Content uploaded by Monica A. Haddad
Author content
All content in this area was uploaded by Monica A. Haddad on Nov 22, 2019
Content may be subject to copyright.
BOLSA FAMI
´LIA AND THE NEEDY: IS
ALLOCATION CONTRIBUTING TO EQUITY
IN BRAZIL?
MO
ˆNICA A. HADDAD*
College of Design, Iowa State University, Ames, Iowa, USA
Abstract: This paper examines whether the Brazilian social programme Bolsa Famı
´lia is
contributing to greater social equality within the country. As a measure of success, we rely on
the programme’s impact on public-school enrolment, which we consider an input for social
equity. Our findings show that policy makers should continue with the same system of
allocation used in 2006, which proved to be contributing to greater social equality. We suggest
policy maker’s direct attention to the 2006 residual map depicting municipalities that have
extremely over- and under-predicted values, which may represent misallocation of public
funds, and therefore may require some scrutiny. Copyright #2008 John Wiley & Sons, Ltd.
Keywords: Brazil; Bolsa Famı
´lia; spatial econometrics; social equity; cash transfer
1 INTRODUCTION
Social inequality is a continuing problem in Brazil, but as president, Luiz Ina
´cio Lula da
Silva has focused more attention on the issue than ever before. When he took office in
January 2003, Lula created the new Ministe
´rio do Desenvolvimento Social e Combate a
`
Fome (Ministry of Social Development and Fight Against Hunger) to manage the country’s
investments aimed at reducing poverty, combating hunger, increasing school attendance
and improving the number of people —especially children— who receive basic health
care. One such programme, Bolsa Famı
´lia (BF) (Family Fund), is based on a direct transfer
of funds to low-income families that agree to keep their children in school and provide
them with basic health care.
Based on the Ministe
´rio’sdatabase, BF benefited 3.6 million families in its first year
(2003) at a cost of R$3.4 billion; in 2005 the programme helped 8.7 million families at a
Journal of International Development
J. Int. Dev. 20, 654–669 (2008)
Published online in Wiley InterScience
(www.interscience.wiley.com) DOI: 10.1002/jid.1445
*Correspondence to: Mo
ˆnica A. Haddad, Department of Community and Regional Planning, 583 College of
Design, Iowa State University, Ames, IA 50011-3095, USA. E-mail: haddad@iastate.edu
Copyright #2008 John Wiley & Sons, Ltd.
cost of R$5.7 billion. Furthermore, in 2005, more than 21 per cent of the federal budget was
allocated to social welfare programmes like BF, while in 1987, only 3 per cent of the
federal budget was. There is no denying that this is a significant change. While this
programme has been the topic for important studies (Brie
`re and Lindert, 2005; Hall, 2006;
Lavinas, 2006; Soares et al., 2006; Oliveira et al., 2007), only one study assesses BF from a
spatial perspective, examining whether BF investments are being allocated in
municipalities that need them the most (Haddad, in press).
As Lula began his second term in January 2007, he promised to continue to allocate a
substantial part of the federal budget to social programmes. The federal government set
aside approximately R$8.6 billion for BF in 2007. Therefore, it is crucial to assure that
these resources are contributing to greater social equality within the country. Focusing on
BF in 2006, the most recent full year for which data are available, and 2003, the first year of
Lula’s administration, this paper addresses the following research question: Are
investments contributing to greater social equality in Brazil? If investments are not
contributing to greater social equality, Lula’s compensatory policies may be spending
public funding to achieve misleading goals.
Why should we address this issue? When Oliveira et al. (2007) analyse results of a
survey focusing on the impact of BF, they conclude that ‘there are no significant differences
between the proportion of children—between 0 and 6 years old —with updated vaccines’
schedules among the BF beneficiaries and beneficiaries of other programmes’ (p. 38).
1
In
addition, the econometric results of Cardoso and Souza (2004) —when analysing social
programmes in Brazil—suggest that they ‘have not been effective in fighting child labour
in Brazil’ (p. 1). Finally, Farrington and Slater (2006) point out that one of the views about
cash transfer is that ‘social assistance payment(s) have purely consumption effects with no
bearing on productive activities, and so represent ‘money on the drain’’ (p. 500). Based on
these previous studies, if new allocations are about to be made, it is important to assure that
BF is contributing to greater social equality in Brazil.
To better examine Bolsa Famı
´lia allocation and its relationship with social equity, we
introduce the spatial dimension in our analyses. By introducing the spatial dimension we
overcome the limitations of analyses that neglect to consider the dependency of each
municipality on its geographic location. Hence, to answer our research question posed
above, we rely on Exploratory Spatial Data Analysis (ESDA) and Confirmatory Spatial
Data Analysis (CSDA) methods. Our intention is to spatially analyse this contemporary
social programme and devise responses to policy makers. With our spatial analyses, policy
makers will be able to better comprehend how social equity is being addressed by Lula’s
main social programme with maps that are easy to understand. This examination may guide
policy makers towards a new system of allocation, if needed, or the continuation of the
same system.
The paper proceeds as follows. Section 2 focuses on social development, social policies,
and a description of the BF programme. In Section 3 we explain the methodology, focusing
on the data, the spatial weight matrices, spatial autocorrelation test and model
specifications. Section 4 focuses on the modelling results, including some maps as an
1
The preliminary analysis presented by Oliveira et al. (2007) is based on the survey ‘Avaliac¸a
˜o de Impacto do
Programa Bolsa Famı
´lia’, which compared three different groups: dwellings with families who are beneficiaries of
Bolsa Famı
´lia; dwellings with families who are part of Cadastro U
´nico, but not yet beneficiaries of Bolsa Famı
´lia;
and dwellings without Bolsa Famı
´lia beneficiaries and without families who are part of Cadastro U
´nico. More
information about Cadastro U
´nico is included in the next section.
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
Bolsa Famı
´lia and the Needy 655
aid to understand the results. The last section concludes, suggesting some recommen-
dations for policy makers.
2 THINKING SOCIALLY
Traditional economic development theory advocates that economic growth generates
income gains for the poor and promotes welfare benefits to them, such as access to school
and health care. The overall outcome of this process is an improvement in the social
development status of the population. Nevertheless, as Ghai (2000) states, ‘It is commonly
observed that levels of economic development tend to be roughly correlated with levels of
social development in countries throughout the world. This observable cross-country
correlation, however, does not necessarily indicate a direct causal relationship between
economic and social development in any particular case’ (p. 1). In the same direction, there
are studies that indicate that growth alone is not sufficient to overcome social problems
(Haddad et al., 2002; Morley and Coady, 2003; Hall and Midgley, 2004; Arbache, 2006;
Haddad and Nedovic
´-Budic
´, 2006). Another example is Devas et al. (2001), which verify
in ten case studies from developing countries that ‘economic growth alone does not ensure
access for all basic needs, and can in turn increase inequality’ (p. 6). Therefore, attention to
social development should be the part of the public policy agenda, and President Lula is on
the right track when deciding to allocate public resources to social programmes such as BF.
Complementary to that, Hall and Midgley (2004) propose that ‘the best hope of raising
standards of living and eradicating poverty lies in an approach that combines a
commitment to economic development with the introduction of social policies that
specifically and directly address the poverty problem’ (p. 45). During his first term
(2003–2006), President Lula dedicated a lot of attention to social policies, leaving
economic development strategies aside. But as Lula entered his second term as president,
he delivered an economic development package called Program de Acelerac¸a
˜odo
Crescimento (Programme for Accelerating Growth— PAC). Its objective is to reach a
sustained annual GDP growth rate of 5 per cent by stimulating private investment, and
commanding and controlling higher rates of public investment in economic infrastructure.
This new combination of economic development and social policies may lead Brazil to a
new developmental stage in the coming years.
Mkandawire (2004) also advocates social policies as key instruments that work in
combination with economic policy to ensure equitable and socially sustainable economic
development. Mkandawire defines social policy as ‘collective intervention in the economy
to influence the access to and the incidence of adequate and secure livelihoods and income’
(2004, p. 1). Bolsa Famı
´lia ensures children’s access to school and health care and provides
income to their families, a good example of how a social programme can be implemented
in the real world. Social policy may target different objectives such as rural development
and urban environment improvement (Hall and Midgley, 2004). BF lies in the categories of
poverty reduction and social inequality minimisation.
Conditional cash-transfer programmes —a specific type of social policy— appeared in
Brazil prior to Lula’s first term as president. The idea behind BF is not a new approach for
Brazilians. As Morley and Coady (2003) point out, conditional transfer-for-education
(CTE) programmes have been in place in many different countries since the mid-1990.
Examples are Progressa in Mexico, Food for Education in Bangladesh, Red de Proteccio
´n
Social in Nicaragua and Bolsa Escola (School Fund) in Brazil. Bolsa Escola started as a
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
656 M. A. Haddad
municipal programme whose rules and financing were decided locally. In 2001, Bolsa
Escola was implemented at the national level (under Cardoso’s administration).
In October of 2003 Bolsa Famı
´lia was created, and the existing cash-transfer
programmes were slowly merged with the new BF.
2
The same system initiated by Bolsa
Escola is now being used by BF beneficiaries. The system lowers the cost—transportation
and time—of ‘making transfer payments to both the donor and the recipient. [...] The
mother of each beneficiary family is given an electronic card and an account at Caixa
Econo
ˆmica Federal. Monthly payments are directly credited to this account from the
national treasury, and the mother can make electronic withdrawals at any of the local
outlets of the bank or in thousands of other authorised commercial outlets’ (Morley and
Coady, 2003, p. 33). BF monetary values for the beneficiaries vary according to poverty
level and number of children, pregnant women and nursing mothers per family.
3
Brazil’s social agenda under Lula is quite expensive. According to Arbache (2006), one
way to fund it is through tax increases, but ‘the tax burden plus the nominal deficit
amounted to about 38.5 percent of the GDP’ (p. 341). An assessment of the cost of Lula’s
social agenda—based on the cost transfer ratio —should be elaborated to verify how much
funding is being absorbed by administrative costs (Calde
´s and Maluccio, 2005). In that
way, we would know the proportion of the budget that is reaching the beneficiaries.
Each municipality is responsible for the implementation of BF in its territory. Eligible
families apply by contacting the responsible party in the municipality where they live, and
presenting personal documents (social security number or voter’s card). These families
apply only once to the Cadastro U
´nico dos Programas Sociais do Governo Federal, which
is a federal government tool to collect data in order to identify all the existing poor families
in Brazil. This unique application allows families to be officially eligible for all federal
social programmes available, including BF.
Bolsa Escola— predecessor to BF—has been a topic of interest for some studies
(Bourguignon et al., 2003; Barrientos and DeJong, 2004, 2006; Cardoso and Souza, 2004;
de Janvry et al., 2005). To illustrate, Cardoso and Souza (2004) find a positive and
significant impact of Bolsa Escola on school attendance. In the same direction,
Bourguignon et al. (2003) simulated the effect of Bolsa Escola and found that it had a
big impact on school enrolments, but not so much on poverty levels, since families lose the
income of children who leave the labour force to attend school. As they state, ‘[R]esults
suggest that 60 per cent of poor 10- to 15-year-olds not in school enrol in response to the
programme’ (p. 229). Contrary to that, Barrientos and Dejong (2006), when examining
different social programmes —including Bolsa Escola—concluded that cash-transfer
programmes targeting children in poor households are an effective way of reducing poverty.
Different scholars propose different ways to assess Bolsa Famı
´lia. Below are some
examples that deserve attention. Marques et al. (2004) examined the July 2004 BF
beneficiaries for some Brazilian municipalities comparing their regional location,
2
The existing cash-transfer programmes were Bolsa Escola,Bolsa Alimentac¸a
˜o,Auxı
´lio Alimentac¸a
˜oand Auxı
´lio
Ga
´s.
3
Besides being in the per capita range of R$120.00 or lower, families who benefit from the programme need to
have children from 0 to 15 years old, and/or pregnant women, and/or nursing mothers as members. For families
considered under extreme poverty (up to R$60.00), one member receives R$76.00, two members receive R$94.00
and three or more members receive R$112.00. For this category, even if there are no eligible members, a family
receives R$58.00. For families considered under poverty (between R$60.01 and R$120.00), one member receives
R$18.00, two members receive R$36.00 and three or more members receive R$54.00. As announced in September
2007, the targeted population for Bolsa Famı
´lia will be expanded starting in 2008, when 16- and 17-year-olds will
also be included.
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
Bolsa Famı
´lia and the Needy 657
population size, Human Development Index, urban or rural status and economic-sector
predominance. By using descriptive statistics, they found that the majority of the
beneficiaries are located in the northeast region, in contrast with a lower number of
beneficiaries in the south region. On the other hand, Neri (2005) observes that the
proportion of people living below the poverty line fell significantly from 2003 to 2004,
reaching the lowest figure since 1992; he credits this result to economic growth and BF.
In summary, based on this literature review, we can state that President Lula seems
justified in investing significant public resources in a social programme, since economic
development will do no good for the poor per se. BF as an effective vehicle for social equity
is yet in need of assessment, especially knowing that the programme will not only continue
during Lula’s second mandate but will also be expanded.
3 METHODOLOGY
As stated previously, BF lies in the social policy categories of poverty reduction and social
inequality minimisation. The question —Are investments contributing to greater social
equality in Brazil?—is related to social inequality minimisation. In this section, we explain
the methodological approach to address this research question.
3.1 Research Framework
The central element of the research framework (Figure 1) is what we consider an input for
social equity: public-school enrolment that is either contributing to social equity, or not.
This practice is influenced by policy makers, who can decide on BF allocation changes
based on enrolment. In other words, policy makers may change BF allocation as a tentative
step to affect school enrolment. A feedback loop is constructed to allow for adjustments in
public policies, if policy makers are willing to make changes in enrolments, when focusing
on BF allocation.
3.2 Data
This study focuses on most of the Brazilian municipalities.
4
The dependent variable is the
natural log of the total enrolment in public schools per municipality—as a proxy for social
Figure 1. Research framework
4
Because we are using some data from the 2000 IBGE Census, we use a sample size of 5504 municipalities
(removing islands), instead of 5564 (which is the current number of municipalities in Brazil).
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
658 M. A. Haddad
equity (Source: INEP, Ministry of Education).
5
We use data for 2003 (first year of the
programme) and 2006 (most recent full-year data available) to examine changes in
public-school enrolment. In other words, BF may be contributing to increased enrolment
from 2003 to 2006, and we consider an increase in the number of children attending school
as an input that contributes to greater social equality. We control enrolment for success and
failure rates, using 2002 data for the year 2003, and 2005 data for the year 2006. Figure 2
depicts the spatial distribution of the 2003- and 2006-dependent variables, and we cannot
observe regional concentrations of low and high values.
The independent variables are eligible children, natural log of BF, and parents’
education.
6
Other variables that could affect enrolment are related to the socio-economic
characteristics of families such as family income, dwellings with electricity and television
and dwellings with sewage services from the public network. The BF application
procedure already captures these socio-economic characteristics of the families, which is
why our models do not include them. Data for the 2003 and 2006 BF beneficiaries come
from the Ministe
´rio do Desenvolvimento Social e Combate a
`Fome. It is very important to
highlight that these data sets are sensu stricto BF, not including the beneficiaries from the
existing cash-transfer programmes (pre-Lula) that were merged with BF. Table 1 depicts
the summary statistics of all variables used in this study.
3.3 Spatial Weight Matrices
To conduct ESDA/CSDA it is necessary to define a spatial weight matrix W. This matrix
imposes a neighbourhood structure on the data and can be defined in a variety of ways.
Figure 2. Natural log of public-school enrolment relative to the sample average of Brazilian
municipalities
5
This variable is a sum of Ensino Fundamental and Ensino Me
´dio enrolments per municipality.
6
Eligible children is measured as the number of children who were 0–8 years old in year 2000, and the number who
were 3–11 years old in year 2000. These numbers correspond to the eligible children who could attend school in
2003 and 2006, respectively. We divided these variables by 10 000 to scale it to the dependent variable values
(Source: 2000 IBGE Census). Parents’ education is measured as per cent of people whowere 25 years old or older
and were attending college in the year 2000 (Source:Atlas de Desenvolvimento Humano, Fundac¸a
˜oJoa
˜o
Pinheiro).
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
Bolsa Famı
´lia and the Needy 659
Because we have a large dataset (n¼5504 observations), we faced some software-related
constraints when choosing our matrices. Therefore, we opted for simple binary queen
contiguity and simple binary rook contiguity. We use two matrices in our analyses to test
the robustness of our results.
The simple binary queen contiguity matrix is composed of 0 and 1: if municipality ihas a
common boundary and/or vertex with municipality j, then they are neighbours and w
ij
¼1;
if municipality idoes not have a common boundary and/or vertex with municipality j, then
they are not neighbours and w
ij
¼0. The diagonal elements are set to 0. The rook matrix
follows the same logic except that it is based on common boundary only, not including
vertex. Both matrices are row standardised so that each row sums up to 1.
3.4 Spatial Autocorrelation
ESDA methods are used ‘to describe and visualise spatial distributions, identify atypical
locations (spatial outliers), discover patterns of spatial association (spatial clusters) and
suggest different spatial regimes and other forms of spatial instability or spatial
non-stationary processes’ (Anselin, 1998, p. 258). By using these methods we can test for
spatial autocorrelation. Spatial autocorrelation occurs when value similarity and locational
similarity coincide (Anselin, 2001). Positive spatial autocorrelation exists when high
values correlate with high neighbouring values and when low values correlate with low
neighbouring values. Negative spatial autocorrelation exists when high values correlate
with low neighbouring values, and vice versa, and no clustering pattern can be observed.
Among statistics of global spatial autocorrelation, Moran’s Iis widely used. It provides a
formal indication of the degree of linear association between the observed values and the
spatially weighted averages of neighbouring values. Moran’s Ishows if there is clustering
in the spatial distribution of a variable, and is defined as:
I¼n
SO
:
P
i
P
j
wijðximÞðxjmÞ
PðximÞ2(1)
where x
i
is the observation in municipality i;mis the mean of the observations across
municipalities; nis the number of municipalities and w
ij
is one element of the spatial
weight matrix Wwhich expresses the spatial arrangement of the data. S
O
is a scaling factor
equal to the sum of all elements of W.
Table 1. Summary statistics of all variables
Variables Mean St. Dev. Minimum Maximum
Parents’ education 0.567 0.537 0.001 5.410
Natural log public school enrolment (2003) 7.848 1.106 4.965 14.335
Natural log public school enrolment (2006) 7.787 1.105 4.806 14.267
Natural log Bolsa Familia (2006) 6.803 1.207 1.099 12.323
Natural log Bolsa Familia (2003) 5.488 1.562 0.693 10.897
Eligible children (2003) 0.539 2.807 0.014 154.665
Eligible children (2003) 0.790 3.981 0.017 217.641
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
660 M. A. Haddad
Table 2 displays the global Moran’s Istatistics for the 2003 and 2006 natural log of
public-school enrolment, using the two spatial matrices described above. The spatial
autocorrelation tests for these two variables lead to the rejection of the null hypothesis of
spatial randomness. All coefficients are statistically significant at the 0.001 level based on
the permutation approach with 999 random permutations.
The 2003 and 2006 enrolment spatial distributions are characterised by a significant
global positive spatial autocorrelation, which suggests that their values are spatially
clustered. This means that the municipalities with high enrolment values are located close
to municipalities with high enrolment values, and the municipalities with low enrolment
values are located close to districts with low enrolment values. These results indicate that
location plays an important role when examining the spatial distribution of 2003 and 2006
public-school enrolment, that is, when municipal values are compared to the country’s
mean. Thus, we cannot talk about spatial randomness or lack of spatial dependence when
focusing on the distribution of enrolment in public schools in Brazil.
3.5 The Models
Because spatial autocorrelation is present in the spatial distributions of the 2003 and 2006
public-school enrolment, we use CSDA methods as an attempt to answer our research
question, taking into account each municipality and its geographic location relative to other
municipalities in the country. These methods help avoid misspecification of models,
inefficient coefficients and erroneous statistical inferences that occur when spatial
autocorrelation and spatial heterogeneity are not addressed (Anselin and Rey, 1991).
To address the research question we propose two models: one for 2003 and one for 2006.
This time lag frame permits us to assess for impacts in enrolment and corresponds with
Lula’s first term in office and/or the first 4 years of the BF programme. The dependent
variables for the models are the natural log of 2003 and 2006 public-school enrolment. The
independent variables are eligible children, natural log of BF, and parents’ education.
Similarly, we take the following steps in specifying the model, addressing Florax’s et al.
2003 ‘specific to general’ model specification approach. Since the model specification is
the same for both years, we describe only the 2003 model below. Step 1 is on ordinary least
square OLS-based model estimated by:
PSEi¼b0þb1x1iþb2x2iþb3x3iþ"ii¼1;...;5;504 and "Nð0;s2
"IÞ(2)
where PSE
i
is the natural log of 2003 public-school enrolment for each municipality i;x
1i
,
x
2i
and x
3i
are the independent variables eligible children, natural log of BF, and parents’
Table 2. Moran’s Istatistics for 2006 2003/2006 public-school enrolment (in natural log)
Queen Rook
Moran’s ISt. value Morans’ ISt. value
Natural log public school enrolment (2003) 0.356 45.650 0.357 44.160
Natural log public school enrolment (2006) 0.367 45.400 0.369 45.580
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
Bolsa Famı
´lia and the Needy 661
education; b
0
,b
1
,b
2
, and b
3
are the unknown parameters to be estimated; and eis the vector
of errors.
In Step 1 we carry out tests to detect the presence of spatial dependence, revealing that
the spatial error model is the most appropriate specification for both years. Therefore, in
Step 2 the spatial error model is applied to include spatial dependence in the model
specification. In the spatial error model, the misspecification is handled by the error process
with errors from different municipalities displaying spatial covariance (Le Gallo and Ertur,
2003). According to Messner and Anselin (2004) ‘a spatial error model indicates that
clustering reflects the influence of unmeasured variables’ (p. 138). The spatial error model
is written as:
PSEi¼b0þbx1iþb2x2iþb3x3iþui
ui¼lWuiþ"ii¼1;...;5;504 and uNð0;s2
"IÞ(3)
where all the elements are defined as previously, and lis the scalar parameter expressing
the intensity of spatial correlation between regression residuals.
4 MODELLING RESULTS
The results of the models presented in this paper were calculated using GeoDa software
(Anselin, 2003). The results for all model estimations based on the queen matrix
are presented in this section. Reporting the model estimates based on the rook matrix would
be redundant as it leads to the same results. This fact points to the robustness of our results
with regard to the choice of spatial weight matrices.
To assess social equality we rely on the programme’s impact on public-school
enrolment, which we consider an input for social equity: the higher the number of children
attending school, the greater the possibility that these children will have more opportunities
and their quality of life will improve in the future. We believe this is a good proxy for social
equity. Table 3 presents the estimation results for OLS and spatial error models, using the
queen matrix. As one can observe, the Akaike Information Criterion (AIC), the Schwartz
Table 3. Estimation results for the public-school enrolment models for Brazilian municipalities,
2003 and 2006
2003 2006
Non-spatial Spatial Non-spatial Spatial
OLS ERROR (ML) OLS ERROR (ML)
b
0
(constant) 5.121 (0.000) 5.177 (0.000) 2.505 (0.000) 1.708 (0.000)
b
1
(eligible children) 0.07 (0.000) 0.03 (0.000) 0.02 (0.000) 0.004 (0.000)
b
2
(parents’ education) 0.62 (0.000) 0.726 (0.000) 0.447 (0.000) 0.217 (0.000)
b
3
(Bolsa Familia-log) 0.425 (0.000) 0.39 (0.000) 0.736 (0.000) 0.868 (0.000)
R
2
-adjusted 0.63 — 0.85 —
Lamda — 0.636 (0.000) — 0.823 (0.000)
AIC 11203.7 9594.13 6282.92 2362.19
SC 11230.1 9620.58 6309.38 2388.64
Log likelihood 5597.83 4793.1 3137.46 1177.09
OBS: p-values are presented in parentheses.
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
662 M. A. Haddad
Criterion (SC) and the Log Likelihood test indicate that the spatial model performs better
than the non-spatial model.
For the OLS model, the coefficients for eligible children, parents’ education, and BF are
all significant at the 0.1 per cent level— for both 2003 and 2006—revealing that a high
public-school enrolment coincides with a high number of eligible children, parents’
education and BF. For 2003, the model’s R
2
-adjusted is close to 64 per cent, and for 2006,
the model’s R
2
-adjusted is close to 86 per cent. The spatial error model—for both 2003 and
2006—has significant coefficients at the 0.1 per cent level, indicating that a high
public-school enrolment coincides with a high number of eligible children, parents’
education and BF. Also, for both years, the spatial error coefficient is positive and
significant at the 0.1 per cent level.
The magnitude of BF coefficients increases from the 2003 error model to the 2006 error
model showing a stronger relationship between the social programme and public-school
enrolment as the programme evolves. Since the dependent variables and the
BF-independent variable are in a natural log format in these error models, we can
interpret the coefficients as elasticity. In other words, an additional family receiving BF
will increase enrolment by 0.868 in 2006. In 2006 (0.868) the effect of BF is more elastic —
in fact, more than double —than in 2003 (0.39). From a municipal perspective, one would
expect that if one of the goals of this cash-transfer programme is to minimise social
inequality, municipalities characterized by higher public-school enrolment should have
higher BF investments. This can be confirmed by the models presented above showing that
there is evidence that BF is contributing to greater social equity—when focusing at the
municipal level—since we can observe an increase in public-school enrolment directly
associated with BF allocation.
The use of spatial econometrics methods allows us to map some results for comparison
and better understanding of specific locations that may need scrutiny. Figure 3 displays four
maps elaborated based on the results of the 2006 spatial error model: a predicted map, a
simulation predicted map, a map that displays the dependent variable —that is, the natural
log of 2006 public-school enrolment—and a residual map. The predicted map indicates the
distribution and intensity of the three independent variables (in this model, eligible
children, parents’ education and BF) across the Brazilian municipalities (Haddad and
Nedovic
´-Budic
´, 2006). Higher predicted values (darker grey) show more intense
independent variables. We can observe a slight concentration of darker values in the
northeast-north part of the country, but no very clear pattern can be detected.
By comparing this predicted map with the actual public-school enrolment map, we do
notice a similar pattern in the spatial distribution of higher values slightly concentrated in
the northeast-north part of the country. This slight concentration may be caused by the
BF-independent variable. Therefore, we create a simulation predicted map to examine
what would happen with public-school enrolment intensity if the entire population targeted
by BF would receive benefits from the programme.
7
By observing the frequency of the
largest range from the maps’ legend— from 8.62 to 14.26— we notice an increase of 14 per
cent of municipalities in the simulation predicted map that are within this higher range of
enrolment. Based on these findings we can state that BF is indeed contributing to greater
social equality in Brazil, when using school enrolment as an input towards that end.
7
The simulation predicted map is created by plugging in the calculated coefficients and predicted errors, the
existing independent variables and a new independent variable— natural log of BF target population— which
substitutes the independent variable of natural log of families who benefited from the programme in 2006.
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
Bolsa Famı
´lia and the Needy 663
The residual map from Figure 3 uses standard deviation classification, in which
municipalities are classified as under-predicted (positive values) and over-predicted
(negative values) based on the spatial error model. This map represents the difference
between the predicted value of a municipality and the dependent variable in that same
municipality. Table 4 depicts the idea of under- and over-prediction in a more detailed
manner. Concerning the ‘extremely high’ and ‘very high’ residual values, if policy makers
are willing to increase social equity, by relying on BF among other policies, they should
direct special attention to the darker shades (i.e. darker grey) since they represent a greater
mismatch between predicted value and the actual public-school enrolment.
First, policy makers should concentrate on ‘extremely high over-predicted’
municipalities (0.5 per cent of Brazil total) because they are characterised by very low
enrolment, but may have very high BF. Second, policy makers should focus on ‘very high
over-predicted’ municipalities (4.3 per cent of Brazil total) since they are characterized by
Figure 3. Maps based on the public-school enrolment model for Brazilian municipalities, 2006
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
664 M. A. Haddad
low enrolment, but may have high BF allocation. In these two cases, BF allocation may not
indicate the positive impact on enrolment as the overall model does. The other two
classes—‘extremely high under-predicted’ and ‘very high under-predicted’ (6.4 per cent
of Brazil total)—are characterised by very high/high enrolment and may have very low/
low BF allocation. These classes do not represent misallocation of public funds and
therefore do not require any scrutiny.
5 CONCLUSION AND RECOMMENDATIONS
In this paper, we examine the allocation of BF from the perspective of equity. We apply
ESDA and CSDA methods that account for the location of the 5504 Brazilian
municipalities and dependency among neighbouring municipalities. Haddad (in press)
applies the same methodology as described here, concluding that BF proved to be targeting
the most disadvantaged municipalities in Brazil.
The empirical evidence from our models leads to a few conclusions. Our models indicate
that BF affects enrolment from 2003 to 2006. Our findings are similar to the ones by
Bourguignon et al. (2003) and Cardoso and Souza (2004) described in Section 2.
Remember that based on the 2006 model, an additional family receiving BF will increase
enrolment by 0.868; in the 2003 model this elasticity is less than half that (0.39).
Should President Lula continue with the same system of allocation, or should a new
system of allocation be proposed? Based on the spatial analysis presented above, we
believe there is no need to change the BF allocation system overall since it is contributing to
greater social equality—based on increase in public-school enrolment. There are,
nevertheless, specific locations—outliers in the residual maps —that may merit some
scrutiny. These locations suggest the existence of other variables where location and
the independent variables are not sufficient to explain public-school enrolment. These
variables could be socio-economic characteristics of families such as family income,
dwellings with electricity and television and dwellings with sewage services from the
public network. Based on our mapping procedures in Section 4, policy makers are able to
identify these municipalities and assess those that are either characterised by both high
enrolment and low BF, or vice versa. This is our methodological contribution to the
literature.
Policy makers from other countries and regions can also benefit from this methodology.
Conditional cash-transfer programmes are very popular in both developing and developed
countries (Heinrich, 2006), and there is certainly a need to assess these programmes from
Table 4. Percentage of Brazilian municipalities under- and over-predicted based on the public-
school enrolment model, 2006
Interval Scale Percentage (%)
<2.5 St. Dev. Extremely high over-predicted 0.5
2.5 to 1.5 St. Dev. Very high over-predicted 4.3
1.5 to 0.5 St. Dev. High over-predicted 26.3
0.5 to 0.5 St. Dev. Over-predicted and under-predicted 42.6
0.5 to 1.5 St. Dev. High under-predicted 19.9
1.5 to 2.5 St. Dev. Very high under-predicted 4.5
>2.5 St. Dev. Extremely high under-predicted 1.9
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
Bolsa Famı
´lia and the Needy 665
different perspectives. When spatial data are available, this methodology can be very useful
to identify areas that deserve special scrutiny. Knowing that public resources are scarce, the
produced maps can be an efficient way to prioritise areas that merit attention.
Within this context, quality of education becomes a very important issue. As Oliveira
(2004) alerts—when examining 1997 and 1999 student performance for math and
reading—‘the majority of public schools do not reach the minimum standards. In each
grade and subject, from 60 to 80 per cent fail to do so’ (p. 46). In the same direction, when
Soares (2004) reviews 1997 expected proficiency level to various grades, he concludes that
‘the vast majority of Brazilian students have not acquired the cognitive skills expected for
their grades’ (p. 73). Handa and Davis (2006) advocate that to solve the problem of
inequalities in human capital, it is necessary to redirect ‘resources and/or attention [to]
essential investments in health and education [that] may be the only way to sustain the
long-term investment in human resources required to reduce poverty’ (p. 532), and not only
implement cash-transfer programmes.
Long-term investments in human resources are indeed important and should be part of
public agendas, together with increasing educational quality. As Cuesta (2007) suggests,
one alternative could be to integrate long-term investments with cash-transfer programmes.
To illustrate, recipients of such programmes could have access to financial credit or
micro-credit, allowing them to be under ‘a comprehensive and cohesively social protection
umbrella’ (p. 1017). The Brazilian federal government already realised that these programs
are per se unsustainable. Alternatives, such as the promotion of micro- and small-firm
clusters—as a complement for permanent job creation— and labour qualification process
are already in place to improve human resources.
If public schools do not offer a high-quality education, the future of these children who
are benefiting from BF may not be as bright as expected. Currently BF’s design does not
cover quality of education, and we believe this should change. Ways to incorporate quality
of education in the programme might include ‘enhanc[ing] the role of [municipalities] in
monitoring programme performance or influencing school management more generally’,
or ‘transfers can contain a ‘‘voucher component’’ that is transferred to the school via school
fees’ (Morley and Coady, 2003, p. 36). In addition, the BF programme could have a more
ambitious educational goal of extending ‘the traditional conditionality on school enrolment
and attendance into continuation and graduation’ (Cuesta, 2007, p. 1017).
Schwartzman (2005) proposes that some of the billions invested in education-oriented
social programs (such as BF) should be allocated instead to improve the quality of the
Brazilian public educational system, making schools more capable of dealing with children
from deprived families (p. 23 and 26). It seems that President Lula is aware that quality of
education is a complementary social policy to be implemented side by side with BF. As part
of the Program de Acelerac¸a
˜o do Crescimento (PAC), the national congress approved—at
the beginning of April 2007—some funding to increase teacher salaries.
Among Oliveira et al.’s(2007) findings is the fact that, when analysing data about
children who did not attend school in the last month, they notice a lower frequency
among BF beneficiaries in relation to beneficiaries from other programmes. They
explain that ‘conditionality of school attendance required by other programs like Bolsa
Escola and PETI [...] exist[ed] before BF, and as a consequence [those programmes] are
showing a duration effect more consistent’ (p. 43). However, this is not an issue in our
model since we use sensu stricto BF data. The fact that BF elasticity increased from 2003
to 2006 supports BF’s positive impact on enrolment under President Lula’s first
mandate.
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
666 M. A. Haddad
Finally, to advance the type of analysis presented in this paper, it would be recommended
to change the spatial unit of analysis, since intra-municipal variations may be masked when
working at the municipal level (Haddad, 2003). However, spatial disaggregated data—
lower than municipality—is very unlikely to be available in Brazil. Notwithstanding this
data problem, the public sector should make an effort to collect intra-municipal data for the
municipalities that depict ‘extremely high’ and ‘very high’ residual values. This procedure
would help illuminate the misallocation that may be occurring in their territory and
identify—in greater detail —the socio-economic characteristics of their beneficiary
families, which may be affecting public-school enrolment and other measures of social
equity. The ability to map estimation results should be incorporated in other studies using
the smallest spatial unit available. Once this mapping process is completed, policy makers
should then examine whatever locations they find intriguing and relevant.
ACKNOWLEDGEMENTS
We specially thank Ana Fava for her very valuable comments on the modelling, Ro
´ridan
Penido—from the Ministe
´rio do Desenvolvimento Social e Combate a
`Fome —for his
detailed and patient assistance regarding questions about BF, and an anonymous referee for
his/her valuable comments. We would also like to thank Ma
´rcia Azanha and Carlos Azzoni
for their feedback on an earlier version of this study, and Guilherme Moreira and Raul
Santos for helping with the data collection. Previous versions of this study have been
presented at the Brazil: President Lula’s First Administration conference (an international
conference supported by the Hewlett Foundation), University of Illinois, Urbana-
Champaign, 20–21 April 2007; and at the 48th Annual Conference of the Association
of Collegiate Schools of Planning, 18–21 October 2007, Milwaukee, Wisconsin.
REFERENCES
Anselin L. 1998. Interactive techniques and exploratory spatial data analysis. In Geographical
Information Systems: Principles, Techniques, Management and Applications, Longley P,
Goodchild M, Maguire D, Rhind D (eds). John Wiley and Sons: New York; 251–264.
Anselin L. 2001. Spatial econometrics. In A Companion to Theoretical Econometrics, Baltagi B
(ed.). Basil Blackwell: Oxford; 310–330.
Anselin L. 2003. GeoDa 0.9 User’s Guide. Spatial Analysis Laboratory, University of Illinois:
Urbana-Champaign, IL.
Anselin L, Rey S. 1991. Properties of tests for spatial dependence in linear regression models.
Geographical Analysis 23(2): 112–131.
Arbache J. 2006. Has macroeconomic policy been pro-poor in Brazil? In Pro-Poor Macroeconomics:
Potential and Limitations, Cornia Giovanni (ed.). Palgrave MacMillan: New York; 326–348.
Barrientos A, DeJong J. 2004. Childhood Poverty Research and Policy Centre Report no. 4: Child
Poverty and Cash Transfer. Save the Children: London.
Barrientos A, DeJong J. 2006. Reducing child poverty with cash transfer: a sure thing? Development
Policy Review 24(5): 537–552.
Bourguignon F, Ferreira FHG, Leite PG. 2003. Conditional cash transfers, schooling, and child labor:
micro-simulating Brazil’s bolsa escola program. The World Bank Economic Review 17(2):
229–254.
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
Bolsa Famı
´lia and the Needy 667
Brie
`re B, Lindert K. 2005. Reforming Brazil’s Cadastro U
´nico to improve the targeting of Bolsa
Famı
´lia program. Social Protection Discussion Paper Series. The World Bank.
Cardoso E, Souza AP. 2004. The Impact of Cash Transfer on Child Labor and School Attendance.
Working paper No. 04-W07. Vanderbilt University: Nashville.
Calde
´s N, Maluccio J. 2005. The cost of conditional cash transfer. Journal of International
Development 17(2): 151–168.
Cuesta J. 2007. On more ambitious conditional cash transfer, social protection and permanent
reduction of poverty. Journal of International Development 19(1): 1016–1019.
de Janvry A, Finan F, Sadoulet E, Nelson D, Lindert K, Brie
`re B, Lanjouw P. 2005. Evaluating
Brazil’s Bolsa Escola Program: Governance and Decentralization Implementation, (unpublished
report). The World Bank: Washington.
Devas N, Amis P, Beall J, Grant U, Mitlin D, Rakodi C, 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.
Farrington J, Slater R. 2006. Introduction: cash transfer: panacea for poverty reduction or money
down the drain? Development Policy Review 24(5): 499–511.
Florax R, Folmer H, Rey S. 2003. Specification searches in spatial econometrics: the relevance of
Hendry’s methodology. Regional Science and Urban Economics 33: 557–579.
Ghai D. 2000. Social development and public policy: Some lessons from successful experiences. In
Social Development and Public Policy: A Study of Some Successful Experiences, Ghai D (ed.).
UNRID: Chimppenham, UK; 1–45.
Haddad M, Alderman H, Appleton S, Song L, Yisehac Y. 2002. Reducing child nutrition: how far
does income growth take us? Food Consumption and Nutrition Division Discussion Paper No. 137.
International Food Policy Research Institute.
Haddad M. 2003. Human development and regional inequalities: spatial analysis across Brazilian
municipalities, PhD Dissertation, Department of Urban and Regional Planning, University of
Illinois at Urbana-Champaign.
Haddad M. (in press). A spatial analysis of Bolsa Famı
´lia: is allocation targeting the needy? In
(In Press), Baer W, Love J (eds).
Haddad M, Nedovic
´-Budic
´Z. 2006. Using spatial statistics to analyze intra-urban inequalities and
public intervention in Sa
˜o Paulo, Brazil. Journal of Human Development 7(1): 85–109.
Hall A. 2006. From fome zero to bolsa famı
´lia: social policies and poverty alleviation under Lula.
Journal of Latin American Studies 38: 689–709.
Hall A, Midgley J. 2004. Social Policy for Development. Sage Publications: London.
Handa S, Davis B. 2006. The experience of conditional cash transfers in Latin America and the
Caribbean. Development Policy Review 24(5): 513–536.
Heinrich C. 2006. Demand and supply-side determinants of conditional cash transfer program
effectiveness. World Development 35(1): 121–143.
Lavinas L. 2006. From mean-test schemes to basic income in Brazil: exceptionality and paradox.
International Social Science Review 59(3): 103–125.
Le Gallo J, Ertur C. 2003. Exploratory spatial data analysis of the distribution of regional per capita
GDP in Europe, 1980–1995. Papers in Regional Sciences 82(2): 175–201.
Marques R, Mendes A, Leite M, Hutz A. 2004. A importa
ˆncia do Bolsa Famı
´lia nos municı
´pios
Brasileiros. Diretoria do Departamento de Avaliac¸a
˜o e Monitoramento, Secretaria de Avaliac¸a
˜oe
Gesta
˜o da Informac¸a
˜o, Ministe
´rio do Desenvolvimento Social e Combate a
`Fome.
Messner S, Anselin L. 2004. Spatial analyses of homicides with areal data. In Spatially Integrated
Social Science, Goodchild Michael, Janelle D (eds). Oxford University Press: Oxford; 127–
143.
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
668 M. A. Haddad
Mkandawire T. 2004. Social policy in a development context: introduction. In Social Policy in a
Development Context, Mkandawire T (ed.). Palgrave MacMillan: New York.
Morley S, Coady D. 2003. From Social Assistance to Social Development: Targeted Education
Subsidies in Developing Countries. Center for Global Development: Washington, DC.
Neri M. 2005. Mise
´ria em queda; mensurac¸a
˜o, monitoramento e metas. Centro de Polı
´ticas Sociais do
IBRE/FGV.
Oliveira JBA. 2004. Expansion and inequality in Brazilian education. In The Challenges of Education
in Brazil, Brock C, Schuartzman S (eds). Symposium Books: Oxford; 41–68.
Oliveira AMHC, Andrade MV, Resende ANC, Rodrigues CG, Souza LR, Ribas RP. 2007. Primeiros
Resultadosda Ana
´lise da Linha de Base da Pesquisa de Avaliac¸a
˜o de Impacto do Programa Bolsa
Famı
´lia, (Chapter 1). Report prepared for the Ministe
´rio do Desenvolvimento Social e Combate a
`
Fome, Brazil.
Schwartzman S. 2005. Education-Oriented Social Programs in Brazil: The Impact of Bolsa Escola.
Paper prepared for the Global Conference on Education Research in: Developing Countries:
Prague.
Soares F. 2004. Quality and equity in Brazilian basic education: facts and possibilities. In The
Challenges of Education in Brazil, Brock C, Schuartzman S (eds). Symposium Books: Oxford;
69–88.
Soares F, Soares S, Medeiros M, Osorio R. 2006. Cash Transfer Programmes in Brazil: Impacts on
Inequality and Poverty, Working paper number 21. International Poverty Centre-UNDP: New
York.
Copyright #2008 John Wiley & Sons, Ltd. J. Int. Dev. 20, 654–669 (2008)
DOI: 10.1002/jid
Bolsa Famı
´lia and the Needy 669
