Political Turnover and Public Health Provision in Brazilian Municipalities

2.1 Municipal Political Scenery

Brazil is a federative democratic republic, characterized by having a multiparty system. Its territory is divided into 26 states and a federal district. The 5,570 municipalities, belonging to the states, form the local layer of public administration in the country. This local layer is the closest to the citizens, being made up of mayors and councilors. The mayors, heads of the municipal executive branch, are elected every 4 years in a direct democratic manner. In these elections, the councilors of each administration are also elected, forming the legislative body of the municipality.

In municipalities with a population below 200,000 inhabitants, mayors are elected according to the majority rule (absolute majority of votes), while municipalities with a population above this threshold follow the rule of the plurality of votes, and victory may occur in the second round of an election if no candidate achieves victory in the first round. Mayors are elected to 4-year terms, and since 1998, their terms may extend for four more years through re-election.^[3]

2.2 Municipal Public Health Provision and Management

Created from the 1988 Constitution, the SUS is based on three basic principles: universality, integrality, and equity. It was conceived following a hierarchical and decentralized organization, in which the primary health care is destined to the municipalities and the most complex procedures are responsibilities of the states and the federal government. The 1988 Constitution therefore established organizational guidelines for the SUS, which have implications for decisions and implementations of public health policies. Among these guidelines, one aspect is extremely important for the functioning of the public health system: the definition of the role and attributions of SUS managers.

As the Ministry of Health points out (2019), it is up to the three spheres of government to jointly define mechanisms for the evaluation and control of health services, as well as to manage and apply budgetary and financial resources, to develop short and medium-term policies, and to promote their articulation, among other functions.

In this sense, Machado et al. (2011) emphasize that in the three spheres of government, decisions about health policies involve a large number of services and people. At the municipal level, health secretariats (or equivalent bodies) are administered by SUS managers – a term used to refer to the minister and state and municipal health secretaries. In other words, the SUS manager is the health authority in each sphere of government. The position has a very important political meaning, since each manager is appointed by the head of the executive branch.

In addition, according to Machado et al. (2011), the main duties of the SUS manager can be divided into four categories: formulation of public policies; financing; regulation, coordination, control, and evaluation; and direct execution of services. At the municipal level, it is important to note that it is up to the SUS manager to identify the problems and define the municipality’s priorities and carry out the planning of necessary actions for public health. It is also the responsibility of the municipal SUS manager to ensure the application of resources and make investments, and to articulate with other municipalities, guarantee the direct provision of assistance services, manage the basic health units and, consequently, evaluate the results of adopted municipal policies.

2.3 The Link Between Incumbent Performance and the Party Brand

Political parties play a crucial role in the establishment of stable democratic administrations. Their primary function is to gather and consolidate the diverse preferences of individuals, thereby decreasing the informational burdens on voters and facilitating the organization of legislative bodies (Heller & Mershon, 2009). Desposato (2006) asserts that while political parties are essential for a functioning democracy, they also serve the interests of politicians by offering mobilization efforts, financial support for campaigns, and a recognizable political identity upon which they can depend.

Concurrently, a substantial body of academic literature on distributive politics indicates that political parties derive national advantages from their local office control, and conversely, experience disadvantages when they lack such control. Emerging research indicates that the presence of local incumbent contributes to enhancing the perception of their affiliated parties in terms of governance matters (Holland, 2016; Lucardi, 2016). However, it is important to note that this notion is predicated upon one underlying assumption.

The underlying premise posits that the combined influence of party labels and the performance of incumbents shapes the effects of incumbency. However, in contexts where particularism and weak party discipline are prevalent, there is limited expectation for citizens or politicians to establish associations among co-partisans across different tiers of government (Rodden & Wibbels, 2011). In instances where party labels hold significant sway, a favorable performance at the national level is expected to bolster electoral support for co-partisan candidates in local elections. Conversely, when party labels lack strength, voters do not establish connections between local and national politicians, resulting in office control having no discernible impact, either positive or negative, on the vote shares of other candidates running under the same party label.^[4]

Looking at the Brazilian case, the nature of the country’s party system has been hotly debated by comparativists. Previously, parties were commonly described as feeble, centered around individual personalities, and primarily driven by practices of patronage and pork-barrel politics (Desposato, 2006; Samuels, 1999). However, this negative perspective has been convincingly challenged by recent studies that highlight signs of party consolidation in areas such as legislative decision-making within the Congress (Figueiredo & Limongi, 2000), and the shaping of voters’ attitudes on national matters (Samuels & Zucco, 2014).

Despite suggestive indications of party system institutionalization at the national and state levels, there is currently no evidence to suggest that this hypothetical trend extends to the realm of municipal politics. This is further emphasized by the frailty of local political parties. Mayors often run personalistic campaigns (Novaes, 2014), and local coalitions are large and ideologically diverse, diminishing the impact of national party reputations in municipal elections (Lavareda & Telles, 2015).

Such findings can be confirmed by examining our data on the results of the 2008 and 2012 municipal elections. During our analysis period, the incumbent national party from 2003 to 2016 was the Workers’ Party (PT). The strength of the PT’s party label is well-established in the literature, as the party has a coherent program of social change, a disciplined membership (both in Congress and at the local level), deep roots in civil society (Hunter, 2010), and is the only party in Brazil with a large supporter base, as shown by Samuels and Zucco (2014).

However, the results of the 2008 and 2012 municipal elections do not reveal a partisan dominance between national and subnational governments. In 2008, despite the PT being the titular national party, 21% of the elected mayors were affiliated with the PMDB (Brazilian Democratic Movement Party), 14% with the PSDB (Brazilian Party for Social Democracy), and only 10% with the PT. Restricting the sample to only close races, the results seem similar, with 21% of the winning mayors being affiliated with the PMDB, 16.8% with the PSDB, and 11.5% with the PT. In the 2012 municipal elections, these results were maintained, as 18.6% of the elected mayors were affiliated with the PMDB, 12% with the PSDB, and only 11.8% with the PT. In the close elections of the same year, 23.8% of the winning mayors were affiliated with the PMDB, 10% with the PSDB, and only 12.9% with the PT. It is worth noting that in the 2012 elections, the PT was already in its third term as the titular national party.

Lee et al. (2004), Erikson (1971), and Groseclose (2001) highlight the inherent advantage enjoyed by incumbent parties and candidates. Novaes and Schiumerini (2022) reveal that the advantage or disadvantage of incumbency for Brazilian mayors depends on several factors, including differences in access to fiscal resources that systematically shape the performance of incumbents. Conversely, our data from the 2008 and 2012 municipal elections do not demonstrate a clear predominance of re-election or political turnover. For instance, in the closely contested 2008 elections, the re-election rate stood at 56% (while the turnover rate was 44%). Similarly, in the 2012 elections, the re-election rate was 48% (with a turnover rate of 52% that year). Given the small margin of difference between these results, it is not possible to ascertain whether there was a prevailing pattern of re-election or political rotation in the municipal elections.

3.1 Election Data

In this work, we used data from municipal elections that took place in 2004,^[5] 2008, and 2012, obtained through the Superior Electoral Court (Tribunal Superior Eleitoral, TSE). With these, it was possible to calculate the incumbent’s voting margin for the years 2008 and 2012, defined for each year as the share of votes received by the incumbent party minus the share of votes received by its strongest opponent. Like Lee (2008), we consider only the cases in which the incumbent was the most voted or second, since candidates outside these positions generally obtain a small portion of the votes.

From this, it was possible to build our running variable of the regression discontinuity model – the incumbent’s narrow margin of votes. Here, we consider an election with a narrow margin one in which the incumbent almost lost or almost won the election with a margin of less than 5% or, in absolute terms, 10%. The idea is that in competitive elections, the result of winner or loser is random, as the candidates are similar.

3.2 Municipal Health Indicators

The data referring to municipal health indicators were obtained through the SUS Department of Informatics (DATASUS), for the years after the 2008 and 2012 municipal elections. More specifically, we used data from the indicators for the years 2009 and 2011 and for 2013 and 2015 (1 and 3 years after the election of 2008 and 2012). Table 1 shows the indicators that will be tested in this work, as well as their respective definitions.

3.3 Municipal Characteristics

Finally, the data used to form our covariate vector were obtained through the National Treasury Secretariat. We decided to use only budgetary variables – tax revenue per capita and expenditure per capita on health – since these are directly linked to the execution and provision of public services. Furthermore, as Ferraz and Finan (2011) point out, the use of such variables controls possible differences among the municipalities’ budgets.

3.4 The Regression Discontinuity Design

The work of Thistlethwaite and Campbell (1960) introduced the idea of the method that we now know as regression discontinuous design (RDD). As Imbens and Lemieux (2008) point out, the basic idea behind the RDD is that whether an individual is treated or not is determined totally, or partly, by the value of a covariate, which is on both sides of a fixed limit. The covariate may be associated with potential results, but this association is supposed to be mild. Thus, any discontinuity of the result due to the covariate is interpreted as evidence of a causal effect of the treatment.

Lee (2008) points out that for an RDD in the sharp case, it is necessary to assume two hypotheses. The first hypothesis concerns continuity: The conditional regression functions of the results of treatment and control in X _i must be continuous. Formally:

where Y _i (0) and Y _i (1) are the potential results and X is the so-called running variable, a covariate that indicates whether the individual will receive treatment or not.

Imbens and Lemieux (2008) point out that this condition assumes that the conditional distribution function is smooth in the covariate and can be called the conditional distribution function. Let F _y be the conditional distribution function of Y _(t) , given X, where T indicates the treatment. The general version of the hypothesis therefore becomes

are continuous at x for all y.

The continuity hypothesis makes it clear, according to Lee (2008), that the credibility of impact estimates via RDD depends on whether the average result of individuals just below the cutoff point identifies the true counterfactual for those individuals who are just above the discontinuity. In other words, it is the continuity hypothesis that guarantees that we are identifying the causal effect of a treatment.

The second important hypothesis to be assumed is that of local ignorability. If around the point X _i = c the condition of being treated or has not been defined at random, we have local ignorability. Formally:

According to Imbens and Lemieux (2008), this assumption is trivial because, depending on the covariables, there is no change in treatment. In other words, an individual cannot manipulate a covariate in order to be treated or not.

Under the two hypotheses presented earlier, the only explanation for the discontinuity in Y _i in c is that an individual has received treatment, so that we can interpret the jump of T _i at point c as the causal effect of the treatment. Thus, according to Hahn et al. (2001) and Angrist and Pischke (2009), we have T ⁺ = 1 and T ⁻ = 0, and the average local treatment effect can be expressed by

That is, the causal effect of the treatment is given by the difference in results between individuals who are located above the cutoff point and below this point.

3.5 Empirical Strategy

We want to examine whether the fact that a municipality has experienced party turnover has any effect on the health indicators of these municipalities. Following on the work of Akhtari et al. (2022), we will make a comparison between municipalities in which the incumbent party almost lost the election – and, therefore, remained in power – and municipalities in which the incumbent party was almost re-elected – and, in this way, there was party rotation in the municipal leadership. For that, we will make use of a sharp RDD.

To work with the concept of “almost lost” or “almost re-elected,” we will use a variable referring to the incumbent’s narrow margin of votes. The incumbent’s voting margin refers to the portion of votes that the incumbent party received in a given election, considering the total useful votes received in its respective municipality. By restricting this margin to a limit of [−5%; 5%], we have what we call a narrow margin.^[6]

Furthermore, it is important to highlight that as Lee’s (2008) study, we will work only with municipalities where the incumbent party finished first or second in the election, since the share of votes received by third or other places tends not to be significant.

Following Imbens and Lemieux (2008), Brollo and Troiano (2016), and Akhtari et al. (2022), we will work with a linear regression specification, which is restricted to municipalities that are in the range of vote margins specified.

Thus, our specification can be expressed as follows:

where Y _m,t+1 are the outcome variables (health indicators) of municipality m, measured 1 year after the elections (we are working with a pool of data from the 2008 and 2012 elections (period t), so that the period t + 1 corresponds to the data pool for the years 2009 and 2013). The result variable Y _m,t+1, in this case, represents the local average effects (LATE) at the cutoff point c = 0. Our running variable is the incumbent’s margin of votes whereas our treatment variable is the incumbent’s margin of votes less than 0, which indicates that the incumbent lost his attempt at re-election and, therefore, the municipality underwent Political turnover. In addition, standard errors are grouped at the municipal level. In order to verify whether the possible effects of political turnover on municipal indicators are transient or permanent, we will estimate the same model above but consider the 3 years after the elections. In this case, our result variables will be given by Y _m,t+3 (corresponding to the data pool for the years 2011 and 2015).

The intuition of our empirical strategy is that municipalities in which the incumbent party lost the election by a narrow margin of votes form a good counterfactual for municipalities in which the incumbent party remained in power (won the election with an advantage of at most 5%). Therefore, according to Lee (2008) and Brollo and Troiano (2016), the assumption stems from the fact that in close races, the result is defined by random factors, so that the incumbent’s probabilities of being re-elected or losing the election are the same.

Thus, through our empirical efforts, it will be possible to discover whether Political turnover (which occurs when the incumbent party loses the election and another party takes over the municipal leadership) has an effect on municipal health indicators and whether this effect is beneficial or deteriorates public health management in the municipalities.

4.1 Descriptive analysis of the data

In the years 2008–2012, Brazil had 5,565 municipalities in its territory. Of these, we excluded the municipalities in which there was a possibility of a second round of elections (i.e., with a population of 200,000 or more) or went through supplementary elections, the municipalities went through supplementary elections, or both. Thereafter, for each year, we restricted our sample to the municipalities in which the incumbent party was trying to remain in power and finished first or second in its respective elections.

As a result, our sample was reduced to 2,820 municipalities for 2008 and 2,675 for 2012, totaling 5,495 municipalities in the total of our sample. The left part of Figure 1 presents the histogram of the incumbent’s vote margin constructed from the steps mentioned previously.

Figure 1

Incumbent’s vote margin and incumbent’s narrow vote margin – Histogram – 2008 and 2012. Source: TSE, data worked by the research.

Municipalities that are left to zero in the histogram are those in which the incumbent party lost the election (in 2008 or 2012), and there was Political turnover. The right side of the histogram represents the municipalities in which the incumbent party remained in power. Figure 1 shows that the distribution of the incumbent’s voting margin seems normally close to zero, indicating symmetry between the two sides of the distribution.

When we consider only the incumbent’s narrow vote margin, in which the party won or lost with a margin of no more than 5%, our sample is reduced to 686 municipalities in 2008 and 711 municipalities in 2012, totaling 1,397 municipalities. The right part of Figure 1 shows the histogram of this sample section.

Continuing, Table 2 presents descriptive statistics of the two groups of municipalities with which we deal.

Table 2 shows that within our sample, the municipalities that suffered Political turnover and those that kept the parties in power were quite similar to each other.

Finally, we performed the test by Cattaneo et al. (2018) for our discontinuity, in order to check if there is manipulation in our running variable. The results are shown in Table 3.

The results show that there is no evidence of manipulation in our variable, since the values found for the t-statistic lead us not to reject the null hypothesis of non-manipulation. Figure A1, in the Appendix, shows this result graphically.^[7] From that result, we proceed to our main objective.

4.2 Political Turnover and Health Outcomes

Next, we present graphically the results of the regression discontinuity estimations carried out in order to verify how political turnover affects the provision of public health in the municipalities. Therefore, we estimated equation (1) using the incumbent’s narrow vote margin as the model’s running variable. The graphs that compose the figures presented below were generated using the uniform selection method of bins. The results can also be seen in Tables A2 and A3, in the Appendix.^[8]

In each figure,^[9] presented from this point on, the right side (incumbent’s narrow vote margin <0) refers to the municipalities in which the incumbent was almost re-elected – and there was political turnover – while the left side represents the municipalities in which the party almost lost but remained in power.

First, Figure 2 presents the results for SUS workers, visits from the PSF, and coverage of primary care for the period immediately after the elections (t + 1) and, in a measure of the persistence of the effects, for the period of 3 years after the elections (t + 3). When we analyzed 1 year after the elections (left column), the three variables were found to be significant and minor in municipalities where the incumbent party almost lost and, thus, there was no Political turnover. In other words, the change of party in municipal power has a positive effect on such variables. Thus, when compared with municipalities where the incumbent remained in power, municipalities with a new ruling party had more SUS workers, more PSF visits were made and the coverage of primary care was greater 1 year after the elections.

Figure 2

Political turnover (2008 and 2012) and municipal health indicators: 1 (2009 and 2013) and 3 years after the elections (2011 and 2015) – Part 1. Source: DATASUS and TSE, data worked by the research. Note: Sample average within the bin. Kernel: Triangular. Polynomial fit of order 1.

However, when we analyze 3 years after the elections (right column), the significant result is not maintained for two of the three variables, so that only the number of SUS workers per thousand inhabitants remains significant.^[10] Three years after the elections, the results for the visits made by the PSF and the coverage of primary care indicate that there is no significant difference between municipalities that underwent political turnover and those that did not. We have indications, therefore, that the improvement in some indicators is transient, since the result was verified only in the year in which the political change took place.

Proceeding, Figure 3 shows the results for the ratio of cervical exams performed, immunization coverage, and infant mortality rate. For 1 year after the elections, only the variable referring to cervical exams performed is significant, indicating, once again, that in municipalities where there was no political turnover, this indicator was lower. For rates of immunization coverage and infant mortality, the results are not significant. For 3 years after the elections, none of these variables shows statistical significance.

Figure 3

Political turnover (2008 and 2012) and municipal health indicators: 1 (2009 and 2013) and 3 years after the elections (2011 and 2015) – Part 2. Source: DATASUS and TSE, data worked by the research. Note: Sample average within bin. Kernel: Triangular. Polynomial fit of order 1.

Through these results presented, it is possible to perceive a different behavior for the tested response variables. The indicators shown in Figure 2 reflect administrative variables, related to the configuration and functioning of SUS activities. In these indicators, the role of the SUS municipal manager is a protagonist.

As noted earlier, we obtained significant results for the number of SUS workers, the visits made by the PSF, and the coverage of primary care, indicating that 1 year after the elections, municipalities where there was a political turnover present such higher indicators.

We have also said that the position of SUS manager is appointed by the head of the municipal power. Thus, the change of party in power is related to the change of the municipal health manager. Our results indicate that the change in the municipal chief of power seems to benefit the health administration, but that these benefits do not seem to last after the year of the change. That is, the adoption of new strategies and the realization of new actions in the health sector – given the entry of a new party in the municipal power – are effective, but only in the current period when political change is taking place.

The indicators presented in Figure 3 reflect health variables that can be considered structural. Although the performance of cervical exams^[11] and the rate of immunization coverage directly depend on the management and SUS workers, they also depend on the interest of the service users. The infant mortality rate, in turn, depends on factors such as assistance to pregnant women, instruction and education policies, and adequate sanitation conditions. The results for immunization coverage and the infant mortality rate are not significant in the analyzed periods – 1 and 3 years after the elections. The lack of effect of political change on such indicators reflects the difficulty in reaching structural public health factors in the short term.

Our positive results found for the administrative variables can be related to the practice known as the report of the 100 days.^[12] It is common that on the hundredth day of government, mayors publish reports on the strategies to be followed in their mandates, dictating the pace of the government in the next 4 years, in addition to publishing the first balance sheets on their management. Therefore, the first year of management is a milestone for mayors, as it is an opportunity for them to demonstrate their management skills to the population. The improvement seen in the administrative variables can be explained by the need to adopt practices that, simultaneously, improve the government’s management indicators. This creates an illusion of greater administrative competence for the electorate. Furthermore, the idea of transient competence shocks is also explained by the more competitive political environment (close elections) that amplifies political pressure on the government.

Despite the lack of impacts on individual indicators (Figure 3), turnover can have indirect effects that are not reflected in individual indicators. Additionally, in Figure 4, we observe the effects of political turnover on composite indicators. The first is a Health Index that measures the quality of municipal health services estimated by the Rio de Janeiro Industries Federation (FIRJAN), and the second is a participation rate of SUS workers in relation to the number of basic health units. The health index “focuses on basic health and includes indicators whose control is the responsibility of the municipality: proportion of adequate prenatal care, deaths from ill-defined causes, infant deaths from preventable causes and hospitalizations sensitive to basic care” (Firjan, 2018). The number of workers per health unit tries to capture the capacity of municipal health care.

Figure 4

Political turnover (2008 and 2012) and municipal health indicators: 1 (2009 and 2013) and 3 years after the elections (2011 and 2015) – Part 2. Source: DATASUS and TSE, data worked by the research. Note: Sample average within bin. Kernel: Triangular. Polynomial fit of order 1.

We observe that the effects of political turnover on both indicators are positive and persistent. These results show that political turnover can improve the quality of municipal health services. The number of SUS workers per health unit is directly associated with the accessibility dimension of municipal health services. A larger number of professionals attending health units suggest a more collaborative approach and greater care in providing health services. The health index is multidimensional and captures complex interactions of municipal public health interventions. It allows for a more subtle evaluation of policy effects, considering the multifaceted nature of public health challenges.

4.3 Robustness and Sensitivity

To test the validity of our evidence, we estimated the same Equation 1 using two other estimation methods, in addition to the conventional one shown previously. Presented by Calonico et al. (2014), these two other methods are characterized by correcting the bias and robustness of the estimator. In addition, we also estimated equation (1) by including the covariates shown in Table 8 and changing the polynomial specification of the model. Tables 4 and 6 show the results for the outcome variables 1 year after the elections, and Tables 5 and 7 show the results for 3 years after.

In the context of Tables 4–7, we have three different methods and three different specifications for each of the outcome variables. In lines (a), (b), and (c), we have the methods “Conventional,” “Bias-corrected,” and “Robust.” In columns (1), (2), and (3), we have the specifications without covariates and polynomial of order 1, with covariates and polynomial of order 1, without covariates and polynomial of order 2, respectively.

The “Conventional” estimator is the initial estimate based on standard methods and assumptions. It represents the treatment effect without any adjustments or corrections. The “Bias-corrected” estimator is designed to address potential biases that may arise in the conventional estimator. It employs statistical techniques to correct for systematic biases, improving the accuracy and reliability of the treatment effect estimation. The “Robust” estimator accounts for potential issues with the data or model assumptions. It incorporates robust standard errors or bootstrapping techniques to provide valid inferences even in the presence of data imperfections or model misspecifications. The robust estimator increases the reliability and robustness of the treatment effect estimation, enabling more trustworthy conclusions.

Estimating and comparing these three estimators is important to assess the sensitivity of the treatment effect to different assumptions and approaches. By examining the variation across estimators, it becomes possible to identify potential biases, uncertainties, or violations of assumptions. This comprehensive analysis helps in drawing more reliable conclusions and enhances the overall robustness of the findings in RDD studies.

For 1 year after the elections (Table 4), the new results obtained for the number of SUS workers, the visits made by the PSF, and the coverage of primary care remain consistent with the idea that municipal political turnover favors health indicators related to its administration, since, in general, the three variables still prove significant.

When we analyze 3 years after the elections (Table 5), the results for the coverage of primary care are significant to the specification changes, indicating that the improvement in this variable may not be transient; that is, the improvement seemed maintained throughout the term of new managers. We also detect a change in the infant mortality rate, which in some specifications proves significant. However, for this variable, the result is not consistent, since it changes as we change the polynomial specification or include covariates in the model. Thus, we do not have precise conclusions on the impact of political turnover on the infant mortality rate.

The robustness analysis confirms our results found for the health variables that are most directly associated with the control of municipal management. These results show that political turnover has short-term (transitory) impacts on public health administrative variables, so that these benefit from the change of party in the municipal leadership. Thus, it is interesting to try to understand the mechanisms used by the new municipal management that can favor or imply the results found.

The results in Tables 6 and 7 also corroborate the effects observed in the health index. It provides a more comprehensive picture of the long-term effects of the policy by considering trends and patterns across multiple dimensions of public health. The composite indicator provides a more holistic perspective on the policy’s overall effects, considering both direct and indirect consequences. For example, the policy may have had a positive impact on health system efficiency, resource allocation, or coordination of care, which may not be visible in specific health outcomes.

4.4 Mechanisms

Since we find that political turnover has improved some administrative health indicators, we wonder how it happens. At first, we analyzed variables regarding the characteristics of municipal health management. Table 8 shows the average incidences and the tests on the equality of means performed on selected variables, considering the municipality that underwent political turnover as treated.

We obtained significant results for the tests of difference of means performed for the two variables referring to the education of the municipal manager of SUS. Thus, the results presented above indicate that in municipalities that underwent political turnover, the average education level of SUS managers was higher. One year after the elections, 70% of the municipalities with political change had SUS managers with higher education, whereas in municipalities where the party remained in power, this incidence is lower, around 65%. In addition, in municipalities with political turnover, there was a higher incidence of health trained managers.^[13] Thus, we can infer that political turnover is beneficial in the sense that as new mayors take office, the health managers they appoint are better qualified compared with those in which the municipal executive power has not undergone changes in leadership. This better qualification may reflect more effective strategies in order to improve or make administrative issues more efficient, which may explain the significant results previously presented in Figure 2.

Some observations should be made regarding the larger number of qualified health managers in municipalities that have undergone political turnover. There is a well-established literature, indicating that political turnover and government transitions can be accompanied by bureaucratic turnover. Iyer and Mani (2012) were the first to examine how political turnover affects bureaucratic turnover in India, revealing that politicians influence the process of appointing bureaucrats, and political changes are associated with significant increases in the likelihood of bureaucrats being relocated. More recently, in the context of Brazil, Brassiolo et al. (2021) discovered that municipalities experiencing political turnover after closely contested elections also witness an increased turnover at both the top (managers) and bottom (administrative and elementary officials) levels of the hierarchy.

Also observing the Brazilian case, Akhtari et al. (2022) have conducted a study that examines the occurrence of disruptions in municipal bureaucracy associated with political turnover during municipal elections. Furthermore, the authors expand their analysis to investigate the impact of public service provision in the field of education. Their findings indicate that an increase in personnel replacements in schools is linked to a decline in the quality of public education, as evidenced by lower test scores in subsequent years. This finding is particularly significant as it establishes a connection between changes in the bureaucracy and outcomes that directly affect the well-being of citizens. However, the adverse effects observed by the authors seem to disappear with time. In a similar vein, Toral (2022) has also explored political turnover in Brazilian municipal elections, focusing on the transitional periods and specifically examining the composition of the bureaucracy and the delivery of healthcare services. The study reveals that when incumbents are defeated in elections, it leads to reorganization within the bureaucracy and a reduction in healthcare service delivery during the transition period.

As Akhtari et al. (2022) point out, party turnover results in new hires being observed across various sectors, including education, health, and construction. While this extensive bureaucracy overhaul may impact many of the services provided by the municipality, it remains unclear in the literature whether this is ultimately beneficial or detrimental to the quality of such services. Although we did not encounter studies analyzing the effects of improved qualification in management positions following party turnover, it is worth noting that higher qualifications or levels of education are typically associated with better quality public service provision. This factor could potentially serve as one of the mechanisms contributing to the positive results found in this study.

We also sought to investigate whether there is a difference between some variables in the municipal healthcare network. In this case, we calculated the variation^[14] average 1 year after the election of such variables, in order to verify whether there is a difference in the generation of new units of the same between municipalities that have undergone political turnover and those that have not. The results are shown in Table 7.

We obtained a significant result of the test of difference of means for the variable referring to the variation of basic health units. In this case, we can intuit that municipalities that underwent political turnover created more basic health units compared with municipalities in which there was no political change. In fact, the data in Table 9 tell us that for every six municipalities in which the incumbent party lost the elections and there was political turnover, one center of basic health unit was created. As for the sample in which there was no political turnover, one center of basic health unit was created for each 17 municipalities in our sample. This result is consistent with the positive effect found in the previous section, where we verified that municipalities that go through rotation have better rates of coverage of primary care. We can infer, therefore, that this result is explained by the higher occurrence of the creation of new basic health units in municipalities that underwent changes in municipal leadership.

However, although we do not have significant results from the means tests for the other variables in Table 9, we think it is important to highlight how different they are between the treated and control groups. For the variation of jobs of community agents, for example, we found that one new agent was hired from each municipality that underwent political turnover. For municipalities where the party remained in power, however, one new community agent was hired for every three municipalities. Table A5, in the Appendix, shows the maximum and minimum values of the care network variables found for both groups. For the group that underwent political turnover, the maximum creation of jobs for community agents in a single municipality was 68, whereas for the sample of municipalities in which the parties remained in power, this maximum number was 32. Thus, for such indicators, even if we did not obtain significant results in the average tests, it is possible to perceive that the scenario in question in municipalities that undergo political turnover is better.