Article

Matheus Ruschel; Daniel Alvarez Firmino

DOI: 10.22167/2675-6528-202602576

Article derived from a Final Course Work (TCC), with content based on the student’s original work and adapted to the editorial format of the Revista E&S with the support of the ResumeAI tool, an artificial intelligence solution developed by the Pecege Institute for textual synthesis and organization.

Summary

The Bolsa Família Program (PBF) represents one of the most relevant social protection policies globally, which justifies the investigation of its effects given the sharp and heterogeneous regional inequalities in Brazil. The study evaluated the socioeconomic impacts of the program’s transfers on health, education, and the labor market in Brazilian municipalities, from 2004 to 2019. To this end, the K-means clustering technique was applied for territorial segmentation, and econometric panel data models with fixed effects were estimated, both at the national level and segregated by clusters. The results revealed the counter-cyclical nature of the PBF in the labor market, with a negative relationship between transfers and formal employment ties in four out of the five clusters, suggesting that resources were more intense where the formal market failed. In health, the program was associated with a significant reduction in preventable infant mortality in municipalities with greater socioeconomic balance, but it showed no detectable effect in clusters with greater structural precariousness, indicating that income transfer is necessary but insufficient without functional health infrastructure. In education, the analysis by sub-periods showed a progressive attenuation of the national coefficient, reflecting the convergence of enrollment rates to a level of high temporal inertia. It was concluded that the PBF fulfilled its objective of social protection in a counter-cyclical and territorially focused manner, but its capacity to transform structural indicators depended on synergy with investments in public infrastructure.

Keywords: Bolsa Família; Infant Mortality; Brazilian Municipalities; Data Panel; Public Policy.

1. Introduction

Income inequality in Brazil is a structural phenomenon, deeply rooted in historical and socioeconomic disparities between its diverse regions and social groups. In this context, income transfer policies, such as the Bolsa Família Program, emerge as crucial social protection instruments, demonstrating significant impacts on poverty reduction and the improvement of social indicators. Previous studies highlight the program’s relevance in poverty reduction (Soares et al., 2010; Medeiros, Britto, Soares, 2008), in promoting social mobility through increased schooling and future income potential of beneficiaries (Glewwe, Kassouf, 2012), and in contributing to the fall in the Gini index and advances in reducing infant mortality (Silva et al., 2021).

However, the effectiveness of these policies is not manifested homogeneously throughout the national territory. The results and scope of the Bolsa Família Program are conditioned by local structural conditions, the availability of public services, and the socioeconomic composition of municipalities. This territorial variability suggests that the capacity of a public policy to mitigate inequalities is intrinsically linked to the context in which it is implemented. The existing literature, however, still addresses the dimension of heterogeneity in the response of different municipal profiles to income transfer programs in a limited way.

Given this gap, the present study is justified by the need for a deeper understanding of how territorial specificities modulate the impacts of the Bolsa Família Program. To this end, an innovative methodological approach is adopted that integrates data science techniques and econometric methods. The research begins with the application of K-means clustering, using demographic, educational, economic, and social assistance variables to segment Brazilian municipalities into groups with similar socioeconomic patterns. This stage is fundamental to recognizing and categorizing the country’s structural diversity.

Next, econometric panel data modeling with fixed effects (Wooldridge, 2010; Baltagi, 2021; Stock, Watson, 2019) is employed to estimate the intramunicipal impact of the Bolsa Família Program. This technique allows for controlling time-invariant characteristics that could bias the estimates, isolating the effect of public policies. By articulating machine learning and econometrics, this work seeks to contribute to a more precise and contextualized understanding of the effects of income transfer policies. The main objective is to analyze the heterogeneity of the effects of the Bolsa Família Program on education, health, and labor market indicators, considering the socioeconomic specificities of Brazilian municipalities, in order to offer empirical subsidies to improve territorial targeting and formulate more effective public policies.

2. Material and Methods

This study was characterized as a quantitative research, with a methodological approach that integrated Data Science techniques and econometric methods. The unit of analysis comprised all Brazilian municipalities, covering all regions and states of the country, with the objective of preserving the heterogeneity and representativeness of the data. The temporal analysis was delimited to the period from 2004 to 2019 for econometric modeling, while the clustering stage used more updated data, referring to the year 2023.

Data collection was carried out from secondary sources of public knowledge. Information on Population and Gross Domestic Product (GDP) per capita was obtained from the Brazilian Institute of Geography and Statistics (IBGE). Infant mortality data were extracted from DATASUS, and formal labor ties from the Annual Social Information Report (RAIS). The Anísio Teixeira National Institute for Educational Studies and Research (INEP) provided the School Census data for basic education. Finally, the Ministry of Development and Social Assistance, Family and Combating Hunger made available data on the Bolsa Família, Auxílio Brasil, and Auxílio Emergencial programs.

The variables used were treated to ensure comparability and consistency. Municipal GDP per capita and income transfer values (Bolsa Família, Auxílio Brasil, and Auxílio Emergencial) were deflated to current values as of December 2023, using the National Broad Consumer Price Index (IPCA). The enrollment rate in basic education considered students aged 4 to 17 years, and the infant mortality rate included deaths of children aged 0 to 4 years from preventable causes, according to the Brazilian List of Preventable Deaths from the Ministry of Health, calculated as a five-year average to mitigate volatility. The proportion of active labor ties was calculated in relation to the population aged 16 to 64 years.

In the data treatment, missing values were kept as zero, justified by the nature of the cases, such as municipalities emancipated after 2004 or the effective absence of occurrences. The resulting data panel was unbalanced, which is compatible with the fixed-effects estimator used. No outliers were removed in the regression models. For the clustering step, the variables were standardized via Z-score, ensuring that distinct scales contributed equitably to the calculation of Euclidean distances and reducing the weight of atypical values.

The territorial segmentation of municipalities was carried out using the K-means clustering technique, an unsupervised learning algorithm. The optimal number of clusters was determined with the aid of the Elbow method. To interpret the clusters, the Random Forest Classifier algorithm was used to measure the relative contribution of each variable in separating the groups (Feature Importance), and the centroid vector of each cluster was calculated, extracting the average of the standardized scores of the original variables. A structural stability test of the clustering, using Spearman’s coefficient between 2004 and 2023, confirmed the persistence of the segmentation variables, such as formal labor market (p=0.82), GDP per capita (p=0.88), and Bolsa Família transfers (p=0.70).

To measure the association of income transfer programs on health, employment, and education indicators, the Fixed Effects Panel Data Regression technique was employed (Wooldridge, 2010; Baltagi, 2021; Stock; Watson, 2019). This modeling allowed for tracking the same cross-sectional units (municipalities) over time, controlling for unobserved heterogeneity and time-invariant characteristics. The empirical strategy was based on the estimation of a fixed effects panel data model, where the dependent variable (Yit) represented the analyzed social indicator, the variable of interest (BFit) the Bolsa Família transfers, and Xity the set of control variables, in addition to municipal (αi) and temporal (λt) fixed effects and the error term (εit).

The validity of the statistical inference was ensured by the use of a robust covariance matrix with clustered standard errors at the municipal level, addressing potential issues of heteroscedasticity and serial autocorrelation. The fixed-effects estimator performed an internal transformation on the data, known as demeaning, by subtracting the temporal mean of each variable per unit, which allowed for the isolation of the programs’ impact based on within-municipality variation. Estimations were processed separately for each cluster and also at the national level. Robustness tests with temporal lag (t-1) were conducted to verify the persistence of the effects, confirming the stability of the coefficients in sign and statistical significance for the three outcomes analyzed.

3. Results and Discussion

The analysis of the results began with the segmentation of Brazilian municipalities, using the K-means clustering technique on socioeconomic indicators from 2023. This approach was crucial for identifying distinct territorial profiles and understanding the country’s structural heterogeneity, which conditions the effectiveness of income transfer policies. The Elbow method, although initially suggesting three clusters, led to the choice of five groupings, as this configuration provided a more detailed segregation of vulnerability dimensions, allowing the isolation of a cluster with critical indicators of structural mortality that would otherwise be aggregated into more generic groups. This granularity is fundamental for evaluating the specific effects of the Bolsa Família Program in various contexts of public need.

The importance of the variables in the formation of clusters was evaluated by the Random Forest Classifier algorithm, revealing that all indicators (GDP per capita, Bolsa Família transfers, basic education enrollments, formal employment, and death rate) contributed significantly to the composition of the groupings. This high relevance of the variables demonstrates the study’s capacity to differentiate Brazilian regions based on distinct and significant socioeconomic aspects. The qualitative interpretation of the clusters, based on Z-score averages, transformed the mathematical structure of the model into concrete socioeconomic typologies, highlighting the profound disparities that characterize the national territory.

The geographic distribution of the five statistically robust clusters corroborates the literature on Brazilian regional disparities. A clear spatial segmentation was observed, with the predominance of Clusters 0 and 2 in the North-Northeast axis and the concentration of Clusters 3 and 4 in the Center-South. This geographic proximity acts as a relevant determinant in the socioeconomic homogeneity of municipalities. This spatial configuration validates the subsequent application of the segmented regression model, as it allows capturing how income transfers impact distinct regional realities, offering a solid basis for the analysis of the program’s heterogeneous effects.

Clustering

Cluster 0, named “Income Vulnerability”, concentrates 32.85% of the municipalities and 18.28% of the population, predominantly in the North and Northeast. It is characterized by the highest Bolsa Família transfer (1.06) and the lowest formal employment (-0.76), in addition to a low per capita GDP (-0.55). In this group, the Bolsa Família Program acts as a central gear of the local economy, indicating a high dependence on income transfer for family subsistence and economic dynamization, in a context of an incipient formal labor market.

Cluster 1, “High Income Concentration”, comprises only 1.63% of municipalities and 1.47% of the population, dispersed throughout Brazil. This group represents the economic elite, with the highest GDP per capita (5.19) and the highest formalization (RAIS: 2.91). The value of Bolsa Família is negative (-0.62), suggesting that these municipalities depend less on income transfers. The low number of observations and high internal homogeneity limit the model’s ability to capture significant variations, as social indicators already operate at high and stable levels.

Cluster 2, “Health Crisis”, encompasses 12.08% of municipalities and 5.94% of the population, concentrated in the North and Northeast. This group presents the highest point of deaths (1.62) and an above-average education rate (0.67), but with low GDP. The analysis suggests that these municipalities face significant structural bottlenecks in sanitation or public health, where isolated income transfer is insufficient to promote detectable improvements in infant mortality indicators, even with relatively better educational levels.

Cluster 3, “Developing Interior”, predominates in inland regions of the South, Southeast, and Center-West, representing 32.64% of municipalities and 25.03% of the population. It is characterized by balanced indicators, without extremes of poverty or wealth. The development trajectory of these municipalities seems to depend mainly on the dynamization of the formal labor market and local economic growth, with the Bolsa Família Program acting as a supplement to income during periods of low employability, but not as the main driver of development.

Finally, Cluster 4, “Social Efficiency”, concentrates 20.79% of the municipalities and 49.28% of the population, located in economic hubs in the South, Southeast, and Center-West. This is the most balanced cluster in terms of socioeconomic indicators, with high labor formalization, good educational indicators, and lower dependence on transfers. The balanced development between human capital and economic dynamism is the distinctive feature of this group, where the Bolsa Família Program operates as a stabilizer during economic downturns, but not as a primary driver of structural changes.

Health: Infant Mortality Avoidable Deaths

The investigation of the relationship between the transfers from the Bolsa Família Program and the municipal infant mortality rate due to preventable causes (pcapita_obitos) revealed a non-uniform impact, strongly conditioned by the infrastructure and socioeconomic profile of each grouping. At the national level, the Bolsa Família coefficient was negative and significant (β = -2.56e-07, t = -2.12), indicating an association with the reduction of preventable infant mortality in the period from 2004 to 2019. This finding suggests a trend of a reducing impact of the program, which becomes more refined and confirmed when the sample is segregated by clusters.

In Cluster 0 (Income Vulnerability), the coefficient for Bolsa Família was positive and significant (β = 6.08e-07, t = 3.16), associating the program with increased mortality in the most vulnerable municipalities. This result is consistent with the hypothesis that the effectiveness of income transfers in health depends on the availability of functional public services, although causal confirmation requires more granular data on hospital infrastructure and sanitation. The lack of significance of other controls, such as formal employment, GDP, and education, reinforces that the mortality dynamics in these municipalities are determined by structural factors not directly captured by the model, suggesting that income transfer is necessary but insufficient without functional health infrastructure.

In Clusters 1 (High Income Concentration) and 2 (Health Crisis), the coefficients were positive but not significant (β = 1.24e-06, t = 0.94 and β = 3.28e-07, t = 0.88, respectively). In Cluster 1, the model does not explain mortality, as these wealthy municipalities already have structurally low and little variable infant mortality. In Cluster 2, with the highest mortality rates in the sample, Bolsa Família did not manage to move the indicator in a detectable way, reinforcing the thesis that, where the health crisis is structural, income transfer alone is insufficient to mitigate public health bottlenecks.

In Cluster 3 (Developing Interior), the coefficient was negative but not significant (β = -5.14e-07, t = -0.69). Despite the expected direction of mortality reduction, there was insufficient statistical evidence to confirm the effect. On the other hand, in Cluster 4 (Social Efficiency), the coefficient was negative and significant (β = -1.11e-06, t = -2.15), representing the most robust result in the expected direction. In these municipalities with greater socioeconomic balance, more functional public service networks, and a formalized labor market, Bolsa Família is associated with lower rates of avoidable infant mortality, which is consistent with the mechanism that transferred income produces better health outcomes where there is functional health infrastructure.

The results, therefore, suggest that the effectiveness of Bolsa Família in the health dimension is conditioned by the presence of functional health infrastructure. In the municipalities of Cluster 4, where this infrastructure is more consolidated, income transfer is associated with lower rates of avoidable infant mortality, indicating that the program finds conditions for the additional income to be converted into effective access to services. In Clusters 0 and 2, where structural precariousness is more pronounced, this channel appears to be blocked. This pattern suggests that territorial targeting policies that combine income transfer with strengthening primary care may be more effective than isolated financial transfers, especially in contexts of vulnerability and health crisis.

Education: Basic Education Enrollment

The educational dimension, with minimum school attendance as a central conditionality of the Bolsa Família Program, was examined to understand the influence of financial transfers on the per capita enrollment rate. The estimated coefficients for education revealed a pattern of high heterogeneity among clusters, with effects mainly detectable in the national aggregated model and in sub-period analyses. In the aggregated model for Brazil, a negative and highly significant coefficient was observed (β = -2.81e-05, t = -4.63). However, this negative sign does not indicate a reduction in enrollments, but rather an anti-cyclical correlation that demands temporal decomposition for its interpretation.

The sub-period analysis revealed that, between 2004 and 2008, a phase of accelerated program expansion, the negative and significant coefficient (β = -6.82e-05, p = 0.000) reflects an anti-cyclical correlation. Transfers grew preferentially in municipalities with greater room for school inclusion, resulting in a negative association that does not denote deterioration, but rather adequate targeting. In the period from 2009 to 2014, the coefficient maintained its negative sign with reduced magnitude (β = -4.27e-05, p = 0.000), suggesting that, as the program reached maturity, municipalities with lower educational performance converged to higher levels, progressively attenuating the anti-cyclical relationship.

From 2015 to 2019, the coefficient lost statistical significance (β = -8.89e-06, p = 0.238), indicating that the enrollment variable reached a plateau of high temporal inertia. This was confirmed by the almost complete overlap of the 2004 and 2019 distributions, indicating that, at this stage, the program sustains enrollments but does not marginally move them within municipalities. This result is consistent with the hypothesis of convergence to a structural enrollment ceiling. The Annual Social Information Report (RAIS) and per capita GDP were highly significant (t = 6.63 and t = 3.95, respectively), confirming the argument that minimum income, municipal investment capacity, and demand for qualification act jointly in basic education.

In Clusters 0 (Income Vulnerability), 1 (High Income Concentration), 2 (Sanitary Crisis), 3 (Developing Interior), and 4 (Social Efficiency), the Bolsa Família coefficients were mostly non-significant. In Cluster 0, the program does not move enrollment in a detectable way, as the basic education rate was already universally high since 2004, with GDP per capita being the main explainer. In Clusters 1 and 4, the enrollment rate already operates at structurally high levels and with low temporal variation, and economic dynamism and labor formalization are the central determinants of schooling.

In Cluster 2, the program does not move enrollment, which is consistent with the profile of structural bottleneck, not income. In Cluster 3, RAIS is the most relevant control, reinforcing that the demand of the formal labor market is the main driver of enrollment. In summary, in the educational dimension, the results indicate that the program sustains school attendance in contexts where enrollment rates still showed room for growth, especially in the expansion phase. As enrollments converge to levels close to the structural ceiling, the marginal role of Bolsa Família becomes less detectable, suggesting that the public policy challenge shifts from inclusion to quality and attendance, demanding instruments complementary to financial transfer.

Work: Formal Employment Relationships

The investigation of the interaction between income transfers and the dynamics of the formal labor market in Brazilian municipalities revealed that the Bolsa Família Program has an anticyclical nature and precise targeting, reacting inversely to formal hiring cycles. In the aggregate model for Brazil, the coefficient was negative and highly significant (β = -5.66e-05, t = -12.93). This result indicates that the volume of Bolsa Família resources is greater in periods and locations where the supply of formal employment declines, acting as a social safety net and reaching families in contexts of lower access to the formal labor market with greater intensity.

In Clusters 2 (Health Crisis), 3 (Developing Interior), and 4 (Social Efficiency), the coefficients were negative and highly significant (β = -3.44e-05, t = -3.96; β = -6.54e-05, t = -5.74; and β = -1.00e-04, t = -3.16, respectively). These results are robust and consistent with the counter-cyclical nature of the program, indicating that Bolsa Família maintains its social safety net function even in territories with different development profiles. Basic education and GDP per capita emerged as significant and positive controls in multiple clusters, reinforcing that economic dynamism and qualification are structural drivers of formalization, with Bolsa Família operating as a stabilizer during periods of contraction.

In Cluster 0 (Income Vulnerability), the coefficient was positive and not significant (β = 6.27e-06, t = 1.52). In the most vulnerable municipalities, the formal labor market is incipient, and variations in the program’s transfer do not produce a detectable movement in the indicator. GDP per capita (t = 6.16) is the most relevant control, suggesting that local economic dynamism is the main driver of formalization in these territories. In Cluster 1 (High Income Concentration), the coefficient was negative and not significant (β = -3.00e-04, t = -1.29), with basic education (t = 2.46) being the most relevant control, indicating that qualification is the main determinant of formalization in more sophisticated and selective labor markets.

The negative relationship observed between the amount transferred by the Bolsa Família Program and the level of formal employment allows for two main interpretations. From an economic perspective, the program may have an anti-cyclical character, acting as an automatic stabilizer in municipalities where the labor market is more fragile. However, it is necessary to recognize the possibility of reverse causality, where increased unemployment and a fall in labor income expand the number of families eligible for the program, raising the volume of transfers. Although the fixed-effects model controls for constant municipal characteristics, the dynamic nature of this relationship suggests that the coefficients should be interpreted as robust correlations.

A relevant cross-cutting finding is that basic education and per capita GDP emerge as significant controls in the labor market models across multiple clusters, suggesting the existence of a long-term impact chain, where human capital fuels formalization, which in turn mitigates vulnerability. This impact chain, however, is only partially captured by the temporal horizon and municipal aggregation of this analysis. Additional robustness tests, replacing contemporary Bolsa Família transfers with a one-period lag (t-1), confirmed the stability of the estimates, with all three outcomes maintaining a negative sign and statistical significance, suggesting that the identified effects are not sensitive to the adopted temporal specification.

In summary, the study revealed that the impact of the Bolsa Família Program on Brazilian municipal development is heterogeneous and conditioned by the structural characteristics of each locality. Aggregated national-level metrics conceal distinct, and in some cases opposite, territorial dynamics of public policy effectiveness. The program’s counter-cyclical nature in the labor market was the most robust finding, confirming that transfers tend to be higher where the formal market fails. In health, the program’s effectiveness depended on public service infrastructure, with a reduction in infant mortality in more balanced municipalities, but not in contexts of structural precariousness. In education, the program sustained enrollment, but its marginal role diminished as rates converged towards a structural ceiling, shifting the challenge to school quality and retention.

4. Conclusion

This study sought to analyze the heterogeneity of the effects of the Bolsa Família Program on education, health, and labor market indicators, considering the socioeconomic specificities of Brazilian municipalities. It was found that the program had a robust counter-cyclical nature in the labor market, with transfers being higher in periods and locations where formal employment opportunities declined, acting as a social safety net in four out of the five identified clusters. In the health dimension, it was observed that the program was associated with a significant reduction in preventable infant mortality in municipalities with greater socioeconomic balance, where public service infrastructure was more functional. However, in clusters with greater structural precariousness, the effect was not detectable or showed a positive association, indicating that income transfer, although necessary, proved insufficient without synergy with investments in health infrastructure. In education, the analysis by sub-periods revealed a progressive attenuation of the national coefficient, reflecting the convergence of enrollment rates to a level of high temporal inertia, suggesting that the program sustained enrollment, but its marginal role diminished as rates approached a structural ceiling. The methodological approach, which integrated K-means clustering with econometric panel data models, proved essential for unraveling the territorially distinct and, in some cases, opposing dynamics of public policy effectiveness, offering empirical subsidies for improving territorial targeting and the formulation of more effective public policies.

Despite the analytical robustness, the study presented limitations arising from the nature of the variables used, such as the absence of more granular data on health and sanitation infrastructure, which prevented the causal confirmation of some mechanisms. Additionally, the aggregation of different income transfer programs and the high temporal inertia of the basic education enrollment variable limited the ability to capture long-term effects on human capital. For future studies, it is suggested to explore strict causality mechanisms through instrumental variables or quasi-experimental designs, such as difference-in-differences with multiple periods. It is also recommended to use more granular data, including beneficiary microdata and detailed information on public service infrastructure, as well as to apply spatial econometrics models to capture territorial interdependencies. These advances would allow for an even deeper understanding of the effectiveness of the Bolsa Família Program and its synergy with other public policies, especially in contexts of vulnerability and health crisis.

Bibliographic References

Baltagi, B.H. 2021. Econometric Analysis of Panel Data. 6ed. Springer Nature, [S.I.].

Glewwe, P.; Kassouf, A.L. 2012. The impact of the Bolsa Escola/Família conditional cash transfer program on enrollment, dropout rates and grade promotion in Brazil. Journal of Development Economics 97(2): 505-517.

Medeiros, M.; Britto, T.; Soares, F. 2008. Transferência de renda no Brasil: impactos sobre a desigualdade e a pobreza. Instituto de Pesquisa Econômica Aplicada [IPEA], Brasília, DF, Brasil. (Texto para Discussão n. 1250).

Silva, M.O.S.; Yazbek, M.C.; Di Giovanni, G. 2021. A política social brasileira no século XXI: a prevalência dos programas de transferência de renda. Cortez, São Paulo, SP, Brasil.

Soares, F.V.; Ribas, R.P.; Osório, R.G. 2010. Evaluating the impact of Brazil’s Bolsa Família: cash transfer programmes in comparative perspective. Latin American Research Review 45(2): 173-190.

Stock, J.H.; Watson, M.W. 2019. Introduction to Econometrics. 4ed. Pearson, [S.I.].

Wooldridge, J.M. 2010. Econometric Analysis of Cross Section and Panel Data. 2ed. MIT Press, Cambridge, MA, USA.

Article originating from the Course Conclusion Work of the Specialization in Data Science and Analytics of the MBA USP/Esalq

To learn more about the course, click here and access the MBX Academy platform