ArticlePDF Available

Using Spatial Statistics to Analyze Intra-urban Inequalities and Public Intervention in Sao Paulo, Brazil


Abstract and Figures

Like many cities in developing countries, Sao Paulo, Brazil, is characterized by major intra-urban inequalities with respect to human development. The center-periphery spatial regimes are the most obvious spatial manifestation of this phenomenon. In this paper we apply confirmatory spatial data analysis to examine these inequalities and their relationship to public interventions. Using district-level data, we examine the relationship between public interventions and the level of human development, while controlling for population density, spatial heterogeneity and spatial autocorrelation. Our results suggest that public interventions reinforce the existing differences between center and periphery. Specifically, public services and utilities and social programs are allocated more intensively in districts with higher human development levels. These findings call for a more careful consideration of distribution of societal resources and effectiveness of public programs and policies.
Content may be subject to copyright.
Using Spatial Statistics to Analyze Intra-
urban Inequalities and Public Intervention in
Sa˜o Paulo, Brazil
Moˆni˙ca A. Haddad is Assistant Professor in the Department of Community
and Regional Planning at Iowa State University, and Zorica Nedovic
is Associate Professor of Urban Planning and GIS at the University of Illinois
at Urbana-Champaign
Abstract Like many cities in developing countries, Sa˜o Paulo, Brazil, is
characterized by major intra-urban inequalities with respect to human
development. The center-periphery spatial regimes are the most obvious
spatial manifestation of this phenomenon. In this paper we apply
confirmatory spatial data analysis to examine these inequalities and their
relationship to public interventions. Using district-level data, we examine
the relationship between public interventions and the level of human
development, while controlling for population density, spatial
heterogeneity and spatial autocorrelation. Our results suggest that
public interventions reinforce the existing differences between center
and periphery. Specifically, public services and utilities and social
programs are allocated more intensively in districts with higher human
development levels. These findings call for a more careful consideration of
distribution of societal resources and effectiveness of public programs and
Key words: Spatial statistics, Intra-urban inequality, Public programs,
Urban policy, Developing countries, Human development, Economic
growth, Spatial autocorrelation, Spatial heterogeneity
Urban areas are characterized by heterogeneous spatial patterns, which are
often related to distribution of population, their ethnicity, race, and socio-
economic attributes. Urban heterogeneity, however, tends to be associated
with intra-urban inequalities, particularly in developing countries. There,
the ‘‘differential in wealth … and its associated problems are increasingly
Journal of Human Development
Vol. 7, No. 1, March 2006
ISSN 1464-9888 print/ISSN 1469-9516 online/06/010085-25 #2006 United Nations Development Programme
DOI: 10.1080/14649880500502102
more visible and … more intense in urban settlements. A large — or rather
the largest — number of rich and poor people are physically concentrated
in relatively small geographical areas’’ (Werna, 2000, p. 2). Brazilian cities
exemplify this uneven spatial distribution of rich and poor people and the
heterogeneous spatial patterns. Favelas (Brazilian slum areas) and
expensive condominiums are adjacent to each other in many metropolitan
regions in the country. The income variable, which is probably the major
determinant of intra-urban inequalities in Brazilian cities, is also
manifested in the differential provision of and access to public
infrastructure (Lima, 2001).
Previous studies on the relationship between intra-urban inequalities
and public intervention suggest that the success of such interventions
depends on local politics as much as on the approach taken to address
them. For example, Werna concludes that the public sector’s attention to
intra-urban differences in the municipality of Sa˜o Paulo ‘‘has been
extremely variable, entailing an oscillation between governments with a
strong concern with such differentials and others with different primary
interests’’ (1995, p. 134). More generally, projects, grants, and interven-
tions for improving human condition in developing countries focus on
economic growth, assuming that it generates income gains for the poor
and promotes welfare benefits such as access to school and health care.
However, there is accumulating evidence that ‘‘economic growth alone
does not ensure access for all basic needs, and can in turn increase
inequality’’ (Devas et al., 2001, p. 6) in urban and rural communities alike
(Shah and Sah, 2004).
Extending on efforts to overcome the dominance of economic
variables, income in particular, in understanding intra-urban inequalities
and promoting overall development, we subscribe to a broader view of
development based on the human development paradigm. In addition to
income, this paradigm focuses on social, cultural, and political factors, but
others are also proposed. For example, Arimah suggests inclusion of
‘‘public expenditures on education, primary school enrollment, female
educational enrollment, expenditure on health and good governance’’
(2004, p. 399).
In this study we conduct spatial econometric analysis of the intra-
urban inequality among the districts of Sa˜o Paulo municipality (SPM) as
the variation in the Human Development Index (HDI) devised by
the United Nations Development Programme. We explore the relation-
ship of this index with the provision of public services and utilities
and investments in social programs. To achieve these objectives we
use confirmatory spatial data analysis (CSDA) methods, which control for
spatial heterogeneity (e.g. between center and periphery) and for spatial
autocorrelation. Such analytical techniques can be used to evaluate the
spatial distribution of societal resources and effectiveness of existing
programs and to help policy-makers and decision-makers at the local,
regional and national levels address inequalities.
M. A. Haddad and Z. Nedovic
Development and public intervention
Human development–economic growth linkage
Human development is closely interlinked with the overall development
and economic growth in urban areas, and thus requires direct attention
(Ranis et al., 2000). Even when successful, developments aimed at
economic growth only have ‘‘contributed with disturbing regularity to
increase inequality, conflict, unemployment’’ (Alkire, 2002, p. 14).
Conversely, education and health are found to support generation of
economic growth (Anand and Sen, 2000). Regardless of this close
association between the human and economic development, the two
phenomena have been treated and measured independently. The HDI is a
widely accepted measure of human development. It summarizes several
variables into three indicators: education, which is represented by adult
literacy rate and gross enrollment ratio; longevity, which is represented by
life expectancy at birth; and income, which is represented by the Gross
Domestic Product (GDP) per capita (Purchasing Power Parity (PPP) in US
dollars), an established measure of economic strength. Alkire (2002),
however, identifies aspects of development, such as human capital and
health for which GDP is an inadequate proxy. Foster et al. (2005) point to
the deficiencies with regard to sensitivity of the indices to distribution of
human development.
To address this recognized inadequacy of the GDP, Haq (1998)
introduced the human development paradigm, which is defined by four
components: equity, sustainability, productivity, and empowerment. He
suggests four ways of linking economic growth and human development:
emphasis on investment in the education, health, and skills of people;
more equitable distribution of income and assets; well-structured social
expenditures by government; and the empowerment of people. The
author considers equity as crucial for enlarging people’s choices, and
suggests that equitable access to opportunities is necessary for its
achievement. He states that the ‘‘emphasis on investments in the
education, health and skills of the people can enable them to participate
in the growth process as well as to share its benefits, principally through
remunerative employment’’ (Haq, 1998, p. 21).
Sen’s (1999) capability approach is also useful for understanding
urban inequalities and relating the expansion of human capabilities to
public interventions. He defines a person’s capability ‘‘as the alternative
combinations of functionings that are feasible for her to achieve. Capability
is thus a kind of freedom: the substantive freedom to achieve alternative
functioning combinations (or, less formally put, the freedom to achieve
various lifestyles)’’ (Sen, 1999, p. 75). A sample of capabilities we apply to
this study includes school attendance, adequate health care, sufficient
income (without child labor), access to safe water and sanitation,
electricity, safe environment, and nourishment. The achieved functionings
are represented by variables, which allow us to measure of the effects of
Analyzing Intra-urban Inequalities and Public Intervention
public interventions. Student-teacher ratio, proportion of dwelling with
bathroom, proportion of dwellings with public sewage connection, and
home-based health services are examples of functionings.
Assessing public intervention needs and effects
The empirical work on the relationships between the human development
and public interventions to expand human capabilities is scarce. Previous
research relates inequalities to the impacts of globalization processes, the
effects of the implementation of neoliberal economic policies that
advocate a smaller public sector, and other macro-economic initiatives
(Burgess et al., 1997; Musterd and Ostendorf, 1998; Marcuse and Van
Kempen, 2000, 2002; Andersen and Van Kempen, 2001). With more
sensitivity to the local context of Sa˜o Paulo municipality, Sposati (1996,
2000) applies the Social Exclusion/Inclusion Index to identify districts in
need of social assistance. In examining the districts of Sa˜o Paulo
municipality, Torres and Oliveira (2001) find differential access to primary
education and conclude that the absence of land regulation creates
difficulties in constructing infrastructure in the periphery.
Most studies, however, are descriptive and rely primarily on maps,
graphs, or tables to display data and their spatial distribution. These
studies tend to ignore spatial associations, or are too general to assess
public intervention. Instead, they present only the outcomes of various
forces that shape urban space and quality of life. Statistical analyses and
tests that increase the reliability of findings are rarely performed. For
example, Torres and Marques (2004) only map social indicators to display
spatial concentrations of rich and poor people in Sa˜o Paulo metropolitan
area. Departing from this descriptive approach, Ramos’ (2002) exploratory
spatial data analysis demonstrates how the visualization and analyses of
metropolitan level data can be useful in guiding decision-making
processes; Caˆmara’s et al. (2003) CSDA and mapping of Social
Exclusion/Inclusion variables reveals social dynamics at the district level.
However, these authors too do not relate the mapped indices to public
interventions or any other explanatory factor.
Summary of the literature base
Haq’s human development paradigm and Sen’s capability approach form
the theoretical basis of this study. Both approaches track the intra-urban
inequalities to the lack of equitable access to opportunities. In this paper
we demonstrate the intra-urban inequality as the variation in human
development between districts of SPM. If ‘pro-equality’ forces dominated
the SPM urban policy agenda, public intervention would reduce intra-
urban inequalities by targeting the areas with lower human development
levels. Therefore, the assessment of public investments can be used to
examine to what extent the public sector intervention is related to the
process of ‘enabling’ people and reducing intra-urban inequalities.
M. A. Haddad and Z. Nedovic
We find that most empirical assessments of the relationship between
public interventions and their intended consequences are largely devoid
of spatial dimension and statistical analysis. Inclusion of spatial effects in
assessing intra-urban inequalities and application of statistical tests would
allow for more reliable discovery and comprehension of urban processes.
The insights from the spatially enabled statistical analysis might help direct
public intervention and allocation of investments in a more strategic and
effective manner. In the next section we provide a detailed account of our
application of the confirmatory spatial data analysis in studying intra-urban
inequalities among Sa˜o Paulo’s districts and their relationship to the
provision of public infrastructure and social programs.
Research framework and methodology
Research framework
This study is based on three assumptions: there are some basic capabilities
that should be part of everyone’s life (for instance, the capability to attend
elementary school); urban economic growth is necessary but not a
sufficient condition to improve human development level and to reduce
intra-urban inequalities; and in most developing countries the public
sector still plays an important role in providing these basic capabilities,
although other actors also produce and provide them.
The central element of the research framework (Figure 1) is the
human development that is either equally or unequally distributed. It is
influenced by public intervention — the provision of public services and
utilities and investments in social programs. A feedback loop is
FIGURE 1. Research framework.
Analyzing Intra-urban Inequalities and Public Intervention
constructed to allow for adjustments in public policies to the changed
levels of human development.
In this study we ask the following questions: What is the relationship
between provision of public services and utilities and human development
level? Are the investments in social programs allocated in districts that
need them the most?
This study is undertaken with data for SPM, which is part of the Sa˜o Paulo
metropolitan area, the largest in Brazil, and contains 39 municipalities.
SPM has over 10.5 million inhabitants, representing more than 10% of the
national population. In 2000, about 17% of the National GDP and
approximately 21% of the industrial domestic product were generated in
SPM. There is a large presence of the public sector in SPM, with about
150000 public employees directly working for the municipal government,
and about an additional 50 000 employees working for other public sector
entities (Werna, 2000). The municipal government is decentralized into 96
administrative districts, 31 Sub-Prefeituras (sub-administrations), 13
Nu´cleos de Educac¸a˜o (Education Nuclei), and 39 Distritos de Sau´de
(Health Districts).
An analysis of regional inequalities with respect to human develop-
ment at the municipal level in the Brazilian state of Sa˜o Paulo indicates
that SPM is well above other municipalities within the state (Haddad,
2003). That is not surprising given that SPM is the most affluent
municipality in the country. However, we suspect that aggregated
municipal human development data may be masking the intra-urban
variability that occurs even inside the wealthy municipalities. As Portnov
alerts, ‘‘similarly to spatial disparities in regional development, intra-urban
inequalities, once occurring, may become persistent and self-perpetuat-
ing’’ (2002, p. 149). Therefore, a change in the scale of analysis is applied
to assess intra-urban inequalities.
Table 1 compares HDI levels and two measures of inequality — Gini
and Theil indices — for Brazil, Sa˜o Paulo state, SPM, and other
municipalities within Sa˜o Paulo state. The HDI for the SPM is above the
Brazilian average and Sa˜o Paulo state average. Within the state, SPM ranks
above the 75th percentile (0.802). Gini and Theil indices for SPM have
Table 1. The HDI and the Gini and Theil Indices, 2000
HDI 2000 Gini 2000 Theil 2000
Brazil 0.766 0.650 0.760
Sa˜o Paulo State 0.820 0.590 0.610
Sa˜o Paulo Municipality 0.841 0.620 0.680
Minimum within Sa˜o Paulo state 0.645 0.420 0.290
Maximum within Sa˜o Paulo state 0.919 0.730 1.060
Average within Sa˜o Paulo state 0.779 0.530 0.470
M. A. Haddad and Z. Nedovic
values close to the Brazilian average and higher than the Sa˜o Paulo state
average. Within the state, they rank below the 90th percentile (0.590 and
0.580, respectively), showing a high level of inequality.
The unit of analysis for this study is an administrative district with its
intra-urban diversity and human development level, which substitute for
Sen’s focus on interpersonal diversity and individual preferences,
respectively. The HDI for the 96 administrative districts is calculated by
the Secretaria Municipal do Desenvolvimento, Trabalho e Solidariedade
(SMDTS, 2002) using the United Nations Development Programme’s
methodology (Figure 2). The Secretariat’s report states that ‘‘40 percent of
the districts of the richest city in the country present low levels of human
development’’ and that ‘‘the HDI for SPM districts reach values that can be
found in both Europe and Africa at the same time’’ (SMDTS, 2002, p. 12).
Thirty-eight districts have very low HDI (below 0.5), thus ‘‘forming poverty
pockets across the municipality’’ (SMDTS, 2002, p. 4). The HDI in districts
with low human development, located in the east, north, and south, is less
than 90% of the sample average; the HDI in districts with high human
development, located in the center of the municipality, is over 110% of the
FIGURE 2. HDI relative to the sample average of SPM Districts, 2000.
Analyzing Intra-urban Inequalities and Public Intervention
sample average. These facts suggest substantial intra-urban inequalities
and variability of human development within SPM.
Data on the provision of public services and utilities are derived from
three sources: education data from the Censo Escolar 2000, from the
Ministry of Education; longevity data from Pesquisa de Assisteˆncia Me´dico-
Sanita´ria 1999, conducted by the Instituto Brasileiro de Geografia e
Estatı´stica (IBGE, Brazilian Bureau of Statistics); and infrastructure data
(water, sewerage, garbage collection, electric power, and others), from
Censo Demogra´fico 2000, also conducted by IBGE.
Data on investments in social programs come from four federal
sources: Programa Bolsa Alimentac¸a˜o (Nutrition Children Program)
supplements the monthly salary of low-income families with funds to be
used to feed children up to the age of 6 and their mothers during the
pregnancy period and the child’s first year; Programa Merenda Escolar
(School Snack Program) provides nutritional meals for children that attend
municipal schools; Programa Sau´de da Famı´lia (Family Health Program)
brings health professionals to low-income families by providing basic care
in their homes; and Programa Renda Mı´nima (Minimum Wage Program)
targets low-income families with children of school age to furnish an
income supplement and allow the children to attend school instead of
working to help increase their family income (SMDTS, 2003).
Spatial weight matrix
Spatial statistical analysis requires a spatial weight matrix W. This matrix is
a formal expression of spatial adjacency between observed districts. Its
selection affects the results of diagnostic tests (Le Gallo and Ertur, 2003,
p. 180) and we use three matrices to test the robustness of the results. By
examining the empirical work on the types of spatial weight matrices, the
context in which they are used, and their applicability to this study
(Anselin, 1995; Talen and Anselin, 1998; Pereira et al., 1998; Messner et
al., 1999; Baller et al., 2001; Baumont et al., 2003, 2004; Le Gallo and
Ertur, 2003; Guillain et al., 2006), we opted for the simple binary queen
contiguity and two k-nearest-neighbors matrices.
The simple binary queen contiguity matrix has two values: if district i
has a common border and/or vertex with district j, then they are neighbors
and w
51; and if district idoes not have a common border and /or vertex
with district j, then they are not neighbors and w
50. The diagonal
elements are set to 0. The k-nearest-neighbors weight matrix is defined as:
ij kðÞ~0ifi~j
ij k
ðÞ and wij k
ij k
ij k
ij kðÞ~0ifdijwdikðÞ
where d
(k) is a critical cut-off distance defined for each district i;d
M. A. Haddad and Z. Nedovic
the kth order smallest distance between districts iand jsuch that each
district ihas exactly kneighbors. For this study, k55 and k56 are applied.
These values are chosen because they represent the highest frequency in
the distribution of connection between SPM districts, based on the exami-
nation of the simple binary queen contiguity matrix; that is, the majority of
SPM districts have five or six neighbors (22 and 23 districts, respectively).
All matrices, the two k-nearest neighbors and the simple binary queen
contiguity are row standardized so that each row sums up to 1.
Confirmatory spatial data analysis and models
To better understand intra-urban inequalities and their relationship to
human development and public intervention, we introduce spatial
dimension in our analyses. Spatial dependence and spatial heterogeneity
are found to characterize the distribution of HDI across the SPM districts
(Haddad, 2003). Spatial dependence occurs when the values of observa-
tions in spatial units that are close or adjacent are similar or correlated
(Anselin, 1998). This association can be measured by different statistics;
one of them is the Moran’s Istatistic. This statistic gives a formal indication
of the degree of linear association between HDI values and HDI spatially
lagged values, calculated as spatially weighted averages of neighboring
values. For the SPM districts, the Moran’s Istatistic for the HDI, based on
999 permutations, is 0.6569 (p50.001) when using the simple binary
queen contiguity matrix, is 0.6004 (p50.001) when using the six-nearest-
neighbors matrix, and is 0.6293 (p50.001) when using the five-nearest-
neighbors matrix. It indicates a positive spatial autocorrelation in the
distribution of HDI: districts of high (or low) HDI are surrounded by
districts of high (or low) HDI, respectively.
Spatial heterogeneity implies unstable relationships between values of
observations, and detectable spatial regimes. These relationships may be
described by multiplicity of functional forms and parameters that vary
across the data set (Anselin, 1988). In this paper, following Le Gallo and
Dall’Erba (2006), we use the Getis-Ord statistics to measure local spatial
autocorrelation and to detect the presence of spatial heterogeneity among
the SPM districts. As Le Gallo and Ertur suggest, ‘‘these statistics are based
on spatial accumulations and can thus help to deepen the analysis for
detecting spatial clusters around each [district] iwithout being affected by
the value taken by the variable in that [district] i’’ (2003, p. 178). Getis-Ord
statistics are calculated for each district. The spatial distribution of HDI is
not stable across SPM districts and forms a characteristic spatial pattern of
human development displayed as center–periphery spatial regimes: a
cluster of districts with positive Getis-Ord statistics (the center) and a
cluster of districts with negative Getis-Ord statistics (the periphery).
Figure 3 displays these spatial regimes.
We use CSDA methods to examine the development level measured
by the HDI, provision of utilities and services, and administration of social
Analyzing Intra-urban Inequalities and Public Intervention
programs in each district by taking into account its geographic location
relative to other districts in the municipality. The CSDA methods enable
specification of spatially-explicit regression models that incorporate spatial
autocorrelation and spatial heterogeneity. These methods help avoid
misspecifications of models, inefficient coefficients, and erroneous
statistical inferences that happen when spatial dependency and spatial
heterogeneity are not addressed (Anselin and Rey, 1991).
To examine the relationship between the public interventions and
human development levels, we use the following three regression models:
the education model (M-EDU), the social programs model (M-SP), and the
infrastructure model (M-IS). The HDI is the dependent variable in all three
models. Also, to control for districts’ population distributions, all models
FIGURE 3. Spatial regimes — center and periphery — calculated based on the Getis-Ord statistics.
M. A. Haddad and Z. Nedovic
have a density variable calculated by dividing the total population of a
district by its area.
NM-EDU has the following independent variables: density and public
school variables — number of students per classroom, number of
students per computer, level of education of teachers (e.g. teachers with
bachelor’s degree), and proportion of students per teacher.
NM-SP has the following independent variables: density and ratios
between the population that received the program benefits and the
target population of the four social programs — Nutrition Children,
School Snack, Family Health, and Minimum Wage.
NM-IS has the following independent variables: density, infrastructure
development, and health development — the two latter variables are
derived from 12 variables by a factor analysis.
We take the following steps in specifying the models, addressing Florax’s
et al. (2003) ‘‘specific to general’’ model specification approach. Because
the steps are the same for all models, we illustrate the process using the M-
EDU model with only two independent variables. Step 1 is a simple non-
spatial regression model estimated by:
HDIi~b0zb1x1izb2x2izeii~1, . . . ,96 and e*N0,s2
where HDI
is the Human Development Index for each district I,x
and x
are independent variables representing the number of students per
computer and the number of students per classroom, and b
, and b
are the unknown parameters to be estimated; eis the vector of errors.
In Step 1 we carry out tests to detect the presence of spatial
dependence, which lead us to further apply the spatial error model or the
spatial lag model. For all three models, the tests reveal that the spatial
lag model is the most appropriate. Step 2 is also an ordinary least
squares (OLS)-based regression model that controls for spatial hetero-
geneity. Therefore the spatial regimes — center and periphery,
identified by calculating the Getis-Ords statistics — are introduced in the
i~1, . . . ,96 and e*N0,s2
 ð3Þ
where all other elements are defined as previously, but divided in two
groups; that is, for each element, there is one dummy variable for the
‘center’ D
and one for the ‘periphery’ D
In Step 3 the spatial lag model is applied to include the spatial
dependence in the model specification, as an additional covariate in
the model, the so-called spatial lag. The coefficients of the spatial lag
model are estimated by the Maximum Likelihood (ML) method and the
Instrumental Variables (IV) method (the latter only if the assumption
of the normality of errors is rejected based on the results of
Jarque-Bera test).
For M-EDU, the spatial lag model in matrix form is
Analyzing Intra-urban Inequalities and Public Intervention
written as:
zrWHDI Nx1ðÞ
where HDI is a vector of dimension n596 of the Human Development
Index for each district i,x
and x
are vectors containing values
for independent variables (e.g. students per computer and students per
classroom in 2000), Sis the sum vector, b
, and rare the unknown
parameters to be estimated, and ris the spatial autoregressive parameter
that indicates the extent of interaction between the observations according
to the spatial pattern exogenously introduced by means of the
standardized weight matrix W. The spatial lag variable WHDI
the HDI multiplied by the weight matrix: for a district iof vector HDI
and the corresponding line of the spatial lag vector contains the spatially
weighted average of HDI. uis the vector of errors with the usual
Finally, in Step 4 the spatial lag model is enhanced by including the
center-periphery spatial regimes to the equation:
 ð5Þ
where all the elements are defined as previously, but divided into two
groups; that is, for each element there is one dummy variable for the
‘center’ D
and one for the ‘periphery’ D
If after controlling for the spatial regimes there is a remaining
problem of heteroskedasticity, the model has to be estimated using
groupwise heteroskedasticity. Groupwise heteroskedasticity means that
the variance across error terms in the center regime is different from the
one across error terms in the periphery regime.
In this section we present the results of model estimations based on the
six-nearest-neighbors weight matrix. Reporting the models estimates
based on the simple binary queen contiguity and the five-nearest-
neighbors weight matrices would be redundant as they led to the same
results. This fact points to the robustness of our results with regard to the
choice of the spatial weight matrix. It is important to emphasize that for all
three models the Akaike Information Criterion and the Schwartz Criterion
show that the spatial models perform better than the non-spatial models.
For all models, the tables display the results of the OLS (non-spatial)
model and the more appropriate spatial specification.
Finally, because
there is no time lag between data for the dependent variable and the
independent variables, the models do not measure the effects of the
selected public services, utilities and social programs on human develop-
ment, but only verify if they are located where the needs are.
M. A. Haddad and Z. Nedovic
Regression results for the education model (M-EDU)
As displayed in Table 2, the OLS model coefficients for ‘students per
classroom,’ ‘students per computer,’ ‘education teachers’ and ‘students
per teacher’ are all significant, indicating that a high HDI coincides with a
lower number of students per classroom and students per computer and
with a higher level of teachers’ education and higher number of students
per teacher. The model’s R
-adjusted value is close to 30% but, because of
the presence of spatial autocorrelation, the OLS estimators are inefficient
and inconsistent. The Jarque-Bera test rejects the assumption of normality
of errors, and therefore, when specifying the spatial lag model, the IV
method is used in addition to ML to estimate the lag model and confirm
that the ML-based estimates are reliable. The White test and the Breusch–
Pagan test reject homoskedasticity. We presume that the heteroskedasticity
is possibly associated with structural instability across regimes, and
address this by estimating a spatial regime lag model specified with
dummies for center and periphery.
Table 2. Estimation results for the education model for Sa˜o Paulo municipality Districts, 2000
Estimation Non-spatial OLS Spatial (ML)
Center Periphery
(constant) 0.389 0.19 0.225
(0.001) (0.102) (0.113)
(density) 20.0000006 20.000005 0.0000008
(0.757) (0.062) (0.436)
(students/classroom) 20.003 20.0004 20.0003
(0.000) (0.591) (0.479)
(students/computer) 20.0002 0.0002 20.0003
(0.084) (0.121) (0.000)
(education teachers) 0.283 0.197 0.078
(0.000) (0.004) (0.400)
(students/teacher) 0.01 0.001 0.001
(0.020) (0.783) (0.598)
-adjusted 0.299
Spatial lag 0.467
Akaike information criterion 2156.892 2250.709
Schwarz information criterion 2141.506 2217.373
Jarque–Bera estimated residuals
normality test
18.607 –
Spatially-adjusted Breusch–
Pagan test for
28.320 –
Likelihood Ratio test 20.976
White test for heteroskedasticity 58.775
Note:pvalues presented in parentheses.
Analyzing Intra-urban Inequalities and Public Intervention
Further testing of the spatial regime lag model with the Lagrange
Multiplier Test on Spatial Error Dependence indicates the presence of
omitted spatial autocorrelation. Also, the spatially-adjusted Breusch–Pagan
heteroskedasticity test versus the dummy center is significant, suggesting
that there is remaining heteroskedasticity in the specification. Finally, the
Chow–Wald test strongly rejects the joint null hypothesis of structural
stability, requiring an estimate of a spatial regime lag model with
groupwise heteroskedasticity. The Likelihood Ratio and the Chow–Wald
tests reveal a significant presence of groupwise heteroskedasticity that is
also associated with structural instability across regimes.
The spatial regime lag with groupwise heteroskedasticity model (ML)
takes into account the fact that public intervention could be different
within each of the two regimes. The spatial lag coefficient is positive and
significant, revealing that there is spatial autocorrelation between HDI
levels in neighboring districts, which is not captured by the covariates used
in the OLS-based model. At the periphery regime the only significant (and
negative) coefficient is for ‘students per computer,’ showing that lower
numbers of students per computer are associated with higher HDI, as
expected. At the central regime, two coefficients are significant —
‘education teachers’ (positive) and ‘density’ (negative) — suggesting that
higher HDI is associated with higher level of teachers’ education and with
lower density districts, respectively.
Table 3 shows that the Chow–Wald test strongly rejects the joint null
hypothesis of structural stability for the spatial regime lag with groupwise
heteroskedasticity model (ML). When examining the tests of stability of the
individual coefficients, ‘density’ and ‘students per computer’ coefficients
turn significantly different across regimes. These results indicate that the
correct model for M-EDU is a spatial lag with heterogeneity taking the form
of spatial regimes and groupwise heteroskedasticity. In other words,
Table 3. Chow–Wald tests for overall stability and individual stability, education model
Spatial (ML)
Chow–Wald overall stability 37.947
(constant) 0.042
(density) 4.074
(students/classroom) 0.018
(students/computer) 10.91
(education teachers) 1.04
(students/teacher) 0.01
Note:pvalues presented in parentheses.
M. A. Haddad and Z. Nedovic
public intervention concerning education (when controlling for ‘density’)
in central districts is significantly different from the peripheral districts. In
examining the significance and signs of the four public intervention
variables in the center and in the periphery, we cannot affirm that the
public sector intervention is driven by ‘pro-equality’ forces. The only
significant coefficient for the ‘students per computer’ variable has a
negative sign and points to a possible bias in allocation of school-related
funds toward districts with higher HDI.
Regression results for the social programs model (M-SP)
In interpreting the signs of the M-SP coefficients, we assume that they are
positive if the programs are implemented in districts that have a high HDI
(i.e. where they are not needed); the coefficients are negative if the
programs are implemented in districts that have a low HDI (i.e. where they
are needed). The model estimation by OLS has an R
-adjusted value of just
over 46% and results in two significant coefficients — ‘minimum wage’
with the negative sign and ‘nutrition children’ with a positive sign
(Table 4). We specify the spatial lag model using both ML and the IV
methods. However, the Breusch–Pagan test detects heteroskedasticity,
which we consider possibly associated with structural instability across
regimes. To address that, we estimate the spatial regime lag model,
including the dummies in the specification — one for center and one for
Further tests do not indicate the presence of omitted spatial
autocorrelation, but do suggest heteroskedasticity (spatially-adjusted
Breusch–Pagan against the dummy center is significant) and structural
instability (Chow–Wald is significant)(Table 5). Based on the results of the
spatially-adjusted Breusch–Pagan and Chow–Wald tests for the spatial
regime lag model, there is a need to estimate a spatial regime lag with
groupwise heteroskedasticity model. The model results, based on the
Likelihood Ratio test and on the Chow–Wald test, reveal the significant
presence of groupwise heteroskedasticity and structural instability.
Indeed, heteroskedasticity is associated with structural instability across
the center–periphery regimes.
The spatial regime lag model with groupwise heteroskedasticity (ML)
has several significant coefficients. The spatial lag coefficient is positive
and significant. Two coefficients are significant at the periphery regime:
‘school snack’ is negative and significant (i.e. allocated in districts with low
HDI), and ‘family health’ is positive and significant (i.e. allocated in
districts with high HDI). One coefficient is significant in the center regime:
the ‘minimum wage’ coefficient is negative and significant, revealing that
this program is being allocated in districts of low HDI. This program,
however, is not significant in the periphery where the overall level of
human development is lower than in the center and more resources for
the ‘minimum wage’ program are needed. Finally, the tests of stability of
Analyzing Intra-urban Inequalities and Public Intervention
Table 4. Estimation results for the social programs model for Sa˜o Paulo municipality districts, 2000
Estimation Non-spatial OLS Spatial (ML)
Center Periphery
(constant) 0.582 0.184 0.148
(0.000) (0.003) (0.002)
(density) 20.000002 20.000003 0.0000004
(0.211) (0.104) (0.700)
(minimum wage) 20.228 20.213 20.043
(0.000) (0.000) (0.112)
(school snack) 20.057 0.111 20.135
(0.412) (0.294) (0.000)
(family health) 20.0000005 0.000008 0.000006
(0.919) (0.353) (0.014)
(nutrition children) 0.058 0.018 0.027
(0.000) (0.129) (0.294)
-adjusted 0.462 –
Spatial lag 0.68
Akaike information criterion 2182.372 2272.603
Schwarz information criterion 2166.986 2239.266
Jarque–Bera estimated residuals
normality test
10.290 –
Spatially-adjusted Breusch–Pagan test for
15.667 –
Likelihood Ratio test 13.513
White test for heteroskedasticity 35.216
Note:pvalues presented in parentheses.
Table 5. Chow–Wald tests for overall stability and individual stability, social programs model
Spatial (ML)
Chow–Wald overall stability 20.06
(constant) 0.373
(density) 2.566
(minimum wage) 7.825
(school snack) 4.778
(family health) 0.065
(nutrition children) 0.109
Note:pvalues presented in parentheses.
M. A. Haddad and Z. Nedovic
individual coefficients point to the ‘minimum wage’ and ‘school snack’
coefficients as significantly different across the center-periphery regimes.
Therefore, the correct model for M-SP too is a spatial lag with
heterogeneity model taking the form of spatial regimes and groupwise
heteroskedasticity. The model shows that the implementation of social
programs (when controlling for ‘density’) among central districts is
significantly different from the implementation among peripheral districts.
When examining the social programs variables in the center and in the
periphery, we find evidence of the public sector’s ‘pro-equality’ concerns
in distributing program recourses among SPM districts only in the ‘school
snack’ program. There is some evidence of targeting the ‘minimum wage’
program toward districts with low HDI, but only within central districts.
Overall, except for the ‘school snack,’ social programs are not implemen-
ted where the needs are. If this were the case (i.e. if the programs are well
targeted toward the districts in need), all of the social programs
coefficients would be significant and negative in the periphery regime.
This may suggest that in SPM, as a provider of social programs, the public
sector’s efforts are insufficient and mis-targeted to enhance human
development levels and reduce intra-urban inequalities.
Regression results for the infrastructure model (M-IS)
The OLS coefficient estimates for ‘infrastructure development’ and ‘health
development’ are significant and have positive signs, indicating that the
HDI is higher in districts with higher levels of infrastructure and health
development (Table 6). The model fit is good with R
-adjusted values close
to 60%, but with inefficient and inconsistent OLS estimators due to the
presence of spatial autocorrelation. As in the previous two models, after
estimating the spatial lag and performing further diagnostic tests, we
detect heteroskedasticity that we suspect is due to the center–periphery
instability, and estimate the spatial regime model by including the
dummies for center and periphery in the specification.
For this model the tests also point to the presence of omitted spatial
autocorrelation and structural instability, and a spatial regime lag with
groupwise heteroskedasticity model has to be estimated (ML). The
Likelihood Ratio test for this model estimation points out the significant
presence of groupwise heteroskedasticity. The spatial lag coefficient is
positive and significant. The public intervention variables — ‘infrastruc-
ture development’ and ‘health development’ — are positive and significant
at both the center and periphery. This reveals that the better the provision
of physical infrastructure and health services is associated with higher
overall level of human development across all the districts in the center
and in the periphery. Population density is significant and inversely related
to the HDI only in the periphery, indicating that the HDI in lower-density
districts is on average higher than in more dense districts.
Analyzing Intra-urban Inequalities and Public Intervention
From Table 7 one can observe that the Chow–Wald test does not
reject the joint null hypothesis of structural stability, and the tests of the
individual coefficient stability are not significantly different across the
center–periphery regimes. In other words, the ‘infrastructure develop-
ment’ and ‘health development’ (when controlling for ‘density’) among
central districts are not significantly different from those among peripheral
Table 6. Estimation results for the infrastructure model for Sa˜o Paulo municipality districts, 2000
Estimation Non-spatial OLS Spatial (ML)
Center Periphery
(constant) 0.6 0.315 0.307
(0.000) (0.000) (0.000)
(density) 20.000004 20.000003 20.000002
(0.005) (0.128) (0.079)
(infrastructure development) 0.054 0.111 0.036
(0.000) (0.071) (0.000)
(health development) 0.079 0.049 0.071
(0.000) (0.000) (0.000)
-adjusted 0.598 –
Spatial lag 0.467
Akaike information criterion 2212.153 2279.270
Schwarz information criterion 2201.895 2256.191
Jarque–Bera estimated residuals
normality test
27.226 –
Spatially-adjusted Breusch–Pagan
test for heteroskedasticity
16.492 –
Likelihood Ratio test 23.572
White test for heteroskedasticity 22.468
Note:pvalues presented in parentheses.
Table 7. Chow–Wald tests for overall stability and individual stability, infrastructure model
Spatial (ML)
Chow–Wald overall stability 5.878
(constant) 0.05
(density) 0.311
(infrastructure development) 2.511
(health development) 0.949
Note:pvalues presented in parentheses.
M. A. Haddad and Z. Nedovic
districts. These results suggest that the public sector serves the whole
municipality with infrastructure and health services without discriminating
between center and periphery, and equally distributing the investments
across them. However, we would expect that, given the inequalities in the
level of human development, more resources should be invested in the
periphery and in districts with low HDI.
Summary remarks
Two out of the three models for SPM presented in this section —
education, and social programs — show that the public interventions are
different across the center–periphery regimes. The modeling results also
show that, except for ‘school snack,’ the valuable public resources set
aside for education, social programs, and infrastructure are not always
targeted toward the population in the most needy districts.
Predicted value maps for the M-EDU, M-SP and M-IS (Figure 4) show
the predicted values for HDI-EDU, HDI-SP, and HDI-IS based on the
independent variables for public intervention (while controlling for
‘density’). These predicted values indicate the distribution and intensity
of the outcomes of the public attention to specific conditions across the
SPM districts. Higher predicted values indicate more intensive public
intervention. Assuming that the public intervention is intended to expand
people’s capabilities and create more equitable access to opportunities,
we expect that the higher values for the predicted HDI would be in the
districts located in the periphery. The three maps in Figure 4 display a
common trend — higher values in the central districts, and lower values in
the southern and eastern districts — perpetuating the existing disparities
in human development among SPM districts. However, a positive trend
towards the districts surrounding the central area can be noticed more
strongly in the HDI-IS, and moderately in the HDI-EDU.
In this paper we examine the relationship between intra-urban inequalities
in the level of human development and public intervention in the SPM. We
apply CSDA that accounts for the location of the 96 municipal districts and
dependency among the neighboring districts. The empirical evidence from
the CSDA leads to a few conclusions about the public intervention in SPM.
The valuable public investments in education and social programs are
allocated differentially among the central and peripheral districts, with
central districts on average benefiting more from those investments.
Investments in infrastructure and health services are allocated without
distinction across the whole municipality, but more prominently in
districts with higher HDI. Knowing that the needs vary across the SPM
territory, this finding indicates that the funds for these services are unlikely
Analyzing Intra-urban Inequalities and Public Intervention
to be allocated in districts where they are needed. Mapping of the
predicted HDI values for all three models, in general, demonstrate that the
manner in which public interventions are distributed (when controlling
FIGURE 4. HDI predicted maps based on the M-EDU, M-SP, and M-IS models.
M. A. Haddad and Z. Nedovic
for population ‘density’) is unlikely to address the observed inequalities
since they largely still follow the center–periphery regimes.
When examining the difference between center and periphery, we
find based on the M-EDU model that, on average, the central districts with
higher HDI have teachers with higher education, and that in the periphery
the districts with higher HDI values have smaller ratio of students per
computer. The M-SP model suggests that the ‘minimum wage’ program is
allocated in the center, in districts of lower HDI and hence higher need.
However, it is important to highlight that this program is not significant in
peripheral districts, where it is also needed. The ‘school snack’ program is
correctly allocated in the periphery, in districts of higher needs (i.e. lower
human development level). The M-IS model indicates that ‘infrastructure
development’ and ‘health development’ have a positive association with
the HDI in both the center and the periphery, showing that the periphery
is not neglected.
To examine the idea of expansion of capabilities and its link with
public intervention we consider only the peripheral districts. This idea is
manifested with regard to the ‘school snack’ variable, which is the only one
displaying a desirable direction, with the number of program recipients
increasing as the HDI gets lower. Other social programs and provisions of
public services and facilities do not contribute significantly to expanding
the capabilities of those affected by the intra-urban inequalities. The
models suggest that already affluent and well-developed central districts
perhaps benefit more from the public investments than the peripheral
districts where the needs are more prominent.
We offer a few possible explanations of the discovered reinforcement
of the center–periphery regimes by public intervention. First, in SPM it is
not unusual to see ‘wealthy people influencing government officials to
attract public investments and services to the neighborhoods they live in,
at the expense of low-income neighborhoods’’ (Werna, 2000, p. 4). These
wealthy people tend to live in the center of the municipality. Second, the
inequalities in human development are also related to the ‘‘processes of
political representation, [and] action rationale of the bureaucratic
segments responsible for service provision’’ (Torres and Oliveira, 2001,
p. 16). Most of the time, residents of the less developed districts do not
have enough education and power to participate in political processes.
Third, the southern part of the municipality is characterized by a high level
of urban violence and homicides that are ‘‘concentrated in the periphery
of the Municipality of Sa˜o Paulo, spilling over the borders to neighboring
municipalities of the Metro area’’ (Cardia, 2000, p. 11). This condition may
present an obstacle to public intervention. Fourth, the geography of the
municipality probably contributes to this condition by partially isolating
the southern districts with a body of water. This physical barrier may affect
the decisions regarding the geographical allocation of public investments.
Fifth, three of the four examined social programs are universal (i.e. they
serve all the districts). Only one program (Minimum Wage) is based on
Analyzing Intra-urban Inequalities and Public Intervention
allocation to the districts in need. However, according to the M-SP model
results, the peripheral districts do not benefit from it. Finally, the Strategic
Master Plan of the Sa˜o Paulo Municipality 2002–2012 (SEMPLA, 2004)
points out that the peripheral districts had the highest population growth
in the past decade. This plan also designates these districts as zones for
environmental protection (macrozona de protec¸a˜o ambiental). Clearly,
there are multiple and conflicting forces at work in these districts —
population growth pressures, the demands for environmental protection,
and an already existing low level of human development.
We conclude that policies and programs which explicitly target
less developed areas are more likely to be more effective and successful
in reducing intra-urban inequalities. If the public sector is willing to
achieve ‘pro-equality’ goals, attention to spatially-oriented social policies is
needed. If the allocation of public resources is decided by power politics
and ability to participate in the decision-making processes, then the less
educated and poor will continue to be on the shorter end of the
distribution. The inequalities identified in this paper are probably due
to these kind of issues, although without time-series data this study
could not determine the causes of specific outcomes of public intervention
and its correlates. To evaluate the effectiveness of public policies and
programs and to identify the factors that are causally linked to their
outcomes, future studies should rely on data collected at multiple periods
of time.
The authors are grateful for the comments and suggestions provided by
three anonymous referees. They would also like to thank Julie Le Gallo for
her comments and suggestions on an earlier version of this paper, the
Brazilian institutions for their help with data acquisition and, last but not
least, Geoffrey Hewings and Sandy Dall’Erba for their comments and
suggestions along the way. The support by the Hewlett Program of the
University of Illinois is greatly appreciated. However, the authors remain
solely responsible for any opinions, errors or misconceptions.
1 For Children Nutrition, one group is targeted: children from 0 to 6 years old in families
with per-capita income lower than US$ 30.00 per month. For School Snack the target
population was not provided by the program coordinators, so we assumed that it
includes children from 4 to 14 years old in families with an income of three minimum
wages or less (around US$ 240.00). For Family Health we assumed that the target
population is families with an income of three minimum wages or less. For Minimum
Wage the target population is children from 0 to 15 years old in families with an income
of three minimum wages or less (Pochmann, 2002, p. 100).
2 A list of all the variables used in the factor analysis and the factors they belong to is as
M. A. Haddad and Z. Nedovic
3 For all three models, there was a need to estimate using both methods, ML and IV. For
all three models, the results for the IV models confirm the signs of the coefficients and
their significance are consistent with the results of the ML models.
4 According to Anselin, ‘‘a spatial lag is constructed as a weighted average (using the
weights in the spatial weighted matrix) of the values observed for the neighbors of a
given location’’ (1998, p. 260). This type of model suggests a possible diffusion process:
events in one place predict an increased likelihood of similar events in neighboring
places. In this model, spatial dependence is best described with reference to the
influence of the dependent variable in neighboring locations. If this form of spatial
autocorrelation is ignored, the results are similar to the consequences of omitting a
significant explanatory variable in the regression model.
5 The three models do not present problems of multicolinearity, having the following
multicolinearity condition numbers: 36.86 (M-EDU); 10.43 (M-SP); and 4.68 (M-IS).
Concerning endogeneity, Hausman tests were performed and only two variables are
endogenous at 1%: ‘school snack’ and ‘children nutrition.’ The other variables are all
Alkire, S. (2002) Valuing Freedoms: Sen’s Capability Approach and Poverty Reduction,
Oxford University Press, Oxford.
Anand, S. and Sen, A. (2000) ‘Human development and economic sustainability’, World
Development, 28(12), pp. 2029–2049.
Andersen, T. and Kempen, R. (2001) Governing European Cities: Social Fragmentation,
Social Exclusion and Urban Governance, Ashgate, Aldershot.
Anselin, L. (1988) Spatial Econometrics: Methods and Models, Kluwer Academic
Publishers, Dordrecht.
Anselin, L. (1995) ‘Local indicators of spatial association — LISA’, Geographical Analysis,
27, pp. 93–115.
Anselin, L. (1998) ‘Interactive techniques and exploratory spatial data analysis’, in P.A.
Longley, M.F. Goodchild, D.J. Maguire and D.W. Wind (Eds.), Geographical Information
Systems: Principles, Techniques, Management and Applications, Wiley, New York.
Anselin, L. and Rey, S. (1991) ‘Properties of tests for spatial dependence in linear regression
models’, Geographical Analysis, 23(2), pp. 112–131.
Arimah, B. (2004) ‘Poverty reduction and human development in Africa’, Journal of
Human Development, 5, pp. 399–415.
Proportion of dwellings with water supply from the public network Infrast.
Proportion of dwellings with water supply from spring or well Infrast.
Proportion of dwellings with sewage services from the public network Infrast.
Proportion of dwellings with sewage services using concrete cesspit Infrast.
Proportion of dwellings with garbage collection from the public network Infrast.
Proportion of dwellings with garbage collection by collective tanks Infrast.
Proportion of dwellings with bathroom (toilet) Infrast.
Proportion of dwellings with eletric power and TV Infrast.
Nutrition Program for children between 0 and 6 years of age Health
Nutrition Program for pregnant women between 15 and 49 years of age Health
Snack Program for children between 4 and 14 years of age Health
Medical equipments/district population ratio (public and private) Health
Analyzing Intra-urban Inequalities and Public Intervention
Baller, R., Anselin, L., Messner, S., Deane, G. and Hawkins, D. (2001) ‘Structural covariates
of U.S. county homicide rates: incorporating spatial effects’, Criminology, 39(3),
pp. 561–588.
Baumont, C., Ertur, C. and Le Gallo, J. (2003) ‘Spatial convergence clubs and the European
regional growth process, 1980–1995’, in B. Fingleton (Ed.), European Regional Growth,
Springer, Berlin.
Baumont, C., Ertur, C. and Le Gallo, J. (2004) ‘Spatial analysis of employment and
population density: the case of the agglomeration of Dijon, 1999’, Geographical
Analysis, 36, pp. 146–176.
Burgess, R., Carmona, M. and Kolstee, T. (1997) The Challenge of Sustainable
Cities: Neoliberalism and Urban Strategies in Developing Countries, Zed Books,
Caˆmara, G., Monteiro, A., Ramos, F., Sposati, A. and Koga, D. (2003) ‘Mapping social
exclusion/inclusion in developing countries: social dynamics of Sa˜o Paulo in the 1990s’,
in M.F. Goodchild and D.G. Janelle (Eds.), Spatially Integrated Social Science: Examples
in Best Practice, Oxford University Press, Oxford.
Cardia, N. (2000) ‘Urban violence in Sa˜o Paulo’, Comparative Urban Studies Occasional
Papers Series, Woodrow Wilson International Center for Scholars, Washington, DC.
Devas, N., Amis, P., Beall, J., Grant, U., Mitlin, D., Rakodi, C. and Satterthwaite, D. (2001)
Urban Governance and Poverty: Lessons from a Study of Ten Cases in the South, The
School of Public Policy, University of Birmingham, Birmingham.
Florax, R., Folmer, H. and Rey, S. (2003) ‘Specification searches in spatial econometrics: the
relevance of Hendry’s methodology’, Regional Science and Urban Economics, 33,
pp. 557–579.
Foster, J., Lopez-calva, L. and Szekely, M. (2005) ‘Measuring the distribution of human
development: methodology and an application to Mexico’, Journal of Human
Development, 6, pp. 5–25.
Guillain, R., Le Gallo, J. and Boiteux-Orain, C. (2006) ‘Changes in spatial and sectoral
patterns of employment in Ile-de-France, 1978–1997’, Urban Studies, (forthcoming).
Haddad, M. (2003) ‘Human development and regional inequalities: spatial analysis across
Brazilian municipalities’, PhD Dissertation, University of Illinois at Urbana-Champaign,
Urbana, IL.
Haq, M. (1998) Reflections on Human Development, Oxford University Press, Calcutta,
Le Gallo, J. and Dall’Erba, S. (2006) ‘Evaluating the temporal and spatial heterogeneity of
the European convergence process, 1980–1999’, Journal of Regional Science, (forth-
Le Gallo, J. and Ertur, C. (2003) ‘Exploratory spatial data analysis of the distribution of
regional per capita GDP in Europe, 1980–1995’, Papers in Regional Sciences, 82,
pp. 175–201.
Lima, J.J. (2001) ‘Socio-spatial segregation and urban form: Bele´m at the end of the 1990s’,
Geoforum, 32, pp. 493–507.
Marcuse, P. and Van Kempen, R. (2000) ‘Introduction’, in P. Marcuse and R. van Kempen
(Eds.), Globalizing Cities: A New Spatial Order?, Blackwell Publishers, Oxford.
Marcuse, P. and Van Kempen, R. (2002) ‘States, cities and the partitioning of urban space:
conclusions’, in P. Marcuse and R. van Kempen (Eds.), Of States and Cities: The
Partitioning of Urban Space, Oxford University Press, New York.
Messner, S., Anselin, L., Baller, R.D., Hawkins, D.F., Deane, G. and Tolnay, S.E. (1999) ‘The
spatial patterning of county homicide rates: an application of exploratory spatial data
analysis’, Journal of Quantitative Criminology, 15(4), pp. 423–450.
Musterd, S. and Osterndorf, W. (1998) Urban Segregation and the Welfare State:
Inequality and Exclusion in Western Cities, Routledge, London.
Pereira, J.M.C., Carreiras, J.M.B. and Vasconcelos, M.J.P. (1998) ‘Exploratory data analysis
of the spatial distribution of wildfires in Portugal, 1980–1989’, Geographical Systems,5,
pp. 355–390.
M. A. Haddad and Z. Nedovic
Pochmann, M. (2002) Desenvolvimento, Trabalho, e Solidariedade: Novos Caminhos
Para a Inclusa˜o Social, Cortez Editora, Sa˜o Paulo.
Portnov, B. (2002) ‘Intra-urban inequalities and planning strategies: a case study of Be’er
Sheva, Israel’, International Planning Studies, 7(2), pp. 137–156.
Ranis, G., Stewart, F. and Ramirez, A. (2000) ‘Economic growth and human development’,
World Development, 28(2), pp. 197–219.
Ramos, R. (2002) ‘Ana´lise espacial de estruturas intra-urbanas: o caso de Sa˜o Paulo’,
Masters Thesis, INPE, Sa˜o Jose´ dos Campos.
SEMPLA, (2004) Plano Diretor Estrate´gico do Municı´pio de Sa˜o Paulo 2002–2012, Senac
Sa˜o Paulo Editora, Prefeitura de Sa˜o Paulo, Sa˜o Paulo.
SMDTS, (2002) Desigualdade em Sa˜o Paulo: o IDH, Prefeitura da Cidade de Sa˜o Paulo,
Sa˜o Paulo.
SMDTS, (2003) Programas Sociais e Restric¸o˜es Macroeconoˆmicas, Prefeitura da Cidade de
Sa˜o Paulo, Sa˜o Paulo.
Sen, A. (1999) Development as Freedom, Anchor Books, New York.
Shah, A. and Sah, D.C. (2004) ‘Poverty among tribals in South West Madhya Pradesh: has
anything changed over time?’, Journal of Human Development, 5, pp. 249–263.
Sposati, A. (1996) Mapa da Exclusa˜o/Inclusa˜o Social da Cidade de Sa˜o Paulo, EDUC, Sa˜o
Sposati, A. (2000) Mapa da Exclusa˜o/Inclusa˜o Social/2000: Dinaˆmica Social dos Anos 90,
Catholic University, Sa˜o Paulo.
Talen, E. and Anselin, L. (1998) ‘Assessing spatial equity: an evaluation of measures of
accessibility to public playgrounds’, Environment and Planning A, 30, pp. 595–613.
Torres, H. and Marques, E. (2004) ‘Polı´ticas sociais e Territorio: Uma abordagem
Metropolitana’, Revista Sa˜o Paulo em Perspectiva, 18(4), pp. 28–38.
Torres, H. and Oliveira, G.C. (2001) Primary Education and Residential Segregation in
the Municipality of Sa˜o Paulo: a Study Using Geographic Information Systems,
conference paper, Lincoln Institute of Land Policy, Cambridge, MA.
Werna, E. (1995) ‘Urban management and intra-urban differentials in Sa˜o Paulo’, Habitat
International, 19(1), pp. 123–138.
Werna, E. (2000) Combating Urban Inequalities: Challenges for Managing Cities in the
Developing World, Edward Elgar, Cheltenham.
Analyzing Intra-urban Inequalities and Public Intervention
... To address these questions, this paper conducts an analysis of the Twitter activity related to rainfall in the city of São Paulo, Brazil, where there is an urban environment with a high degree of intra-urban inequality between different neighborhoods (Haddad and NedovićBudić 2006). We derive time series datasets from Twitter and compare them with rainfall radar data. ...
... Among these factors, population size and income have been found to be the dominant driving forces. In particular, these variables have been found to be the major determinants of intra-urban inequality in Brazilian cities such as São Paulo (Haddad and NedovićBudić 2006). Nonetheless, as explained in Section 1, we expect that if full account is taken of how people move around the city, this reference model that includes only geodemographic characteristics of the residential population can be improved. ...
Full-text available
Although it is acknowledged that urban inequalities can lead to biases in the production of social media data, there is a lack of studies which make an assessment of the effects of intra-urban movements in real-world urban analytics applications, based on social media. This study investigates the spatial heterogeneity of social media with regard to the regular intra-urban movements of residents by means of a case study of rainfall-related Twitter activity in São Paulo, Brazil. We apply a spatial autoregressive model that uses population and income as covariates and intra-urban mobility flows as spatial weights to explain the spatial distribution of the social response to rainfall events in Twitter vis-à-vis rainfall radar data. Results show high spatial heterogeneity in the response of social media to rainfall events, which is linked to intra-urban inequalities. Our model performance (R2 = 0.80) provides evidence that urban mobility flows and socio-economic indicators are significant factors to explain the spatial heterogeneity of thematic spatiotemporal patterns extracted from social media. Therefore, urban analytics research and practice should consider not only the influence of socio-economic profile of neighborhoods but also the spatial interaction introduced by intra-urban mobility flows to account for spatial heterogeneity when using social media data.
... In Latin American countries, good-quality infrastructure is unaffordable for lowerincome families (Calder on & Serv en, 2004). The cheapest places to live are usually peripheral areas, typically poor, lacking basic infrastructure, and far from job opportunities (Haddad & Nedovi c-Budi c, 2006). Although some authors have argued that housing expenditures are offset by lower transportation spending and employment accessibility (Leinberger & Rodriguez, 2016), these effects are less evident for individuals who rely on public transportation modes like those part of this study. ...
Walkable neighborhoods are known for bringing social and economic benefits to their residents. One of these benefits is the real estate premium associated with the neighborhood's walkability, which has been explored in studies worldwide. Here, we extend the available evidence by proposing the evaluation of the walkability premium in a new context. We use multigroup structural equation models to evaluate the direct and indirect effects on low-income neighborhoods in two Brazilian metropolitan cities, São Paulo and Rio de Janeiro. The walkability perception is modeled, mediating the impact of the built urban environment elements on real estate pricing. The results confirm that we can expect a premium for walkability investments, even in poor neighborhoods under non-ideal conditions for walking, and this result is invariant between cities. The effects of the built environment on the real estate price are mediated by latent perceptions, showing the importance of incorporating subjective measures when assessing the walkability premium. Additionally, Latin American cities' socioeconomic and environmental conditions provide new interpretations of the studied phenomenon, highlighting the importance of considering the local context.
... The RMSP is the biggest metropolitan region in Latin America by population, with more than 21 million inhabitants and comprises 39 cities. Its spatial characteristics, distribution of infrastructures and public resources (Haddad & Nedović-Budić, 2006), as well as the socioeconomic characteristics of its population, show a clear dichotomy between the center and the periphery. The urban inequalities that exist in São Paulo also affect how citizens commute. ...
Full-text available
Global South cities are vastly underrepresented in the literature that analyzes the relationships between location choice, land-use patterns and travel behavior. This paper aims to reduce that underrepresentation by bringing new evidence from a metropolitan region in the Global South. We estimate a Structural Equation Model to study the relationships between land-use patterns, location choice, car ownership and travel behavior, while controlling for self-selection, in the metropolitan region of São Paulo, Brazil. The model structure is adapted from previous applications to include variables related with specific aspects of the studied region, with the inclusion of informal work and people working two jobs, while simultaneously controlling for cohort effects associated with being a millennial. The results support the claim that land-use patterns influence travel behavior, even in a metropolitan area showing strong income-based spatial segregation levels. More specifically, commuting distance and car ownership act as important mediators in the relationships between the total amount of travel by mode and land-use patterns. In contrast to previous applications of this model framework, income plays a stronger role, an indication of relevant income-based residential sorting. Cohort effects are also visible, as millennials prefer to live in central, accessible, and mixed areas, own fewer cars, travel less by car, and use public transit and non-motorized modes more frequently.
... Environmental justice research, often focusing on distributions of burdens, also called maldistributions (Schlosberg, 2007), investigates inequalities in various exposures and risks, such as air pollution (Bravo et al., 2016), noise (Verbeek, 2019), heat risk (Mitchell & Chakraborty, 2015), and flood risk (Walker & Burningham, 2011). On the other hand, economists and public affairs scholars often study inequality in income and in the distributions of public goods, such as investments in education (Haddad & Nedović-Budić, 2006) and renewable energy (Chan et al., 2017). ...
Full-text available
Cities are increasingly advancing multiple societal goals related to environmental sustainability, health, well-being, and equity. However, there are few comprehensive data sets that address social inequality and equity across multiple infrastructure sectors, determinants, and outcomes, particularly at fine intra-urban spatial scales. This paper: (1) Offers an overarching conceptualization of inequality and equity in multi-sector urban systems; (2) Introduces a broad data framework to assess inequality and equity across social (S), ecological (E), infrastructural (I), and urban (U) form determinants (SEIU) and environment (E), health (H), well-being (W), and economy and security (E) outcomes (EHWE), identifying a universe of >110 SEIU–EHWE data layers (variables) of interest; (3) Provides an illustrative data case study of a US city that synthesizes publicly available sources of the associated SEIU–EHWE data attributes, noting their availability/gaps at fine spatial scales, important to inform social inequality; (4) Discusses analytic methods to quantify inequality and spatial correlates across SEIU determinants and EHWE outcomes; and, (5) Demonstrates several use-cases of the data framework and companion analytic methods through real-world applied case studies that inform equity planning in applications ranging from energy sector investments to air pollution and health. The US data case study reveals data availability (covering 41 of the 113 data layers) as well as major gaps associated with EHWE outcomes at fine spatial scales, while the application examples demonstrate practical use. Overall, the SEIU–EHWE data framework provides an anchor for systematically gathering, analyzing, and informing multiple dimensions of inequality and equity in sustainable urban systems.
... Просторна повезаност моделује се помоћу просторне тежинске матрице где је свака просторна јединица повезана са једном или више суседних просторних једница (статистичких кругова). У овом раду тежинска матрица дефинисана је на бази матрице најближих суседа, тако да је сваки статистички круг, без обзира на локацију, повезан са истим бројем k суседних статистичких кругова (Guillain et al., 2006;Haddad & Nedović-Budić, 2006). Образац тежинске матрице k суседних статистичких кругова израчунава се на следећи начин: ...
Determining how to reasonably protect, upgrade, or rebuild old communities, deciding how to use the material resources they contain, and the generation of construction and demolition (C&D) waste are urgent problems faced in the process of urban renewal in Tianjin, China, that require solutions. In this study, brick and concrete housing structures in the downtown area of Tianjin, China, are considered as an example, and a phased renewal plan is proposed. Based on the nuclear density analysis method and the standard deviation ellipse method, the spatial agglomeration trends and distribution directions of the used bricks that will be produced in the city center during each demolition stage are investigated. The results demonstrate that the overall brick material resources are unevenly distributed; their density will increase over time, and the density core will present a trend of first gathering and then dispersing. Based on the spatiotemporal distribution of resources, it is necessary to formulate an urban resource recycling system to achieve sustainable development.
This paper attempted to develop a methodological framework for capturing regional disparities in social development by adopting the methods suggested by McMeekin et al. (2020). A set of six conditions and their adjoining cases were developed that determined the methodology. A new methodological framework was formulated as an interconnected network of conditions, cases, tools, methods and other aspects. This framework was pilot tested in Kerala, India, to explore the regional inequalities in housing infrastructure and living conditions. Comparison of results with the index of deprivation revealed the importance of adopting multivariate methods and lower areal units of analysis to produce better results.
Full-text available
The urban inequality study in megacities has been thoroughly discussed over causes and symptoms, specifically addressing around the context of Bangkok Extended Metropolitan Region (BEMR). Having been limited by previous studies, however, the researcher attempt to briefly elucidate implying inequality causes and consequences within Thailand's megacity circumstance. Any megacity has inheritably changed overtime. On one hand, it is a geo-economic correlation which fabricates with the international context, termed as 'global city', a capitalism gateway operating upon a terrain of national geography, On the other hand especially in highly-centralized context as in Thailand,
Full-text available
Full-text available
El análisis de las desigualdades regionales en desarrollo humano ha cobrado suma relevancia, sin embargo, hay pocos estudios realizados desde una perspectiva espacial. El objetivo de este artículo es identificar patrones y dinámicas espaciales en el desarrollo humano de los municipios mexicanos para 2010 y 2015. Mediante el uso de métodos de análisis espacial, se analizan datos georreferenciados del índice de desarrollo humano (IDH) a nivel municipal. Los resultados indican la existencia de municipios con dinámicas virtuosas y viciosas de desarrollo humano que muestran un patrón persistente a través del tiempo y espacio, así como aquellos con resiliencia espacial que tienden a convertirse en enclaves de desarrollo humano. La conclusión principal es que la dependencia espacial prevalece entre municipios vecinos, y proponemos que el componente territorial del desarrollo humano sea considerado para el diseño y ejecución de las políticas sociales.
Full-text available
Based on a large number of Monte Carlo simulation experiments on a regular lattice, we compare the properties of Moran's I and Lagrange multiplier tests for spatial dependence, that is, for both spatial error autocorrelation and for a spatially lagged dependent variable. We consider both bias and power of the tests for six sample sizes, ranging from twenty‐five to 225 observations, for different structures of the spatial weights matrix, for several underlying error distributions, for misspecified weights matrices, and for the situation where boundary effects are present. The results provide an indication of the sample sizes for which the asymptotic properties of the tests can be considered to hold. They also illustrate the power of the Lagrange multiplier tests to distinguish between substantive spatial dependence (spatial lag) and spatial dependence as a nuisance (error autocorrelation).
1: Introduction.- 2: The Scope of Spatial Econometrics.- 3: The Formal Expression of Spatial Effects.- 4: A Typology of Spatial Econometric Models.- 5: Spatial Stochastic Processes: Terminology and General Properties.- 6: The Maximum Likelihood Approach to Spatial Process Models.- 7: Alternative Approaches to Inference in Spatial Process Models.- 8: Spatial Dependence in Regression Error Terms.- 9: Spatial Heterogeneity.- 10: Models in Space and Time.- 11: Problem Areas in Estimation and Testing for Spatial Process Models.- 12: Operational Issues and Empirical Applications.- 13: Model Validation and Specification Tests in Spatial Econometric Models.- 14: Model Selection in Spatial Econometric Models.- 15: Conclusions.- References.
In international cross-country studies, evidence for club convergence has often been found using different methodologies (Baumol 1986; Durlauf and Johnson 1995; Quah 1996a, 1997). In the case of the European regions, Ertur and Le Gallo (see Chap. 2) and Le Gallo et al. (see Chap. 3) have shown that the convergence rate among European regions is slow and that GDP disparities seem to be persistent despite the European economic integration process and higher growth rates of some poorer regions, as highlighted as well in the European Commission reports (1996, 1999). Moreover, over the 1980–1995 period, Ertur and Le Gallo (see Chap. 2) found that the geographical distribution of European regions exhibits a persistent polarization pattern between rich regions in the north and poor regions in the south.
Spatial analysis is statistically and substantively important for macrolevel criminological inquiry. Using county-level data for the decennial years in the 1960 to 1990 time period, we reexamine the impact of conventional structural covariates on homicide rates and explicitly model spatial effects. Important findings are: (1) homicide is strongly clustered in space; (2) this clustering cannot be completely explained by common measures of the structural similarity of neighboring counties; (3) noteworthy regional differences are observed in the effects of structural covariates on homicide rates; and (4) evidence consistent with a diffusion process for homicide is observed in the South throughout the 1960-1990 period.