (1971). Inner city residential structure and decline. Annals of the Planning and Development, 337e345.
Association of American Geographers, 61(1), 72e96. Shlay, A. B., & Whitman, G. (2006). Research for democracy: linking community
Hewko, J., Smoyer-Tomic, K. E., & Hodgson, M. J. (2002). Measuring neighbourhood organizing and research to leverage blight policy. City & Community, 5(2),
spatial accessibility to urban amenities: does aggregation error matter? Envi- 153e171.
ronment and Planning A, 34, 1185e1206. Skogan, W. (1990). Disorder and decline: Crime and the spiral of decay in American
von Hoffman, A. (2000). A study in contradictions: the origins and legacy of the cities. Berkeley, CA: University of California Press.
Housing Act of 1949. Housing Policy Debate, 11(2), 299e326. Somin, I. (2006). Blight sweet blight. Legal Times, XXIX(33).
Krumm, R., & Vaughan, R. J. (1976). The economics of urban blight. Washington, DC: Somin, I. (2008). The limits of backlash: Assessing the political response to Kelo. George
RAND Corporation. Mason University Law and Economics, Research paper series. http://ssrn.com/
Miekley, A. E. (2008). Counting broken windows: Pursuing an objective measurement abstract_id¼976298 Accessed 01.08.11.
of blight. Unpublished master’s thesis, Ohio State University. http://etd.ohiolink. Spivack, R. N. (1991). The determinants of housing maintenance and upkeep: a case
edu/view.cgi?acc_num¼ucin1226694198. Accessed 01.09.09. study of Providence, Rhode Island. Applied Economics, 23, 639e646.
Montgomery, D. (1996). Introduction to statistical quality control. New York: Wiley. Stone, D. (2002). Policy paradox: The art of political decision making, revised edition.
Moran, P. A. P. (1948). The interpretation of statistical maps. Journal of the Royal New York, NY: W.W. Norton & Company.
Statistical Society, Series B, 10, 245e251. Swyngedouw, E., Moulaert, F., & Rodriguez, A. (2002). Neoliberal urbanization in
Orford, S. (2004). Identifying and comparing changes in the spatial concentrations Europe: large-scale urban development projects and the new urban policy. In
of urban poverty and affluence: a case study of inner London. Computers, N. Brenner, & N. Theodore (Eds.), Spaces of neoliberalism: Urban restructuring in
Environment and Urban Systems, 28, 701e717. North America and Western Europe (pp. 195e229). Oxford: Blackwell.
Patacchini, E., & Zenou, Y. (2007). Spatial dependence in local unemployment rates. Wang, F., & Meng, Y. (1999). Analyzing urban population change patterns in She-
Journal of Economic Geography, 7, 169e191. nyang, China 1982e90: density function and spatial association approaches.
Pritchett, W. E. (2003). The ‘public menace’ of blight: urban renewal and the private Geographic Information Sciences, 5(2), 121e130.
uses of eminent domain. Yale Law & Policy Review, 21(1), 1e52. Weber, R. (2002). Extracting value from the city: neoliberalism and urban rede-
Rappaport, J. (2003). U.S. urban decline and growth, 1950 to 2000. Federal Reserve velopment. In N. Brenner, & N. Theodore (Eds.), Spaces of neoliberalism: Urban
Bank of Kansas City Economic Review, (third quarter), 15e44. restructuring in North America and Western Europe (pp. 172e193). Oxford:
Robick, B. D. (2011). Blight: The development of a contested concept. Doctoral Blackwell Publishing.
dissertation, Carnegie Mellon University. Wilson, D. S. (2011). The neighborhood project: Using evolution to improve my city, one
Robinson, & Cole. (2007). Urban blight: An analysis of state blight statutes and their block at a time. New York, NY: Little, Brown and Company.
implications for eminent domain. A report prepared for the National Association Wilson, J. Q., & Kelling, G. L. (1982). “Broken Windows.” Atlantic Monthly, March.
of REALTORS. Washington, DC: National Association of REALTORS. Yamada, I., Rogerson, P. A., & Lee, G. (2009). GeoSurveillance: a GIS-based system for
Rogerson, P. A. (1997). Surveillance systems for monitoring the development of the detection and monitoring of spatial clusters. Journal of Geographic Systems,
spatial patterns. Statistics in Medicine, 16, 2081e2093. 11, 155e173.
Rogerson, P. A. (2001). A statistical method for the detection of geographic clus- Yu, D., & Wei, Y. D. (2003). Analyzing regional inequality in post-Mao China in a GIS
tering. Geographical Analysis, 33(3), 215e227. environment. Eurasian Geography and Economics, 44(7), 514e534.