Subnational Politics and Redistribution in a Federal System: Determinants of Progressive Social Spending in Brazilian States

Abstract

What role do subnational governments play in shaping a country’s redistributive efforts? Existing literature suggests that federalism can be a hindrance to redistribution. Such negative effects may be particularly true of Latin America’s federations due to high levels of regional inequality and malapportioned political institutions. However, in order to fully understand redistribution in federal systems in Latin America, we need to examine not only how subnational governments affect centralized redistributive efforts, but also what efforts these subnational governments are making themselves. In this article, I contribute to our understanding of subnational social spending in Latin America’s largest federation, Brazil. My results suggest that, in Brazil, state governments are constrained actors, but they do pursue different levels of redistributive social spending with higher levels being more likely under left parties.

How do subnational governments contribute to redistributive efforts? Latin America’s welfare states are known for their underdevelopment and exclusionary nature (Huber and Stephens 2012; Pribble 2013), yet recent years have seen expanded access to the welfare state through universalistic reforms and the creation of new programs targeted at groups that were previously excluded (Huber, Dunn, and Stephens forthcoming; Pribble 2013). Despite such reforms, Latin America remains a highly unequal region. Why then have governments failed to engage in greater redistribution? When it comes to some of the region’s larger countries, a potential explanation may be their federal structure, something scholars have pointed to as a potential hindrance to redistributive efforts (Castles 1999; Obinger, Leibfried, and Castles 2005). While scholars have shown how federalism can prevent adequate implementation of centralized redistributive efforts (Niedzwiecki 2018), less work has considered the redistributive efforts made by subnational units.

In this article, I investigate which political and economic conditions are associated with subnational governments’ implementation of policies that are favorable to the poor. Specifically, I focus on conditions under which governments may be more likely to increase investment in social policies that disproportionately benefit lower class groups such as public health, education, social assistance, and public housing. Scholars have certainly considered determinants of national-level social spending in the past (Avelino, Brown and Hunter 2005; Brown and Hunter 1999; Huber, Mustillo, and Stephens 2008; Huber and Stephens 2012); yet in federal systems, lower levels of government may hold important competencies both for implementing and funding social programs (Hooghe et al. 2016). Likewise, we know that there are often key differences within countries in terms of the political and economic development of subnational units (Gibson 2012; Giraudy 2015; Montero 2007; Snyder 1999). Such differences seem likely to shape the political will and capacity of state governments to provide equal levels of investment in progressive social policies (Kleider 2018). In this article, I contribute to extending our understanding of the determinants of social spending, particularly progressive social spending, by bringing these subnational units into the discussion.

I focus my analysis on the case of Brazil, a country that has engaged in greater redistribution than Latin America’s other federations (Rogers 2021), but that still has persistently high levels of inequality. Scholars have noted that federal systems may hinder centralized efforts at redistribution (Castles 1999; Obinger, Leibfried, and Castles 2005); therefore, I turn our attention to subnational redistributive efforts. Specifically, I conduct a subnational analysis of progressive social spending by Brazilian states from 2002 through 2017 and find that even in the face of fiscal constraints, politics matter for determining how much of their resources states direct toward progressive social policies. In particular, I find that when left-of-center parties are stronger, state investment in social policies that benefit the lower classes is likely to be greater.

Literature and Theoretical Background

Determinants of Redistributive Spending

Much of our knowledge of the determinants of social spending is drawn from advanced developed democracies, as they generally have the most established welfare systems. One of the most consistent findings from this literature is that welfare state expansion is a function of the historical strength of left parties and organized labor (Esping-Andersen 1990; Huber and Stephens 2001; Korpi 1983; Stephens 1979).

A growing literature has examined how such theories travel to less-developed countries with younger and more fragile democracies such as those in Latin America. Latin American welfare systems have tended to be underdeveloped, benefiting a small group of relative elites, while neglecting or underfunding areas such as public health and education that would benefit the lower classes (Huber and Stephens 2012; Pribble 2013). However, many Latin American governments have begun to change directions and increase their investment in such areas as health, education and social assistance. While a general increase in social spending can be seen, not all countries and not all governments have pushed for social spending to the same extent (Huber, Dunn, and Stephens forthcoming). What then explains these differences?

While left parties and organized labor are central to understanding social spending in advanced industrial democracies, there is debate in the literature as to how much these variables can explain in the Latin American context. In terms of left parties, statistical analyses have not always found left parties to have a statistically significant effect on social spending (Huber and Stephens 2012). There is reason to be skeptical of the role of the left in expanding social policy in the region. In particular, scholars have argued that Latin American parties are, to varying degrees, weak, poorly institutionalized, lacking clear ideological consensuses and reliant on clientelistic and personalistic rather than programmatic linkages (Coppedge 1998; Hagopian 1996; Kitschelt et al. 2010; Mainwaring 2018; Mainwaring and Scully 1995; Roberts 2002).

However, increasing evidence points to the left, at least in some forms, as key to expanding social policy and particularly to making social policy more responsive to lower classes (Anria and Niedzwiecki 2016; Pribble 2013). While the previously described pattern of non-ideological and clientelistic linkages does characterize many parties in the region, some left parties have demonstrated a programmatic commitment to representing the poor. For example, Pribble (2013) argues that left parties in Brazil, Chile, and Uruguay are all sufficiently programmatic to be expected to push for social policy expansion when in office. While left parties in Brazil, Chile and Uruguay are all examples of institutionalized left parties making them perhaps the most likely candidates for implementing programmatic policy, ideological commitments tied with pressure from social movements have also been shown to push more weakly-programmatic left parties to expand social policy. For example, in Bolivia, strong ties to social movements was one of the key factors in pushing the leftist MAS to implement the non-contributory social pension, Renta Dignidad (Anria and Niedzwiecki 2016).

Given the interrupted histories of democracy in Latin America, regime type has also figured centrally in the Latin American literature in a way that it did not in advanced industrial democracies. In particular, scholars have broadly found evidence that democratic governments invest more in social policy than do autocratic governments (Avelino, Brown and Hunter 2005; Brown and Hunter 1999; Huber and Stephens 2012; Huber, Mustillo, and Stephens 2008). Kaufman and Segura-Ubiergo (2001) do not find that democracies invest more in social policies overall than do autocracies, but they do find evidence that democracies spend more on health and education.

Related to the importance of democracy, scholars have found that competitive electoral environments push politicians to be more responsive to lower class interests and, in particular, to implement social policy reforms. For example, Pribble (2013) argues that where electoral competition is fierce, parties are more likely to pursue universalistic social policy reforms even where leaders were otherwise hesitant to reform. Similarly, Garay (2016) argues that social policy expansions in Latin America have occurred when incumbents faced a credible electoral threat. In the face of intense competition, candidates need to court outsider voters, those employed in the informal labor market, and to do so, they will expand social policy. Relatedly, both Fairfield and Garay (2017) and Niedzwiecki and Pribble (2017) find that even right parties that lack a strong ideological commitment to social policy expansion will avoid cuts and even expand social policies in the face of strong political competition.

Yet, this finding is not undisputed, particularly when considering subnational spending in developing countries. In his study of Mexican municipalities, for example, Cleary (2007) finds that electoral competition has no significant effect on public goods provision, pointing to potential institutional hurdles and the lack of re-election as potential explanations. Some scholars have found that competition can actually hinder social spending. For example, Boulding and Brown (2014) find that Brazilian municipalities with less competitive elections allocate more resources toward social spending than do municipalities with greater competition. They argue that where resources are more limited, even politicians that have incentives to spend more, lack the resources to do so. As a result, politicians are less able to mobilize voters and more likely to face an electoral challenge. Similar results have been found in India (Nooruddin and Chhibber 2008) and in Argentina (González 2017).

Redistribution in Federal Systems

While the literature on determinants of social spending in Latin America has grown, it has yet to place much emphasis on the role of subnational spending. Understanding subnational spending in Latin America is important given that three of the four largest countries in the region are federal¹ and within these federations, subnational units maintain important responsibilities in social policy areas (Hooghe et al. 2016). Outside of the region, scholars are divided as to how decentralization will affect social spending and, particularly redistributive social spending. In the context of advanced developed democracies, scholars have identified federalism as a source of additional veto points (Castles 1999; Obinger, Leibfried, and Castles 2005). Such veto points can inhibit the expansion of the welfare state, but can also serve to prevent cuts to the welfare state (Beramendi 2012; Bonoli and Mach 2000; Huber and Stephens 2012). The effect of federalism on redistribution, then, depends on whether a state starts with high levels of redistributive social spending or low levels. In his work, Beramendi (2012) also identifies an interactive effect between federalism and party strength. Strong parties can counter the negative effects of federalism on redistributive spending.

Given that Latin American countries are generally starting from a low level of redistributive social spending, we may expect that federal system will serve as a hindrance to increasing spending. Additionally, political parties in the region are comparatively weak (Mainwaring and Torcal 2006) so are unlikely to counter this effect. Federalism has indeed been identified as an obstacle to redistributive efforts by central governments in the region (Rogers 2021), but the literature has placed much less focus on examining variation in social policy efforts by subnational units themselves. In this article, I turn to the Brazilian case to examine the determinants of progressive social spending by the intermediate federal units, the states.

The Brazilian Case

Brazil is a useful case to examine for a number of reasons. First, Brazil is characterized by high levels of regional inequality and malapportioned political institutions, providing a setting that is arguably unfavorable to broad, centralized redistributive efforts and instead favorable to transfers to states (Beramendi, Rogers, and Díaz-Cayeros 2017). While transfers to states may address inequality if they are targeted at the poorest states, malapportioned legislatures seem to benefit the least populous states rather than the poorest states (Gibson, Calvo, and Falleti 2004; Arretche and Rodden 2004). Likewise, legislative malapportionment has led to a rural-conservative bias in Brazil (Snyder and Samuels 2004). Along with directing transfers to unpopulated states rather than the poorest states, the conservative bias provides an additional hurdle to passing centralized redistributive reforms (Beramendi, Rogers, and Díaz-Cayeros 2017; Snyder and Samuels 2004).

This is not to say that Brazil has failed to engage in centralized efforts at reducing inequality, but rather that there are additional hurdles to implementing sufficient centralized redistributive programs. For example, one of the main success stories of centralized redistribution in Brazil has been the conditional cash transfer program, Bolsa Familia, and Fenwick (2009, 2010) argues, that federalism actually helped promote centralized redistribution in this case. While Bolsa Familia is certainly a notable program that has had a major effect on reducing the severity of poverty and inequality, it should also be noted that it is a quite small program, accounting for only about one half of a percent of GDP, and as a result, is limited in its ability to address poverty and inequality on a larger scale (de Souza et al. 2019). Another key central redistributive program, the noncontributory pensions program Benefício de Prestação Continuada, accounts for a similarly small portion of the GDP, and while important in its impact, does not fully address the problems of poverty and inequality, leaving much more to be done (Medeiros and Souza 2015).

Such a context may mean that the setting is in some ways conducive for subnational governments to act to reduce inequality. In fact, Rogers (2021) shows that states display very different redistributive efforts in Brazil, with Goiás, the least redistributive state, reducing market inequality through redistribution by just 12 percent while Maranhão, the most redistributive state, reduces market inequality by nearly 22 percent. Understanding the drivers of such subnational variation is clearly consequential. While subnational redistributive efforts may not drastically reshape the territorial inequality found in Brazil, they may still be consequential for reducing inequalities within subnational units.

A second reason to look more closely at states is that the constitution grants them important responsibilities for progressive social policies. States are generally responsible for secondary education and, in some cases, primary education. Additionally, states are responsible for providing coordination and financial support to the public health system in their territory. States also often implement their own social assistance programs and provide some support for public housing.

Third, Brazilian states are not solely reliant on intergovernmental transfers, but rather have their own sources of revenue in the ICMS, a sort of value added tax, and, to a much lesser extent, a motor vehicle tax. On average, the ICMS alone accounts for well over half of state revenues and own-revenues combined account for over 70 percent (Ter-Minassian 2012). States do face some restrictions as to their ability to spend their revenues as they please. For example, states are required to spend 12 percent on health and 25 percent on education, but they can, and often do, allocate a higher percentage (Santos et al. 2017). Likewise, many states face high debt loads that limit the fiscal space they have to increase investment in policies. Subnational debt has been particularly relevant in recent years as Brazil entered into a recession and some of Brazil’s largest states have declared fiscal calamity.² However, these particularly bad financial times largely fall outside of the period considered in my analysis. During better financial times, debt has been more of a constraint on some states than others.³

As a result of such constraints, some scholars have expressed skepticism that states can take different paths when it comes to social spending. For example, Arretche (2005) notes that the central government remains the dominant actor in social policy, limiting the autonomy of states and municipalities in terms of tax collection and determining how funds are spent. In her analysis of state-level social spending, Sátyro (2013) seems to find support for this argument. Her analysis of the political and economic determinants of social spending finds little evidence that political actors can influence spending levels within states as they are too constrained both by constitutional spending requirements and their debt loads.

Despite these challenges, qualitative evidence suggests that states are still finding ways to expand social spending. For example, the governor of Brazil’s poorest state, Maranhão, has made waves since coming to office in 2014 for raising teacher salaries to the highest in the country (Gazeta do Povo 2019). In addition, Maranhão has invested in other areas of education such as opening new schools. Likewise, in interviews I conducted in three Brazilian states in 2019, politicians, including former governors, noted that despite tight budgets, they were able to find ways to invest in policies they viewed as important. For example, when asked about social policies during his tenure, a former governor pointed to a new early childhood program his government implemented and noted, “Investments that we made in education and health principally, were investments that, despite the fiscal difficulty of the state, were important (personal interview, April 17, 2019).”

While there is reason to be hesitant about the ability of Brazil’s subnational governments to bring about major social policy change, it is worthwhile to understand what role they may play, even if incremental, in changing the direction of social policy in their territories. I do not expect that states will be able to drastically increase or decrease their social spending, yet it strikes me as unlikely that there is no difference between states with drastically different socioeconomic and political climates. States vary quite substantially on variables the literature suggests should matter for social spending. Importantly, they vary in terms of left party strength.

While Satyró (2013) does not find that left parties affect social spending in Brazilian states, since the end of the period included in her analysis, left parties have grown in importance across Brazil. While the main leftist party, the Partido dos Trabalhadores (PT), won its first governorships in the 1994 elections, its success at the state level remained fairly narrow initially. Through the 2002 elections, the latest considered in Sátyro’s analysis, the PT had only governed in six states, winning a total of eight terms. By 2017, the end of my dataset, the PT had governed in eleven states, winning a combined total of twenty-three terms. In addition to the PT, other leftist parties saw new electoral success during this period. The Partido Comunista do Brasil (PCdoB) won its first governorship in 2014 with the victory of Flávio Dino in Maranhão. The Partido Socialista Brasiliero likewise found renewed success in some states and expanded its success into new states. Taking into account these years where the left has become a much more relevant state actor is important for our understanding of how politics influences state-level social spending. While the limited experiences of having the left in power may not have a significant effect on social spending, we may expect to see more of an effect as the left’s experience governing grows.

In addition to the rise of the left, Brazil’s party system has become more programmatic over time (Figueiredo and Limongi 1999; Hagopian, Gervasoni and Moraes 2009; Mainwaring et al. 2018). Left parties generally espouse commitments to greater equality and, where parties are programmatic, we would expect such values to drive policy. While previously characterized as having an inchoate party system (Coppedge 1998), Brazil has since bucked regional trends and moved toward greater institutionalization (Mainwaring et al. 2018). Relatedly, state and market reforms in the 1990s helped create clear ideological divides in the party systems as well as reduced the resources available for parties to engage in traditional patronage politics (Hagopian, Gervasoni, and Moraes 2009). Likewise, changes to formal rules have increased party discipline (Figueiredo and Limongi 1999). Finally, surveys of Brazilian legislators have shown that parties have remained relatively consistent in terms of their ideology. While an overall shift toward the center can be observed, the ordering of parties on a left to right scale has remained fairly consistent (Power and Zucco 2012). The Brazilian party system still faces challenges such as a lack of societal roots (Samuels and Zucco 2018; Zucco 2015), but the shift toward more programmatic, disciplined and consistent parties should have consequences for policy outcomes. In particular, I expect that parties will work to implement programmatic policies that align with their ideological commitments. This leads to the following hypothesis:

Hypothesis: Compared to states governed by other political parties, states governed by left parties will invest more in progressive social policies.

Data and Methods

To examine under what conditions Brazilian state governments may increase investment in progressive social policies, I have compiled a dataset of state-level spending as well as a variety of political and economic variables that I expect should influence spending. The dependent variables in my analysis are levels of social spending as a percent of state GDP. This allows me to measure the commitments of governments to social policy within the fiscal constraints that states are under. In the first model, I consider social spending as a whole. I restrict my analysis to the categories of spending that are redistributive in nature, that is, spending that disproportionately benefits the lower classes. As such, I exclude spending on social security. In the Latin American context, social security spending is more regressive than progressive (DeFeranti et al. 2004; Lindert, Skoufias, and Shapiro 2006). While the 1988 Constitution created a non-contributory pension aimed at providing support for the disabled or the elderly who lack another pension, this program is run and funded by the central government. State spending on social security is for the payment of civil servant pensions. As such, this spending benefits a disproportionately privileged class and cannot be seen as social spending aimed at responding to the interests of the lower classes. I do include spending on education,⁴ health, social assistance, and public housing.

Following my analysis of overall social spending, I also analyze health and education spending separately. Health and education spending jointly make up the vast majority of state-level social spending when social security is excluded. For these models the dependent variables are health spending or education spending as a percent of GDP. All spending variables are drawn directly from state budget information as reported by the National Treasury’s System for Public Sector Accounting and Fiscal Information.

My main independent variable of interest is the party of the governor. To capture this, I include a series of dummy variables. Since Brazil’s party system is known to be very fragmented, I have created dummies for the larger parties and grouped some smaller parties together based on similar ideological leanings. This results in five dummies: Partido da Social Democracia Brasiliera (PSDB), Partido do Movimento Democrático Brasiliero (PMDB),⁵ left parties including the Partido dos Trabalhadores (PT),⁶ other small parties,⁷ and the Democratas.⁸ As Figure 1 shows, in my dataset the combined left parties are in power for the most cumulative years followed by the PSDB and PMDB, both centrist parties. Finally, smaller parties, generally leaning right of center, and the Democratas have held governorships for the fewest cumulative years. In the Online Appendix, I also include models using a dichotomous variable that simply considers whether the governor is from a left party or not. I focus my attention on the governor given that Brazilian state governments are characterized by executive dominance (Abrucio 1998). Legislatures are generally highly fragmented and in no case in my sample does the governor’s party hold more than 50 percent of the legislative seats. Likewise, state legislatures are generally dominated by the governor in Brazil so even if a governor does not hold a majority of seats, they tend to be able to pursue their legislative agenda (Abrucio 1998). As I have argued, I expect that left parties will be more inclined to increase social spending that disproportionately benefits the lower classes compared to any other group of parties.

All models in my analysis also include a variety of controls. First, I include a series of variables to capture the level of political competition in each state. While some states, such as Minas Gerais and Rio Grande do Sul, have exhibited high levels of political competition since electoral politics returned to the states, others, continue to exhibit more limited competition (Borges 2007). In still other cases, competition has increased over time. For example, in some states, particularly in Brazil’s poor Northeast, dominant parties or actors maintained a strong grip on state power long after the return to democracy, but that grip has begun to decay in more recent years (Borges 2007). As previously cited literature notes, there is a debate as to whether or not greater competition should lead to higher or lower social spending, but evidence suggests it should matter so it is important to include in my models. To capture the level of political competition I use three variables: continuity, first round vote share, and share of seats held by the governor’s party. Continuity takes on a value of one when the elected governor is either the same person or from the same party as the previous governor. First round vote share reflects the percentage of the vote won by the governor in the first round of elections.⁹ Finally, seat share accounts for the number of seats the governor’s party won in the legislative assembly. I am ambivalent as to the direction of the effect competition will have on social spending.

To account for the argument that high levels of debt may prevent states from having sufficient budgetary flexibility to increase investment in social policies, I include the ratio of consolidated liquid debt to consolidated liquid revenue as reported by the National Treasury. I expect that a higher debt burden will lead to a reduction in social spending.

In addition, I include state GDP per capita and the Gini index. Brazil has been characterized as BelIndia due to the disparities in development among different regions of the country; while some have living conditions on par with European nations like Belgium, many face living conditions much more on par with developing countries like India. The well-off are more concentrated in the more developed states of the South and Southeast, while the poor are more heavily concentrated in the less-developed states of the North and Northeast. Taking into account these vast wealth differences is necessary. While there is likely to be greater need for increased spending in states where the GDP per capita is lower, I expect that states with higher GDP per capita will spend more on social policies because they will have more resources available to invest.

While the wealth of a state is of course necessary to consider, it is also important to consider how this wealth is distributed. Brazil is a notoriously unequal country and this inequality is worse in some areas of the country than in others. I expect that the more unequal a state is, the less it will spend on social policies. The Meltzer-Richard theory suggests that more unequal societies should see more redistribution as the median voter’s income will be below mean income making them more likely to benefit from redistribution (Meltzer and Richard 1981). However, this theory fails to account for the fact that political power is not equally distributed. Power resources theory, on the other hand, incorporates the balance of class power, suggesting that we should expect political power to be concentrated in the hands of the upper classes in the absence of strong organizations representing the lower classes (Korpi 1983; Stephens 1979). In Brazil, concentrated wealth has largely coincided with concentrated political power¹⁰ so the Meltzer-Richard theory is unlikely to apply. As a result, I expect these states to be less responsive to the poor and therefore unlikely to increase investment in policies that primarily benefit them, despite the fact that the need for such policies would be stronger.

Likewise, I take into consideration the importance of the private sector for both education and health. A larger private sector is expected to reduce public social spending for two reasons. First, the private sector may be absorbing some of the demand on the public sector creating less need for public spending. Second, where a larger portion of the population uses private services, there may also be less political support for public spending since people are already paying privately and thus may not want to pay taxes to fund public health and education. To measure the size of the private education sector, I measure the percent of total enrollments in each state that are in the private sector. To capture the importance of the private health sector, I include a control for the percentage of the population in each state that has a private health plan.

In addition to the economic and political variables, I also include demographic controls. First, I include the level of urbanization of each state and the level of diversity as measured by the percent of the population that identifies as Black or Indigenous. In the education and health models, I also include a variable to capture the size of relevant age categories. In the education spending models, I include a measure of the percent of the population between the ages of ten and nineteen. I select this age group because states are largely not responsible for the first phase of primary education, but rather the second phase of primary education and secondary education. A state with a larger school-aged population would be expected to invest more in education. In the health spending models, I include a measure of the percent of the population age sixty-five or older as the elderly are expected to make heavier use of the health system. I expect states with larger aged populations to spend more on health than others to account for the likely higher demand.

My dataset is composed of 432 observations from twenty-seven states across sixteen years. Table 1 shows the mean values of the dependent variables for each of the states.

Time series cross-sectional data poses a number of challenges to ordinary least squares regression. According to Hicks (1994) when OLS is used for pooled data five problems are likely to arise; estimates will be temporally autoregressive, cross-sectionally heteroskedastic, cross-sectionally correlated, will conceal unit and period effects, and will reflect causal heterogeneity across space and time. Scholars have debated the best way to address these problems when using pooled data and disagreement remains. Some argue that fixed effects are the best approach as they address omitted variable bias due to the fact that units had different historical trajectories prior to the period of analysis. However, Plümper, Troeger, and Manow (2005) warn that fixed effects do not simply account for omitted variable bias, but also, among other concerns, eliminate variation due to variables that are time invariant and reduces coefficients on variables that differ mainly between units. Rather than fixed effects, I use Prais–Winsten estimation with panel corrected standard errors in line with the recommendations of Beck and Katz (1995). Fixed effects models are included in Table A9 of the Online Appendix as robustness tests and provide similar results regarding the party of the governor though they prove less useful for understanding many other variables which vary more across states rather than within states over time.

Table 2 shows the results for the total state social spending. Model 1 includes just my main independent variables of interest while Model 2 adds in economic controls and Model 3 adds demographic controls and the commodity boom dummy. Finally, Model 4 adds in controls for the size of the private health and education sectors for the fully specified model.

First, looking at partisan politics, compared to when a left party enters the governorship, all other parties are expected to spend less on social policy when they come to power. The coefficient is not only negative, but significant at least at the 0.1 level for all party groupings in the fully specified model.

Looking at the control variables, neither the governor’s first round vote share nor the governor’s party’s seat share have significant effects on social spending. Continuity, however, is positive and significant. An additional term in office, then, may provide added time and flexibility for a party to pursue such changes.

Turning to the other variables in my model, we see that both GDP per capita and the debt to revenue ratio have consistent, statistically significant and negative effects on social spending. While I expected a higher GDP per capita to lead to higher spending, the decline in the size of the coefficient when controls for private health and education are included suggests that this negative effect is at least partly explained by the fact that where citizens are wealthier, they are choosing to use private rather than public services. As a result, they may not pressure the government to increase investment in public health and education. The percent of school enrollments that are in the private sector and the percent of the population with a private health plan both have significant, negative effects on public social spending.

Finally, looking at the demographic controls, having a larger urban population and having a larger Black and Indigenous population both have positive and significant effects on total social spending.

In addition to total spending, I also analyze education and health spending separately. In all states, these two categories account for the vast majority of spending so it is worth examining them each more in depth. First, I look at education spending as a percent of total spending. The only previous piece that I know of that examines determinants of state level spending (Sátyro 2013) combines education and culture spending because prior to 2002, these categories were not separated in the budget data. However, disaggregated spending is now available allowing me to examine just education. While culture spending appears to have generally been just a small portion of the combined total, it is important to be able to examine education spending alone as the determinants of public education spending and culture spending are potentially different.

Table 3 shows full results for the education spending models. As with overall spending, I find support for my left party hypothesis. The coefficients on each of the party variables are negative as with overall spending. In the fully specified model, the PMDB, PSDB, Democratas, and other small parties are all significant at the 0.05 level.

The debt to revenue ratio is statistically significant and negative as in the total spending models. GDP per capita has a negative effect in Model 6, but the effect nears zero and is no longer significant once the school-aged population and private education variables are added in Models 7 and 8.

Finally, looking at demographic controls, a larger urban population has a positive and significant effect on education spending. The school aged population is likewise positive and significant as expected.

Lastly, I turn to health spending. While health spending generally accounts for a smaller portion of state-level social spending than education, it still accounts for the second largest portion of social spending at the state level, making it an important consideration.

Full results for the health spending models are shown in Table 4. Some similar trends again emerge when we look at health spending. All non-left parties continue to take on negative coefficients and these coefficients are statistically significant for the PSDB, PMDB, and the Democratas.

Competition variables again appear to tell us little about social spending. Looking at the economic variables, debt to revenue again has a significant, negative effect on social spending. GDP per capita is significant and negative in Model 10, but once demographic controls and the controls for private health coverage are included in Models 11 and 12, the effect is no longer significant.

Finally, the percent of the population living in urban areas does not have a significant effect on health spending in the fully specified model, while the percent of the population that identifies as Indigenous or Black has a positive effect. The population aged sixty-five or above surprisingly has a significant and negative effect on social spending.

To test the robustness of my findings, I also run all of my models using social spending, education spending and health spending as a percent of total spending rather than GDP. These results provide additional support for my hypotheses, though are somewhat weaker. The differences between these results are likely due to the fact that using spending as a percent of GDP captures how much a state is willing to engage in spending whereas using total spending does not account for this potentially important variation. Full results from this analysis are available in Table A3 of the Online Appendix.

It is important to note that my sample, except for the year 2017, considers a time period where the left was in power at the national level. While own-revenues account for a large portion of state budgets, they also receive mandatory and discretionary transfers from the national government. Research has shown that the president’s co-partisans disproportionately benefit from discretionary transfers (Miranda Soares and Neiva Robson Pereira 2011), suggesting that PT governors likely benefitted more than others under national PT governments and this may have given them some additional flexibility to increase spending compared to other parties. To test the possibility that co-partisanship may be driving my results I run two additional analyses. First, I use discretionary transfers as a control variable to test the possibility that co-partisan governors may receive a disproportionate share of discretionary transfers from the national government, allowing them greater flexibility to invest in social policies. At the time of this writing, data on discretionary transfers to the states was only available beginning in 2011 so these are much smaller samples. However, the results, using both discretionary transfers per capita and as a percent of GDP, are consistent with my previous findings. Most notably, PMDB and PSDB are negative and significant at at least the 0.1 level in all models. Second, to consider the full sample, I separate out the PT from other left parties to see if the PT drives my results. These results show that non-left parties do spend significantly less than the PT, but other left parties do not. Full results from these analyses are available in Tables A4–6 of the Online Appendix.

Discussion

The results presented in the previous section provide interesting insight into the determinants of social spending at the state-level. These results contradict some of what the state-level literature has argued in the past, most notably that the governor’s party is not a significant determinant of social spending at the state-level. Instead I find that the determinants of state-level social spending are more similar to those of national-level social spending than previously argued. In particular, I find that while debt does constrain spending, it does not appear to prevent political considerations from having an effect on social spending.

In terms of partisan politics, I find evidence to support my hypothesis that left parties will spend more on progressive social policies than will other parties. My analysis likewise suggests that these differences in spending cannot simply be attributed to alignment between leftist presidents and governors. In nearly all of my models, the two main centrist parties, the PSDB and PMDB, and the right-of-center Democratas have statistically significant and negative effects on social spending compared to when a left party is in power. Other small parties have a significant, negative effect on total social spending and education spending. When the PMDB is in power, for example, my model predicts 0.005 percent of GDP less to be spent on social policies than when left parties are in power. While 0.005 percent may seem small, this is actually a quite large substantive change. During the period included in my sample, the average state had a GDP of approximately BRL 143 billion. A decrease of 0.005 percent of GDP spent on social policies would be a decrease of almost BRL 715 million. In any context, this would be an important decrease in spending, but it is all the more noticeable in systems where public services are so underfunded as is the case throughout much of Brazil.

Conclusion

In this study, I find evidence that politics matters when it comes to the ways in which states allocate their resources. States may face financial constraints that limit how large an impact any given state politician can have on social spending, but even in the face of such constraints, partisan politics may inspire incremental changes in social spending.

These findings make important contributions to our understanding of party politics and social spending in Brazil as well as to the broader literature on federalism and redistribution. Despite growing evidence that left parties have been central to understanding recent expansions in social policy in the region, some scholars remain skeptical of the importance of partisan politics given the traditional weakness of programmatic party politics in Latin America. While it is true that many Latin American party systems have yet to become institutionalized and many parties continue to rely on clientelistic rather than programmatic linkages, since the early 2000s the Brazilian party system has moved toward greater institutionalization with major parties providing programmatic alternatives to voters (Mainwaring et al. 2018). This context provides greater reason to expect different policy outcomes depending on what parties come to power. When left parties win, we are likely to see investment in policies that align with the left’s ideological commitment to greater equality. Where Latin American party systems have moved toward institutionalization, my results provide evidence that we should expect partisan politics to matter for policy outcomes like social expenditures.

Likewise, my results add to our understanding of how decentralization affects social spending. I find that under certain circumstances states will actually increase their social spending. This finding is in-line with Kleider’s (2018) findings that decentralization actually leads to greater within-country variation on social expenditures. While my results are focused on a single case, they provide evidence that we should look more strongly at the role subnational governments can play in responding to the interests of lower class citizens through investment in progressive social policies. Federations can be found across the globe making an understanding of subnational actors critical.

Finally, beyond contributing to the academic literature, these findings also have important practical consequences for citizens. Recent academic literature suggests that Brazilian states are unable to make major policy decisions and are instead relegated simply to finding ways to cut spending in order to manage their massive debts (Sátyro 2013). Such ideas have translated into public opinion as well. A survey conducted by Arretche, Schlegel, and Ferrari (2015) finds that citizens view states as the least important of Brazil’s three levels of government. If citizens are directing their attentions toward the municipal and central governments while overlooking the state government they may be missing opportunities to push for more responsive government. Especially as the national government in Brazil has moved away from social policy expansion toward social policy retrenchment under both President Temer and President Bolsonaro, lower classes may need to rely more heavily on subnational governments to help make up for at least a small portion of national level cuts. While responsiveness to their interests may decline at one level of government, the lower classes may still be able to find government responsive to their interests at other levels. Likewise, even if a national government is responsive to the interests of the lower classes, state governments can take away from some of these gains by blocking the implementation of policies beneficial to the poor (as shown in Niedzwiecki 2018) or by cutting spending on social policies over which states maintain responsibilities, as shown in this analysis.

Supplementary Data

Supplementary data are available at Publius: The Journal of Federalism online.

I appreciate suggestions received on earlier versions of this article presented at the Pontifícia Universidade Católica do Rio Grande do Sul, UNU-WIDER, and the annual meeting of the American Political Science Association. In particular, I would like to thank Evelyne Huber, Jonathan Hartlyn, Patricia Justino, Eva-Maria Egger, Florian Hollenbach, Steven Sparks, and anonymous reviewers for their helpful comments. I would also like to acknowledge support from a Fulbright-Hays Doctoral Dissertation Research Abroad Fellowship from the United States Department of Education (P022A180038-002) and a Mellon Dissertation Fellowship from the Institute for the Study of the Americas at the University of North Carolina, Chapel Hill.

Footnotes

References