Volume 29, Issue 2
 
"It is the economy, companheiro!": an empirical analysis of Lula's re-election
based on municipal data
 


Claudio Djissey Shikida
IBMEC-MG
Leonardo Monteiro Monasterio
UFPEL
Ari Francisco de Araujo Jr.
IBMEC-MG
Andre Carraro
UFPEL
Otávio Menezes Damé 
UFPel
Abstract
This paper discusses the reasons that led to the Lula's 2006 re-election. Spatial analysis methods revealed that,
contrary to 2002, the President had more votes in less developed municipalities of Brazil. The econometric results cast
doubt on the analyses that attribute to Bolsa Família Programme total responsibility for the re-election. Lula''s electoral
success results from changes in the labor market, low inflation and an export boom that have reduced inequality and
improved the real wages of the Brazilian poor.
We are grateful for the comments and criticisms from João R. Sanson (UFSC), Duilio de Ávila Bêrni (PUC-RS), Pedro Henrique C. G. de
Sant'Anna and Martin D. Brauch. Leonardo Monasterio acknowledges the financial support of Conselho Nacional para o Desenvolvimento
Científico e Tecnológico (CNPq). The authors are responsible for any remaining error
Citation: Claudio Djissey Shikida and Leonardo Monteiro Monasterio and Ari Francisco de Araujo Jr. and Andre Carraro and Otávio 
Menezes Damé, (2009) ''"It is the economy, companheiro!": an empirical analysis of Lula's re-election based on municipal data'', Economics
Bulletin, Vol. 29 no.2 pp. 976-991.
Submitted: Jul 23 2008.   Published: May 16, 2009.
 
     

1. Introduction
The last decade witnessed the rise of cash transfer programmes (CTP) as the main tool
of social policy in less-developed countries, mainly in Latin America. At the same time the
growth of the world economy has caused a surge in regional exports. What are the electoral
consequences of these processes?
Brazil is a good example of these trends. In just four years (2002-2006) the value of
its exports more than doubled and about one-fourth of its families receive monthly transfers
from Programa Bolsa Família (PBF, Family Grant Programme), the federal government CTP.
On the 29th November 2006, more than 59 million Brazilians voted for president Lula’s re-
election, in the largest number of votes in the history of the country. One year before, this
result could have been predicted by few, even among the “companheiros”1. Starting from a
very high level of popularity, Lula saw his popular approval drop during his first
administration, amid corruption denunciations, and recovered during the months preceding
the re-election. In 2006, he got 61% of the valid votes in the run-off, the same percentage that
he had got four years earlier (TSE, 2006).
This electoral performance has generated a more intense discussion on voter’s
behavior than in other campaigns. The profile of Lula’s electorate has changed. Based on
state-level data, analysts have realized that candidate Lula won especially in less developed
regions of the country. This evidence was interpreted in different ways, according to
individual electoral sympathies. For the Partido dos Trabalhadores (Workers Party) voters,
such pattern would be the result of an administration committed to Brazilian poorest
populations. The opposition, on its turn, understood this pattern as the result of Lula’s
paternalist policies and/or a sign that the candidate was the choice of the less educated, the
Brazil of the past. Others emphasized the geographic differences between those who
supported or not the president’s re-election, suggesting a north-south split in the country. We
argue that all of these views are insufficient for understanding the phenomenon.
Despite some exceptions, e.g, Hunter and Power (2007); Nicolau and Peixoto (2007),
still little empirical effort has been made in order to investigate the factors that lead to
candidate Lula’s voting. Thus, the objective of this note is to test how social and economic
variables determined the vote for candidate Lula in the presidential election of 2006, using
spatial econometrics and spatial analysis.
In order to reach those objectives, this note is structured in more four sessions. In the
next one, we present a review of the hypothesis on Lula’s election. The third section is
dedicated to the application of spatial econometric analysis methods, as well as its application
to the 2006 election. In the following section, the determinants of Lula’s re-election are
discussed based on the empirical results obtained. Final considerations end this note.

2. The hypotheses on Lula’s election
Lula’s unexpected recovery and the brutal change in his electoral profile demand a
response. Hunter and Power (2007) divide the explanations for these phenomena in three
categories:
a) Different responses of the voters to scandals: poor and less educated voters would
have less access to the denunciations or would not repudiate corruption as boldly as other
segments of the population.
b) Economic factors: voters from lower income strata would have paid back the
amelioration of their standards of living during the Lula administration with their votes.

1
This was the usual term of address among the Brazilian left in the 1970s and among the members of the union
movement that founded the Workers.

1
c) Focused social policies: especially PBF would have been the key to understand
Lula’s victory.
Let us analyze each one of those hypotheses at a time. PBF belongs to the group of
CTP that was disseminated in Latin America since the 1990s. Heir to the programmes that
already existed in the Fernando Henrique Cardoso administration, PBF, supported by the
World Bank, became the most important social policy of the Lula administration2. According
to the Ministério do Desenvolvimento Social e Combate à Fome (2006), PBF is a direct
income transfer programme, subject to conditionalities, that benefits poor families (monthly
income per capita between US$ 27.50 and US$ 553) and extremely poor families (monthly
income per capita up to US$ 55). Still according to official information, PBF would be
guided by the articulation of three dimensions for overcoming famine and poverty: poverty
reduction through direct income transfer to the families; strengthening the exercise of social
rights in the areas of health and education through the fulfillment of the conditionalities;
complementary programmes for overcoming the situation of vulnerability and poverty.
Depending on the family’s monthly income per capita and the number of children, the
importance paid by PBF can vary between US$ 6.88 and US$ 43.58. In July 2006, the
programme reached more than 11 million families with an average benefit of US$ 28.18. In
effective terms, PBF’s most significant dimension was undoubtedly direct income transfer4.
In Brazil, the hypothesis that attributes Lula’s success to his CTP was proposed even
before the elections. The NGO Contas Abertas raised it three months before the run-off5:

With the elections coming soon, the rush for enlarging the number of people who
benefit from Bolsa Família is in accelerated pace. Payments destined to “income
transfers with conditionalities” increased 56% within a month, from US$ 290
million in June to US$ 454 million in July. The value spent in July was 73%
greater than the average budget of the programme during the last six months,
which was about US$ 263 million. (ContasAbertas, 2006)

Nicolau and Peixoto (2007) analyzed the determinants of the vote for Lula in the last
elections through econometric tests and were convinced about the CTP effect. Hunter and
Power (2007), too, consider the hypothesis that CTP is the most relevant factor in Lula’s
victory.
As for economic explanations, it is worth observing some changes in the Brazilian
economy that may have contributed for Lula’s victory. Data recently reviewed by IBGE show
that the GDP growth, despite far from spectacular, was more significant than expected. In the
first period of Lula’s government, GDP increased 3.3% per year, instead of the 2.6% per year
from the original data series. Even though it is still low, this growth rate was almost one
percentage point above the previous administration by Fernando Henrique Cardoso. In 2006,
GDP increased 3.7%. Part of this growth was stimulated by the notable growth of Brazilian
exports, which almost doubled in the period, from US$ 73 billion to US$ 137.5 billion
(AliceWeb, 2007). Motivated by the growth of the world demand, this exports boom possibly
had electoral effects. Commenting Lula’s voting, Samuel de Abreu Pessôa states:


2
For a broad criticism of the PBF, see Hall (2006).
3
Throughout the paper, we have converted the values from Brazilian Reais to US Dollars using the average
exchange rate for 2006 (2.18 R$/US$).
4
For a review of electoral effects of CTPs see, for example, Molinar and Weldon (1994), Bruhn (1996), Dion
(2000), Menocal (2001) and De La O (2006).
5
Contas Abertas (2006).

2
The president was well succeeded in the elections because he distributed income
without growing, providing the demands of the majority of voters. Most of the life
quality improvement resulted neither from Bolsa Família nor from the minimum
wage increase. It was a combination of an external shock, with an increase in
commodity prices, which appreciated the exchange rate and made food much
cheaper, and of a series of technical advances…which also diminished the cost of
food in Brazil. (Jornal Valor Econômico, 19 jan. 2007)

Thus, there would be a causal chain through which exports would lead to a real wage
increase for the poor, which, recognizing their welfare gains, would support Lula.
Finally, there is the hypothesis that voters would have different behavior towards
corruption accusations. Hunter and Power (2007) consider that, due to the broad regional
disparities and the low market penetration of magazines and newspapers, part of the voters,
especially the poorest, would not get informed about corruption denunciations. A research
showed that in region Northeast 43% of the interviewees considered themselves uninformed
about those accusations, while in the Southeast this figure was not above 25% (Hunter and
Power, 2007, p. 12). Moreover, still according to the authors, pieces of public opinion
research would show that tolerance regarding patrimonialism would drop according to the
interviewees’ educational level. In short, the poor would have voted for Lula due to ignorance
or indifference towards the corruption accusations.

3. Econometric Analysis

3.1. Exploratory Spatial Data Analysis
The data here analyzed on the municipal voting for candidate Lula in the presidential
election of 2006 have as primary source the Tribunal Superior Eleitoral (TSE, Superior
Electoral Court) website. The data are municipal, totalizing 5,507 observations. Exploratory
Spatial Data Analysis (ESDA) is the set of techniques used to describe spatial distributions of
variables, discovering spatial correlation patterns or identifying the occurrence of clusters, or
even indicating outliers (Anselin, 1995). In the present paper, ESDA will be useful in
identifying hotspots, that is, municipalities with high participation of votes for candidate
Lula.
Figure 1 shows the distribution of the “Votes Lula%” variable by municipality in the
2002 and 2006 run-offs. The difference in the spatial distribution of his votes is evident. In
2002, there was a mosaic, without any large spatial concentration of Lula’s voters. In the next
election, a distinct spatial pattern becomes clear and there is a central band of low values that
cuts Brazil from north to south. It is also noticeable that, even though region Northeast
concentrates municipalities with high voting for Lula, there is a broad variation in the
percentage of votes for the candidate within each state.
Helping to recognize spatial patterns, the local Moran indicator identifies the
municipalities in which there were statistically significant clusters of high or low values. The
local Moran’s I is (Fotheringham et al., 2005):

∑
∑
−
−−
=
i
i
j
jiji
i
nxx
xxwxx
I
/)(
)()(
2*
**


Where: xi is the observed value of x (Votes Lula %) at the location x, x* is the mean
of x, n is the number of observations and wij is the element i,j of the contiguity matrix W(1

3
for neighboring municipalities and 0 for non neighbors)6. Figure 2 evidences the central
cluster of Low-Low Values, while the High-High clusters tend to be located within a large
area more to the east of Brazil.

3.2. Spatial Econometrics – Data and Specification
The econometric analysis is based on data from three sources: Brazilian Human
Development Atlas, Superior Electoral Court and Ministry of Social Development (PNUD,
2003; TSE, 2006; Ministério do Desenvolvimento Social, 2006). The Atlas offers the
explanatory variables incorporated in the regressions and has its primary source in data from
the 2000 Demographic Census7. Data are presented on a municipal level. Due to the creation
of municipalities, there are differences between the municipal limits used in the 2000 Census
and present limits. Therefore, there is a lack of information for a small part of the
observations. In spatial econometrics, one way of overcoming this problem is by replacing
the unavailable value by its spatial lag (Bhati, 2004). This procedure, based on Tobler’s
Law8, has the disadvantage of inserting spatial autocorrelation in the data. In the present case,
this cannot be done, since some contiguous observations are missing. Thus, an even simpler
practice was adopted: the average of the observations was imputed to the missing values.
Since this procedure was necessary in but a small number of municipalities, this imputation is
not expected to generate a bias in the estimation.
The initial step is the estimation of the following linear regression model, by ordinary
least squares:
iii Xllula εδα ++=
r
(1)

in which i = 1, ...; n are municipalities; llulai is the natural logarithm of the proportion
of votes for candidate Lula in the municipality i in the 2006 election; iX
r

is a vector of
explanatory variables expressed in logarithmic form; δα ,

are the parameters to be
estimated and iε

is the stochastic error according to the usual hypotheses (non-
autocorrelation, homoscedasticity and normal distribution).
The composition of the X vector, followed by its definitions, is as follows:
a) Distance between the municipality and the state capital – Euclidian distance;
b) Inequality – Income inequality, measured by the Gini coefficient in 2000;
c) Demographic density – Municipal demographic density in 2000, in inhabitants per
square kilometer;
d) Income per capita – Municipal per capita income in 2000, in Brazilian Reais (R$).
e) Child mortality – Percentage of children dying under five years of age in 2000.
f) Illiteracy rate – Percentage of people aged 15 and over who were illiterate in the
municipality in 2000;
g) Lula’s voting in 2002 – Percentage of the valid votes that the candidate got in 2002
run-off.
h) Bolsa Família – Number of families assisted by CTP in July 2006 in relation to the
total, poor and indigent population. The value was multiplied by four, the average number of

6
The neighboring criterion adopted in this work was that of first order Queen contiguity, in which
municipalities are considered neighbors when they have at least one common border point.
7
In ideal terms, data should be available on the characteristics of the municipal population in the year of the last
presidential election. However, since the 2000 Census is the only source of data that is adequate to the present
analysis, it is necessary to suppose that the municipal characteristics continued the same or changed without
bias in the six years between the Census and the election.
8
“Everything is related to everything else but nearby things are more related than distant things” (Tobler, 1970,
p.236).

4
family members, in order to generate an indicator of the number of individuals who benefit
from the programme.
Due do the data characteristics preliminarily explained in the previous section, we
chose to go beyond traditional econometrics and use spatial econometric tools. When existing
spatial effects are ignored, the results are biased estimators and/or erroneous inferences. Two
spatial specifications are tested here: the spatial lag model and the spatial error model.
Formally:

Spatial Error Model:
iii Xllula εδα ++=
r
(2)
ij
j
iji uw +



= ∑ ελε
(3)

Spatial Lag Model:
iij
j
iji Xllulawllula εδαρ +++



= ∑ r
(4)

In the spatial error model, λ is a scalar of the error coefficient and ui ~ N (0, σ2I) and
W is the same binary contiguity matrix used in Moran statistics. This model is recommended
when there are unmeasured variables that are correlated across space or when there is a poor
match between spatial pattern of the phenomenon and the units of observation. In the spatial
lag model, the ρWllulai component is added among the explanatory variables of the classic
model. This component is the average (of the logarithm) of candidate Lula’s voting in the
neighboring municipalities. This specification implies the existence of substantive spatial
dependence, i.e. neighborhood effects in the dependent variable. In both models, the
estimation is carried out by maximum likelihood.
In the search for the appropriate specification, the strategy proposed by Florax et al.
(2006) has been followed. The authors recommend estimating the classic model and testing
for spatial autocorrelation. If it is detected, then it is necessary to choose between the error
and the lag model. Still following the recommendations by those authors, when both spatial
tests (error and lag) of the Lagrange multiplier are significant, the spatial correction related to
the highest value of the test should be made.

3.3. Econometric Results
Table I presents the results of the classic model. The results indicate that Lula’s voting
is always inelastic in relation to the right-hand side variables. The highest among them, in
absolute value, is income per capita (and it is the only one that is inversely related: a 1%
increase in income per capita diminishes the candidate’s voting in 0.154%), followed by
inequality proxies (positive relation with the candidate’s voting), Bolsa Família, demographic
density, child mortality, illiteracy and distance to the state capital.
In the present case, the diagnosis tests have indicated the presence of spatial
autocorrelation in the classic model, suggesting that the spatial error model should be chosen
instead of the spatial lag model. The value of the LM test statistic for the spatial error model
is much higher than for the lag model. In Table II, Model II presents the results of the spatial
error model estimation with the same variables as in the first model. The fact that the spatial
error model is the most adequate indicates that the classic specification ignored that the
evaluated process does not spatially correspond to the analytical units. In this case, it means

5
that municipal border lines do not precisely delimitate the phenomenon of Lula’s voting
determinants9.
It is worthwhile to reflect on the result regarding CTP. In the initial model, the
transfers, measured in per capita terms, would have had a positive effect on Lula’s voting. In
this sense, the result here is similar to that reached by Nicolau and Peixoto (2007). These
statements are rather audacious for a preliminary work based on a questionable econometric
specification. The Bolsa Família per capita variable is correlated with the percentage of poor
in the population. This, in its turn, is correlated with Lula’s voting, and thus it is more than
expected that Lula’s voting is to be correlated with the percentage of people who benefit from
CTP. Correlation does not imply causation and there may be a variable (the percentage of
poor by municipality) that intermediates the relation between the Bolsa Família proxy and
Lula’s voting. The use of the MHDI-Income variable may not be sufficient as a control and it
would be more prudent to use other proxies that captured the programme range. Thus, instead
of considering only total municipal populations, we additionally considered those assisted by
the programme in relation to the poor population (Model III) or to the indigent population
(Model IV).
In general, the low elasticities of the variables persist, as in Model I. Besides, it is
noticeable that elasticities of demographic density, income per capita and distance to the state
capital do not vary much between the models, neither in terms of sign nor in terms of
magnitude. There is also some variation in the absolute value of illiteracy and inequality
elasticities. The “Lula’s voting in 2002” variable is positive and significant. This means that,
despite the changes in the electorate’s profile, with due controls, there was a spatial
continuity in the candidate’s support between the elections. However, its impact on the
candidate’s voting in 2006 is as important as or less important than, for example, the impact
of demographic density or child mortality.
As for the electoral effect of the CTP, in Model II the sign is positive and significant,
as in the spatial model, but the elasticity is even lower. In Model III, the coefficient for Bolsa
Família by poor is positive and significant at the 10% level, but not at 5%. Finally, in Model
IV, the parameter estimated for the Bolsa Família by indigents proxy is not statistically
different from zero and does not have the same sign as in previous specifications. These
results make the positive impact of CTP, with due controls, not as spectacular and statistically
reliable as it would seem in a first analysis.


4. Discussion of the hypotheses on Lula’s election
According to the econometric results, Lula’s voting was higher in poorer, more
unequal municipalities with higher illiteracy rates. Still, there were no robust evidences that
the Programa Bolsa Família per se had an effect on Lula’s voting.
Consider Hunter and Power’s (2007) hypothesis that there would be distinct responses
of the voters to corruption scandals. Actually, it divides into two theses: a) Lula’s voters were
not informed about the scandals; b) if they were informed, they would not be as bothered
with the facts as other social segments. Given our methods and databases, both theses are
quite hard to be refuted (or accepted). We tried to test thesis (a) through the inclusion of the
variable “distance to the state capital” as a proxy for information accessibility. None of the
regressions showed empirical support for this thesis. We admit that our proxy is problematic,
but there is no alternative available. However, against thesis (a) we note that 90.3% of

9
Classic and spatial lag estimations were carried out for all the mentioned models. The tests always suggest the
existence of spatial autocorrelation, and – according to the present criteria – the spatial error model continued
to be the most appropriate. For synthesis reasons, we have chosen not to present such results, which can be
requested to the authors.

6
Brazilian households have television (IBGE, 2005) and that accusations against the Lula
administration were broadly covered by the country’s main communication networks. As for
part (b) of the hypothesis, the methods here adopted do not allow us to make inferences about
its pertinence.
The economic hypothesis raised by Pessôa (Jornal Valor Econômico, 19 Jan. 2007)
and by Hunter and Power (2007) needs to be addressed – that is, the economic welfare gains
of the poor would have been determinant in Lula’s election. Taking price index variations
into account, we notice their differences according to social classes. Observing the period
from the moment Lula took up office in January 2003 to the 2005 elections, the official
Brazilian index of inflation, “Índice de Preços ao Consumidor Ampliado” (IPCA), which
considers household incomes up to 40 minimum wages, increased 24% (IPEADATA, 2007).
Over the same period, basic food basket prices calculated by Departamento Intersindical de
Estatística e Estudos Socioeconômicos (DIEESE) had smaller increases: in the state capitals
of Rio Grande do Sul and São Paulo this index increased 8.5% and 10.4%, respectively. In
Recife and Fortaleza, capitals of Pernambuco and Ceará, basic basket prices varied 4% and -
3% in this period, respectively. In the first state, Lula got 82% of the votes in the first round
and 75% in the second. Even controlling for other factors, the econometric analysis show that
these states were the ones that most supported Lula in 2006. A recent paper showed that the
inequality in the labor market was already decreasing between 2001 and 2004 (IPEA, 2007,
p. 14):

Presented estimations show an annual growth of 7.2% for the poorest 10%, even
though the Brazilian income per capita declined 0.9% per year in the same period.
Considering the average income of the poorest 50%, we observe that it increased
2.4% per year, while the average income of the richest 50% decreased 1.4% per
year.

According to the same study, the most important cause of the inequality reduction
occurred in the labor market. It was estimated that greater educational and spatial
homogeneity, among other factors, would have been responsible for 42% of the decrease in
the Gini Coefficient in the period. CTP would have been responsible for 10% of this
reduction.
Still in what concerns the economic hypothesis, we decided to test the direct impact of
state exports. The econometric methodology used in this paper does not allow us to test this
possibility. We will evaluate a much simpler effect: the possibility that the exports boom, for
being concentrated in some states, would have also had concentrated electoral effects. Taking
2003 as the reference year, Table III shows that the exports increase was quite scattered
among the federative units.
There are data on exports on a municipal level. However, these data tend to present
distortions. For example, municipalities in which there are oil refineries or petrochemical
plants show very high values of exports and the data is misleading. It is also worth noticing
that it was not possible to use state GDP data, since state-level data are only available up to
2004, and thus two key-years for our analysis would be overlooked. In this case, we have
decided to adopt a different strategy. Instead of inserting variables referring to exports in the
spatial regressions, we chose to compare the obtained values with the dummies10.

10
The reference geographic unit for the dummy is the state of Roraima. That is, all the values on the graph are
related to the state effect relatively to that state.

7
On the Graphs 1 and 2, there are the effects of the state dummies11 versus two proxies
for the exports boom: the level of exports per capita by federative unit in 2006 and its growth
rate between 2003 and 2006. We observed federative units such as Amapá, in which there
were strong exports increases, and which, in fact, massively supported Lula. But the dummy
for Amazonas had a similar positive effect, in spite of the low exports growth. In sum, both
graphs show that no linear relation between the variables and, therefore, state dummies
effects do not seem to be related to the exports boom12.
The Graphs 1 and 2 lead us to believe that the possible effects of the exports on Lula’s
election did not occur at a state level. The relation seems to be more connected to Pessôa’s
thesis, that is, exports generated wage increases by impacting the price level.

5. Conclusions
What determined candidate Lula’s re-election as president? The econometric analysis
suggests that Lula’s voting was inversely related to municipal income per capita and directly
related to child mortality, illiteracy and inequality. The distance to the state capital was not
statistically significant. Despite the radical change in the spatial pattern of Lula’s electorate,
his voting in 2002, with due controls, appeared to be related to his voting in 2006. It should
be admitted one of the limitations of the present paper (and of other similar studies of PBF as
well): no causal relationships were identified in the present spatial econometric analysis.
Unfortunately, there were no instrument variables available that could solve the problems of
endogeneity that may have plagued the results. Nevertheless, the Bolsa Família Programme
showed some evidence of positive impact on the election, but the results did not appear to be
robust. Even if it were significant, the value of the estimator would be much smaller than
necessary to explain Lula’s victory.
Among the groups of explanations presented by Hunter and Power (2007), the one
that stresses the economic factors has the strongest data support. The inequality decrease and
the income gains in the poorest sectors of the population seemed to be the determinants. As
demonstrated by a study carried out by the Instituto de Pesquisa Econômica Aplicada (IPEA,
2007), there is a recent tendency of decrease in the inequality of income distribution, and
lower strata would benefit the most from it. A great part of those gains did not directly result
from policies carried out in Lula’s first administration, but from structural and demographic
changes through which the Brazilian economy has been going.
The income gains for the poor were preserved thanks to a very low inflation rate.
More than that, it was shown that prices tended to be more stable for products related with the
goods consumed by the poor and in the capitals of less developed states. Even though there
are no evidences of direct effects of the exports boom on electoral results, this change in
relative prices indicates that the mechanism suggested by Pessôa (Jornal Valor Econômico,
19 Jan. 2007) actually occurred. That is, the increase of exports elevated the real wages of the
poorest through exchange rate mechanisms.
Further quantitative analysis is surely necessary to identify the precise mechanisms
that have lead to Lula’s upturn and re-election. However, his electoral success ceases to be
such a challenging enigma if the explanations presented here are considered. It seems that
voters returned their welfare gains with their votes, in Hunter and Power’s words (2007). This
gratitude, however, was not totally deserved. The structural, demographic and educational
changes that smoothened income concentration were already in progress when Lula came to
power and were intensified thanks to a thriving world economy. It is ironic that his electoral
success is the result of economic policies that are antithetic to the historical left-wing

11
That is, the antilog-1 of the estimators associated to the state dummies. These values are basically similar
among the state models, and specification IV has been chosen as a reference for this section.
12
The relation between state exports and the dummies has been tested and, again, there was no relation.

8
promises of the Workers Party. After all, it was the conservative monetary policy and the
broader integration with the international economy that extended the gains of the poor and
allowed an electoral victory in 2006.

REFERENCES
AliceWeb. Sistema de Análise das Informações de Comércio Exterior via Internet,
(Ministério do Desenvolvimento, Indústria e Comércio Exterior).
http://aliceweb.desenvolvimento.gov.br/default.asp, 2007.
Anselin, L. (1995) Local Indicators of Spatial Association – LISA. Geographical Analysis v.
27, n. 2, p.93-115.
Bhati, A. S. (2004) Robust Spatial Analysis of Rare Crimes, Urban Institute, Washington.
Bruhn, K. (1996) Social Spending and Political Support: The "Lessons" of the National
Solidarity Program in Mexico. Comparative Politics v. 28, n. 2, p.151-77.
Contas Abertas. (2006) Governo aumenta gasto com Bolsa Família em 56%.
http://contasabertas.uol.com.br/noticias/detalhes_noticias.asp?auto=1486.
De La O, A. L. (2006) Do Poverty Relief Funds Affect Electoral Behavior? Evidence from a
Randomized Experiment in Mexico. Mimeo.
Dion, M. (2000) The Political Economy of Social Spending: The Mexican Solidarity
Program, 1988-1994. Estudios Sociológicos v. 18, n. 53.
Fothering, A., Brunsdon, C., Charlton, M. (2005) Quantitative Geography. London, Sage.
Florax, R. J. G. M., Folmer, H., Rey, S. J. (2006) A Comment on Specification Searches in
Spatial Econometrics: The Relevance of Hendry's Methodology: A Reply. Regional Science
and Urban Economics v. 36, n. 2, p. 300-308.
Hall, Anthony. (2006) From Fome Zero to Bolsa Família: social policies and poverty
alleviation under Lula. Journal of Latin American Studies v. 38, p. 689-709.
Hunter, W., Power, T. (2007) Rewarding Lula: Executive Power, Social Policy, and the
Brazilian Elections of 2006. Latin American Politics & Society, v. 49, p.1-30.
IPEA. (2007) Instituto de Pesquisa Econômica Aplicada. Sobre a Recente Queda da
Desigualdade de Renda no Brasil, p.14.
http://www.ipea.gov.br/sites/000/2/publicacoes/NTquedaatualizada.pdf.
IPEADATA. (2007) Instituto de Pesquisa Econômica Aplicada. http://www.ipeadata.gov.br.
Jornal Valor Econômico, (19 Jan. 2007). Contrato social impede crescimento, diz Samuel
Pessôa, http://clipping.planejamento.gov.br/Noticias.asp?NOTCod=332746.
Menocal, A. R. (2001) Do Old Habits Die Hard? A Statistical Exploration of the
Politicisation of Progresa, Mexico's Latest Federal Poverty-Alleviation Programme, under the
Zedillo Administration. Journal of Latin American Studies, v. 33, n. 03, p.513-38.
Ministério do Desenvolvimento Social. (2006) Bolsa Família: o que é,
http://www.mds.gov.br/programas/transferencia-de-renda/programa-bolsa-familia/programa-
bolsa-familia/o-que-e.
Molinar, J., Weldon, J. (1994) Electoral Determinants and Consequences of National
Solidarity, in: Cornelius, W. et al (eds.), Transforming State Society Relations in Mexico: The
National Solidarity Strategy, La Jolla.
Nicolau, J., Peixoto, V. (2007) As bases municipais da votação de Lula em 2006.
http://jaironicolau.iuperj.br/ artigos%20_files/FORUM-2007.pdf.
PNUD. (2003) Atlas do Desenvolvimento Humano, IPEA/PNUD.
Robinson, W. S. (1950) Ecological Correlations and the Behavior of Individuals, American
Sociological Review v. 15, n. 3, p.351-357.
Tobler, W. R. (1970) A Computer Movie Simulating Urban Growth in the Detroit Region,
Economic Geography, v. 46, p. 236.
TSE (2006). Centro de Divulgação da Justiça Eleitoral. http://agencia.tse.gov.br.

9

Figure 1 – Percentage of valid votes for candidate Lula by municipality in 2002 and 2006
run-offs
2002 Elections























2006 Elections





















Source: TSE (2006)

10
Figure 2 – Clusters with local Moran I significant at 99%, based on the percentage of
votes for candidate Lula in the municipalities.




































Source: calculations by the authors.

11
Table I – Determinants of candidate Lula’s voting percentage by municipality in 2006
(classic model)



























Source: Calculations by the authors.
Note: p values in parentheses. State dummies have been omitted.
I
Constant -0.043 (-0.658)
Distance 0.009 (0.017)
Inequality 0.124 (0.000)
Demographic density 0.050 (0.000)
Income per capita -0.154 (0.000)
Child mortality 0.037 (0.000)
Illiteracy 0.024 (0.022)
Bolsa Família 0.066 (0.000)
Lula’s voting in 2002 0.018 (0.063)
LM (lag) 861 (0.000)
LM (error) 1948 (0.000)
R2 0.677
F test 348
Log Likelihood 2005
Akaike Information Criterion -3942
Schwarz Information Criterion -3717

12
Table II – Determinants of candidate Lula’s voting percentage by municipality in 2006
(spatial error model)
II III

IV

Constant -0.448 (0.000)
-0,476
(0,000)
-0,488
(0,000)
Distance -0.003 (0.656)
-0,002
(0,645)
-0,003
(0,655)
Inequality 0.041 (0.129)
0.075
(0.006)
0.063
(0.025)
Demographic density 0.039 (0.000)
0.039
(0.000)
0.038
(0.000)
Income per capita -0.109 (0.000)
-0.130
(0.000)
-0.121
(0.00i0)
Child mortality 0.035 (0.047)
0.017
(0.021)
0.025
(0.053)
Illiteracy 0.053 (0.000)
0.069
(0.000)
0.069
(0.000)
Lula’s voting in 2002 0.023 (0.005)
0.023
(0.005)
0.022
(0.005)
Bolsa Família by
Population
0.041
(0.000)
Bolsa Família by Poor 0.012 (0.062)
Bolsa Família by
Indigents
-0.003
(0.549)
λ 0.643 (0.000)
0.650
(0.000)
0.649
(0.000)
Error lag LR 1240 (0.000)
1288
(0.000)
1278
(0.000)
R2 0.759 0,758 0.757
Log Likelihood 2625 2608 2607
Akaike Information
Criterion -5182 -5149 -5146
Schwarz Information
Criterion -4957 -4924 -4921

Source: Calculations by the authors.
Note: p values in parentheses. State dummies have been omitted.

13
Table III – Brazilian exports by federative unit – 2003-2006 in millions of US dollars

UF 2003 2006
Growth
Rate 2003-
2006
RONDONIA 98 308 215%
ACRE 5 18 233%
AMAZONAS 1300 1523 17%
RORAIMA 4 15 301%
PARA 2678 6708 151%
AMAPA 20 128 554%
TOCANTINS 46 204 347%
MARANHAO 740 1713 132%
PIAUI 59 47 -20%
CEARA 761 957 26%
RIO GRANDE DO NORTE 310 372 20%
PARAIBA 168 209 24%
PERNAMBUCO 411 780 90%
ALAGOAS 361 693 92%
SERGIPE 39 79 103%
BAHIA 3259 6772 108%
MINAS GERAIS 7434 15638 110%
ESPIRITO SANTO 3535 6720 90%
RIO DE JANEIRO 4844 11470 137%
SAO PAULO 23074 45930 99%
PARANA 7153 10002 40%
SANTA CATARINA 3696 5966 61%
RIO GRANDE DO SUL 8013 11774 47%
MATO GROSSO 2186 4333 98%
GOIAS 1102 2092 90%
DISTRITO FEDERAL 15 66 343%
MATO GROSSO DO SUL 498 1004 102%
Source: calculations by the authors and Aliceweb (2007).

14
Graph 1 – Per capita Exports in US dollars, 2006, and the effect of the dummies on Lula’s
voting

AC
AL
AM
AP
BA
CE
ES
GO
MA
MG
MS
MT
PA
PB
PE
PI
PR
RJRN
RO
RS
SCSE
SP
TO
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
0 500 1000 1500 2000 2500

Source: calculations by the authors and Aliceweb (2007).

Graph 2 – Exports growth rate 2003-2006 and the effects of the dummies on Lula’s voting.
AC
AL
AM
AP
BA
CE
ES
GO
MA
MG
MS
MT
PA
PB
PE
PR
RJRN
RO
RS
SC SE
SP
TO
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
0% 100% 200% 300% 400% 500% 600%

Source: calculations by the authors and Aliceweb (2007).