Inequalities and the COVID-19 pandemic in Brazil: Analyzing un-coordinated responses in social assistance and education

This paper contributes to discussions about subnational responses to the coronavirus disease (COVID-19) pandemic in federal countries. In the scholarship on federalism and public policy, few studies seek to understand the factors that shape subnational differences in welfare levels. This article seeks to better understand this issue in Brazil by exploring how, in a context with little national-level coordination, subnationalgovernmentstackletheinequalitiesexacerbatedbytheCOVID-19pandemic. This study analyzes two social policy areas with distinct national-level coordination mechanisms and federalresponsibilities: social assistance and education. Twomulti-levelcases are examined: the states of Amazonas and São Paulo and the cities of Manaus and São Paulo. This analysis relies on quantitative data, mainly social indicators, and qualitative data collected through documents and in-depth interviews. Social assistance and education policy actors in Amazonas and São Paulo faced at times significant obstacles adapting and/or creating policies to tackle inequalities, resulting in a fend-for- yourself federalism and fragmented subnational policies. Differences in subnational responses can be explained by distinct policy legacies and previous capacity, which were key in organizing a useful response to the pandemic. However, to fully explain subnational responses, the role of actors within institutional contexts must also be taken into account. In social assistance, shared responsibilities among different levels of government led to competition and credit claiming dynamics, reinforcing fragmented and uncoordinated responses. In education, decentralization and more stable funding allowed political leadership to activate and mobilize subnational capacities and other actors at the subnational level, producing more sustainable responses. of between cash transfer in both a situation where different actors could claim credit for programs. Local actors were pushed to act, but relief responses were fragmented, uncoordinated, and only met recipients immediate needs. In education, funding and a more decentralized path allowed subnational governments to adopt more sustainable responses. However, these depended on the role of specific actors who activated subnational capacity and mobilized other actors around specific policy ideas. This study highlights the importance of analyzing variations in social policy at the subnational level, including both national and subnational coordination mechanisms, and of considering the intersec-tions between institutional contexts and agency. Future comparative studies are key to better grasping subnational policies, especially welfare policies; their heterogeneity and effects on inequalities; and the factors that explain their existence and content. This study also calls attention to the importance of further exploring institutional structures’ activation, resilience, and capacity for change, and formal and informal relationships built during critical juncture periods.

been put in place. Overall, the federal government has taken on an important role in coordinating policies. These changes were key in reducing regional inequalities by expanding service provision coverage across the country (Arretche, 2012;R. Bichir et al., 2020;Franzese & Abrucio, 2013), even though inequalities in access and quality persist.
The issue of inequality re-surfaced in both the public debate and the government's agenda during the coronavirus disease  pandemic. Vulnerable populations (e.g., low-income families; single mothers; the elderly; Blacks, Indigenous peoples, and other minority groups) have been the most affected by the pandemic. At the end of the first trimester in 2021, the four-trimester accumulated gross dometic product (GDP) variation was −3.8% (IBGE, 2021a). The 12-month accumulated variable of inflation increased at the beginning of 2021, hitting 6.76% in April (IBGE, 2021b). Unemployment in the first trimester of 2021 reached 14.7%, the highest since 2012 when the data series begins (IBGE, 2021c). Such high unemployment rates have increased vulnerability and the demand for social services, cash transfer benefits, and public schooling.
This paper seeks to contribute to the debate on welfare policies in federal countries by exploring how subnational governments have tackled some of the inequalities the COVID-19 pandemic has exacerbated in Brazil. Some scholars highlight the importance of national-level coordination in ensuring similar levels of welfare in federal countries (Greer, 2006;Obinger et al., 2005), while others show that national-level coordination does not fully eliminate differences in social policy within a country. These differences result from different factors (e.g., ideology, actors, ideas, and networks) at the subnational level that are then constrained by institutions, particularly federal dynamics (Niedzwiecki, 2015;Rodrigues-Silveira, 2012;Sellers & Lindström, 2007).
Brazil is an interesting case because national-level coordination (Arretche, 2012), which has been significant in the past, has declined significantly since the beginning of Bolsonaro's government in 2019. This study seeks to understand local governments' responses to the pandemic in two policy areassocial assistance and education-in this context of a lack of national coordination (Abrucio et al., 2020).
Social assistance and education are and will be key policy areas for tackling inequalities in the next decades, but national-level coordination and federal responsibilities for the two policy areas vary. In social assistance, the federal government is responsible for national guidelines, standards, information and monitoring systems, and revenue redistribution, as well as social benefits and cash transfer programs. States' roles are limited to coordinating policies regionally by setting regional parameters and distributing funding, while municipalities provide social services. Education is more decentralized. There is a lower level of national coordination, and states and municipalities share responsibility for service provision. Although the federal government has strengthened its role as a national coordinator in education, especially with budgetary incentives to encourage subnational governments to adhere to specific standards, it has a more restricted role.
We analyzed two states and two municipalities severely affected by the pandemic-the state of Amazonas, the city of Manaus, the state of São Paulo, and the city of São Paulo. Both cities are complex urban settings, yet they have different spatial distributions of income and other measures of equality, as will be seen.
We conducted a quantitative analysis of social indicators and a qualitative analysis of data collected through documents and in-depth interviews with managers at the state and municipal levels. This mixed-methods analysis of both policy areas and cases allows us to better understand the impact national-level coordination mechanisms and subnational-level factors can have on subnational responses. Our research shows that the pandemic has reinforced the lack of national coordination in place since 2019, which has affected both social assistance and education policies in Amazonas and São Paulo.
Amazonas and São Paulo have differing levels of fiscal and service provision capacity, but they both faced challenges adapting and/or creating policies without national coordination, resulting in fend-for-yourself federalism and fragmented subnational policies. Subnational governments relied on policy legacies and previous capacity to create or adapt subnational policies during the pandemic. Local institutional actors also played an important role in these processes. In social assistance, shared federal-state-municipal responsibility in a context where there was little intergovernmental coordination resulted in competition and credit-claiming. While this pushed local actors to act, fragmented, uncoordinated, and emergency-relief responses prevailed. In education, decentralization and guaranteed minimum funding allowed municipal and state political leaders to activate subnational capacity in different ways, including mobilizing other actors around specific policy ideas, leading to more sustainable responses.

Welfare policies in federal countries
Some researchers argue that, although national coordination is important, it does not fully explain the content of subnational policies. Welfare levels vary significantly within federal countries (Niedzwiecki, 2015;Rodrigues-Silveira, 2012;Satyro & Cunha, 2019). The literature points to different political, institutional, ideational, and ideological factors (Niedzwiecki, 2015;Sellers &Lindström, 2007) andsocioeconomic, demographic, andethnic-cultural diversity (Rodrigues-Silveira, 2012) to explain these variations. Sellers and Lindström (2007), for their part, highlight the relevance of local systems of welfare governance and propose moving the debate beyond macro-analytical categories and federal-unitary and centralized-decentralized binaries.
Given that leftist parties have often been responsible for welfare expansion at the national (Huber & Stephens, 2001) and subnational levels (Niedzwiecki, 2015;Vampa, 2016), partisanship has been a central variable in explaining welfare policies. In terms of subnational social policies, the Brazilian case shows that other political and institutional factors, such as competitive intergovernmental relations and credit-claiming by different levels of government (Alves, 2015;Borges, 2008;Fenwick, 2010Fenwick, , 2015Niedzwiecki, 2015), are also relevant. Other institutional factors also play a role in subnational policy-making, such as multi-level governance (Alves, 2015), state capacity, and policy legacies (Rodrigues-Silveira, 2012). Finally, ideology, ideas, and networks help explain subnational policy changes (Segatto, 2018;Sugiyama, 2008).
The scholarly literature on subnational policies in Brazil show that, even though subnational governments have administrative, fiscal, and political powers, the federal government has had an important role in coordinating different policies, which encourages states and municipalities to implement national decisions (Arretche, 2012). Even so, subnational policies vary, not only between policy areas with different levels of national-level coordination, but also within the same policy area Gomes, 2009;Segatto, 2018). Recent studies show that these variations result from differences in state capacity that was, in part, built from national-level coordination mechanisms Lavalle et al., 2019;Lima-Silva, 2019;Lima-Silva & Loureiro, 2020), policy legacies Lavalle et al., 2019;Lima-Silva, 2019;Lima-Silva & Loureiro, 2020), the centrality of the policy area in the governmental agenda , and political leadership Lima-Silva, 2019;Lima-Silva & Loureiro, 2020).
This study seeks to contribute to the debate about subnational policies by exploring whether policy legacies and previous capacity, which have in part been shaped by national-level coordination mechanisms, explain both state and municipal responses to the pandemic, or whether other subnational factors highlighted in the scholarly literature (e.g., ideology, actors, ideas, and networks within institutional contexts) explain subnational responses.
This article includes two policy case studies (social assistance and education) in two Brazilian states (Amazonas and São Paulo) and two cities (Manaus and São Paulo) (Figure 1). These policies were chosen due to their importance in tackling inequalities in the long term. The cities of Manaus and São Paulo were chosen because despite having distinct demographic and socioeconomic characteristics both cities have been seriously impacted by the COVID-19 pandemic. Manaus, the capital of the state of Amazonas in the Northern region, had a population of 2.2 million in 2020, while the city of São Paulo, located in the Southeast in the highly industrialized state of São Paulo, was home to 12.3 million. Per capita GDP for the city of São Paulo was R$58,691.90 in 2018, in contrast with R$36,445.75 in Manaus (IBGE, n.d.a, n.d.b).
Nevertheless, both Manaus and São Paulo were severely affected by the COVID-19 pandemic. Although neither figured among the state capitals with the highest rates of COVID-19 cases, both cities have high numbers of confirmed cases and deaths compared to other state capitals. As of 15 May, 2021, São Paulo city had the highest number of confirmed cases (751,000) and Manaus had the seventh highest (approximately 173,000). 1 In addition, Manaus made headlines worldwide in January 2021 when hospitals ran out of oxygen supplies. It is also where the Brazilian coronavirus variant was first identified.
We divided the analysis into three stages. First, we systematized the scholarship and analyzed documents, particularly legislation and governmental reports. Second, we investigated pre-pandemic contexts in both cities using secondary data on demography, socioeconomic conditions, and inequalities. We used the following data: demographic, income, and geographic data from the Brazilian Institute of Statistics and Geography (Instituto Brasileiro de Geografia e Estatística, IBGE) and the UNDP Human Development Index (HDI); policy indicators using funding data from the National Treasury; data on social assistance capacity and service provision from the Social Assistance Services' Development Index (Indicador de Desenvolvimento do Centro de Referência de Assistência Social, IDCRAS); and data on education from the National Institute for Education Studies and Research Anísio Teixeira (Instituto Nacional de Estudos e Pesquisas Educacionais Anísio de Teixeira, INEP).
In the third stage, primary qualitative data on relevant policies were collected to better understand how policies were created or adapted to tackle inequality during the pandemic. These data were collected through semi-structured interviews with managers from state and municipal departments. The interviewees were selected based on their professional experience and expertise in this field. Sixteen interviews were conducted in total: two managers from the Amazonas Department of Social Assistance (state; interviewees 1 and 2), two managers from the Manaus Department of Social Assistance (municipal; interviewees 3 and 4), one manager from the Amazonas Department of Education (state; interviewee 5), two managers from the Manaus Department of Education (municipal; interviewees 6 and 7), two managers from the São Paulo Department of Social Assistance (state; interviewees 8 and 9), three managers from the São Paulo Department of Social Assistance (municipal; interviewees 10, 11, and 12), three managers from the São Paulo Department of Education (state; interviewees 13, 14, and 15), and one manager from the São Paulo Department of Education (municipal; interviewee 16). This research does not mention the interviewees' names or affiliations to maintain confidentiality and prevent possible professional and/or personal reprisals.
This analysis includes a triangulation of quantitative and qualitative data from different sources. Quantitative data analysis was crucial for contextualizing both states' and cities' socioeconomic inequalities and social assistance and education service coverage and capacity, as well as the quality of these services. This quantitative description allowed us to analyze the qualitative data collected during the interviews.

Brazil's multi-level policy systems: Social assistance and education
The 1988 Constitution included major changes in social policy. The new Constitution classified social assistance as a right. States were mandated to coordinate policies regionally, while municipalities were charged with implementing these policies with the assistance of national-level mechanisms. In education, the new Constitution reinforced a decentralized path but aimed to overcome a previous lack of coordination where states and municipalities had been responsible for service provision in a "dual" model with strong national coordination but little coordination between the state and municipal levels.
Since 1988, states have been responsible for secondary education, municipalities for childcare, and both for elementary education. The federal government was mandated to provide technical and financial assistance, while states and municipalities coordinated service provision. 2 The Calmon Amendment established a minimum spending standard of 25% of revenue on education.

Social assistance
Brazil's social assistance policy path is marked by paternalism, patrimonialism, punctual and fragmented actions, philanthropy, and charity-based programs, with a central role of first ladies as political leaders (Torres, 2002). The 1988 Constitution was crucial in transforming these fragmented actions into social rights (Sposati, 2005). The first post-1988 federal social assistance initiatives were developed during the Cardoso government (1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002), mainly through partnerships with non-governmental organizations. The Community Solidarity Program (Draibe, 2003) is a case in point. The first cash transfer programs appeared at the municipal level; different income transfer programs were then created at the national level.
During Lula's government (2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010), income transfer programs were unified and expanded with the creation of the Bolsa Família Program (Program Bolsa Família, PBF) (in 2003). PBF is meant to tackle beneficiaries' vulnerabilities and to overcome the intergenerational poverty cycle. The program combines cash transfers with education, health, and social assistance conditionalities. Municipalities are responsible for monitoring beneficiaries' continued eligibility for the program. They do this through the Unified Registry for Social Programmes (CadÚnico), created in 2001, which has become an important policy instrument for implementing PBF at the municipal level. Nevertheless, the most important change in terms of institution-building and the regulation of intergovernmental relations took place at the beginning of Lula's government when the central pillars of the Unified System of Social Assistance were defined (Sistema Único de Assistência Social, SUAS) (Bichir & Gutierres, 2019;Jaccoud et al., 2017).
These changes strengthened national coordination in three ways. First, they increased resources available for social assistance nationwide. Second, they put in place the regular and automatic transfer of funds to subnational governments, guaranteeing service continuity. Third, SUAS clearly defined each government's responsibilities. The federal government is responsible for approving national guidelines and standards, funding, and monitoring and evaluating nationally defined programs, as well as defining strategies for capacity-building at the subnational level. Municipal governments are responsible for service provision. States are responsible for regional coordination and for providing financial and technical assistance to municipalities and for the provision of higher-complexity services (such as shelters for street-dwellers). Unlike in education, there is no national minimum spending standard in social assistance. Nevertheless, regular and automatic transfers to subnational governments have functioned as a powerful mechanism to induce service expansion (Jaccoud et al., 2017).
This path led to a division between two important dimensions of Brazil's social assistance policy: (a) the SUAS, based on the provision of social services for vulnerable groups, and (b) direct monetary benefits such as the Continuous Cash Transfer Benefit (Benefício de Prestação Continuada, BPC, a constitutionally protected minimum-income benefit for extremely low-income elderly and disabled people) and the BFP.
At the beginning of the pandemic, Bolsonaro's government, with the support of the National Congress, created the Emergency Aid program to support low-income families (including PBF beneficiaries), workers who had lost their jobs, and informal workers. Initially, aid was approved for 5 months (April to August 2020). Most families received monthly payments equivalent to USD $113.20 (57% of the monthly national minimum wage); low-income, female heads of household received USD $226.41 (114% of the minimum wage). In September 2020, the federal government decided to extend aid for 4 months, but reduced the amount by half and tightened eligibility requirements. The number of beneficiaries dropped from 68 million in April to 43.6 million in September. Between January and March 2021, the federal government did not offer any financial assistance. In April 2021, Emergency Aid resumed with a smaller sum (USD $47.17) and even fewer beneficiaries (39.2 million).
The federal government's decision to implement Emergency Aid was political. It also bypassed SUAS's structure. Although the aid program was implemented with the support of the Ministry of Citizenship (formerly the Ministry of Social Development), citizens who met the income criteria did not need to be enrolled in CadÚnico to receive the new benefit. The Interagency Commissions (CIBs and CIT) were ignored. Furthermore, more than half of the beneficiaries of the first phase registered directly via a special mobile app without any contact with frontline municipal social workers. This allowed many vulnerable sectors of the population, especially those who live in remote places in the Amazon region or even in the giant slums of some big Brazilian cities, to join the program quickly. However, it also meant that the SUAS's accumulated knowledge in dealing with the multidimensional poverty was ignored, making it difficult to connect vulnerable populations with other services such as job searching, alcohol and child abuse prevention, and community relations, among others.

Education
Various national coordination mechanisms have been created since the Cardoso government, including national curriculum guidelines (Diretrizes Curriculares Nacionais and Lei de Diretrizes e Bases da Educação Nacional) and a fund to redistribute revenue according to the number of elementary school students in each state and municipality (Fundef, later Fundeb). Project grants were also created, and national evaluation systems, including standardized tests and school censuses, were strengthened. During the 2000s, different guidelines and project grants were created that focused on tackling inequalities in access to and quality of education, particularly targeting minority groups (e.g., illiterate youth and adults; rural residents; Indigenous, ethnic, and quilombola peoples) (Alves et al., 2021).
The strengthening of national coordination encouraged subnational governments to implement specific policies, thereby establishing minimum national standards, while revenue redistribution reduced fiscal inequalities between regions (Arretche, 2012). However, no CIB or CIT-style forums were created. As a result, institutionalized intergovernmental relationships remained vertical and limited to federal-local revenue redistribution, guideline-setting, and project grants.
In 2016, Dilma was impeached and Michel Temer (2016-2018) came to power. During his government, new restrictions were imposed on federal spending on social policies. This had implications on transfers to subnational governments, particularly for funding through project grants. The federal government's role in coordinating this policy area has changed even more since Bolsonaro's 2019 election (Abrucio et al., 2020). The Ministry of Education decreased or discontinued a few project grants and weakened its role in setting national guidelines. Moreover, the governmental agenda no longer includes "tackling inequality" and project grants have been eliminated.
During the pandemic, the Ministry did not issue public health guidelines related to education, nor were project grants awarded to subnational governments to facilitate distance learning or to increase access to and quality of education. On the contrary, national policies were focused on homeschooling, expanding military schools, and not teaching "gender ideology" in schools. Indeed, the Minister of Education came into conflict with subnational governments at various points. Finally, and significantly, the Ministry of Education did not spend its entire budget in 2020 (Todos Pela Educação, 2020). The lack of national coordination during the pandemic had an impact on education policy as the federal government did not adopt policies to respond to forced closure of schools, nor did the government push subnational governments to implement federal policies. This left space for municipal and state government agencies to make their own decisions about how best to respond to the pandemic. The two cities have different proportions of vulnerable residents, meaning that relative demand for services also differs. In 2018, 6.3% of the population of Manaus (135,000 people) lived on less than USD $1.90 a day. In São Paulo city, 2.6% did (317,000) (Cidades Sustentáveis, 2020). 3 In 2019, approximately 24,000 people lived on the streets in São Paulo, 11,693 of whom used social assistance facilities (São Paulo, 2020). A recent publication by the Prefecture of Manaus indicates a much smaller scale: social services attended to 806 people living on the street in the first semester of 2020 (Manaus, 2020). The pandemic likely made the situation in both cities far worse than these figures indicate. Although the pandemic has affected some groups more than others (Albuquerque & Ribeiro, 2021;Tavares & Betti, 2021), there is no doubt that food insecurity, poverty, and inequality have increased (UN ECLAC, 2020; Rede PENSSAN, 2021). For example, 116,700 families were registered for BFP in Manaus in December 2019 and 398,500 in São Paulo. By December 2020, this had increased by 16.17% and 20.61%, respectively.

Social assistance and education in Manaus and São Paulo
In terms of social service provision, 20 social assistance services were offered in Manaus in 2019 and 55 in São Paulo. Infrastructure had the highest capacity in both cities, followed by the human resources and service provision. However, although infrastructure capacity of public facilities ranked highest in both cities, capacity in the two cities remains highly unequal. In addition, Manaus has greater heterogeneity in human resource capacity than São Paulo. Finally, in Manaus, higher levels of IDCRAS 3 Calculated by the authors using the aforementioned percentage and official population estimates for each municipality in 2018 (IBGE's Estimativas de População). are visible in the southwest of the city, an area with medium and high HDI levels. In contrast, no spatial/geographical pattern can be observed in São Paulo ( Figure 3).
As for education, according to INEP's School Census, there were 981 basic education schools (kindergarten to high school) in Manaus and 7,219 in São Paulo in 2019. Most were public schools or schools partnering with the public system, meaning that they are publicly funded and offer free education. In Manaus, 86.03% of 480,000 children were enrolled in public or public-partner schools; in São Paulo, 76.4% of 2.19 million were. Most basic education teachers have an undergraduate degree. In 2019, 93.4% of students in São Paulo and 94.3% in Manaus were enrolled in schools where more than 90% of the teaching staff had an undergraduate degree. However, those figures change when it comes to graduate degrees or other courses. In the same year, only 27.4% of the enrolments in São Paulo and 11.4% in Manaus corresponded to schools where at least half of the teachers held a graduate degree; 22.6% of the students in São Paulo and 3.1% in Manaus attended schools where at least half of the teachers had completed a minimum of 80 hours of continuing teacher education or other complementary pedagogic training.
The Index of Development of Basic Education (Índice de Desenvolvimento da Educação Básica, Ideb) measures education outcomes (student performance in standardized tests, year-age adequacy, and retention rates) on a scale of 0-10. In 2019, São Paulo registered an Ideb of 6.3 for primary schools and 4.9 for middle schools. In Manaus, these figures were 6.0 and 4.8, respectively. Both cities have values above their targets for Ideb and are slightly higher than the country's overall Ideb values (5.7 and 4.6). However, there is significant disparity within each city. Schools' ranged from 2.2 to 8.3 in São Paulo in 2019 and from 2.9 to 8.4 in Manaus. Quotients of last over first quintile were approximately 1.15 in São Paulo and 1.2 in Manaus.
It is important to underscore that Manaus and São Paulo are located in regions of their respective states with distinct realities concerning Information and Communication Technologies (ICT), which might affect the capacity of subnational governments to cope with a situation that requires distance learning, such as the COVID-19 pandemic. In the North, where Manaus is located, 90% of schools reported challenges related to hardware and Internet connection in students' homes, while 80% in the Southeast, where São Paulo is, did (Comitê Gestor da Internet no Brasil, 2021, p. 66). The North has the lowest adoption rates for ICT measures to guarantee educational continuity: 49% of the schools reported having recorded videos for the students, 30% adopted synchronous online learning (e.g., on Zoom), and 31% used some kind of online platform to make learning resources available. Meanwhile, 88%, 75%, and 66% of the schools in the Southeast adopted those measures, respectively (Comitê Gestor da Internet no Brasil, 2021, p. 68).
In terms of fiscal capacity, São Paulo had the highest municipal revenue (R$60.15 billion) and the second highest per capita municipal revenue (R$4,939.78) in the country in 2019. In contrast, Manaus was in the mid-range among state capitals, with a per capita revenue of R$2,710.60. From 2013 to 2019, São Paulo was the least reliant on transfers among state capitals, while Manaus appears once again in the middle of the ranking. 4 This is significant in terms of spending and particularly spending on the policy areas of interest here. In this, two features stand out. First, the stability of spending over time (up to 2019) despite increases in poverty, inequality, and inflation, and, second, the difference between spending on education and on social assistance. This second feature can be explained by the nationally mandated minimum spending regulation (25% of revenue) in education, which does not exist for social assistance. As a result, from 2013 to 2019, spending on social assistance amounted to 2%-3% in both cities. There is evidence that social assistance funding decreased in 2020 (Licio et al., 2021).

Social assistance
Both São Paulo's state and municipal Departments of Social Assistance's previous policy pathways were fragmented. Programs were "divided" between different state and municipal departments and were a low priority, influencing resource allocation and capacity building. Nevertheless, innovations in subnational, mainly municipal, policies did influence changes in national policy, including cash transfers for vulnerable groups.
One of these innovations was the creation of the Minimum Income Program (officially, the Minimum Family Income Guarantee Program) in the municipality of São Paulo in 2001. This and other programs that emerged in different Brazilian municipalities around the same time were fundamental to the creation of the PBF at the national level (Leite & Peres, 2015). The Minimum Income was part of the "São Paulo Social Inclusion Strategy" (Pochmann, 2002), which, due to the importance of the municipality, had an impact on the federal government's decision to create the Ministry of Social Development in 2003. Ties to the Workers' Party (Partido dos Trabalhadores, PT) were crucial in this. The PT governed São Paulo between 2001 and2004 and then the federal government between 2003 and 2016. The state government, on the other hand, was never governed by the PT and opposed the creation of the PBF, although it implemented its own cash transfer program nevertheless. Unlike the PBF, recipients could only benefit from the state's Renda Cidadã, created in 2005, for a maximum of 36 months.
Before the pandemic, the state of São Paulo's social assistance policy focused on providing financial and technical assistance to help expand municipal social services. Even so, a few new programs were created, including the distribution of meals and milk, and the state agreed to manage contracts with shelters (interviewee 8). This is similar to Amazonas' Department of Social Assistance. Municipal policies in both cases consisted of providing social services and shelters. However, in the case of São Paulo, service provision has two important features: it includes significant non-governmental involvement and provides services that are excluded from SUAS guidelines (interviewee 10). At the beginning of the pandemic, both state and municipal departments had to elaborate their protocols to keep social services open since no national guidelines existed. One of the most important national regulations for social services was only published at the end of March 2020, 1 month after the pandemic began in most states. According to interviewees, bureaucrats from the municipal and state Departments of Social Assistance had to exchange information with other governments and formally or informally consulted with bureaucrats from the federal government to adapt municipal services protocols. These were also 4 Transfers are calculated as the sum of Capital Transfers (Transferências de Capital) and Current Transfers (Transferências Correntes). discussed with various councils and the public at large. Even though they sought to coordinate social services, "there was no unity; each municipality responded in a specific way.
[…] This lack of a pattern is a result of a broader lack of national coordination" (interviewee 10).
An important feature of the state of São Paulo's policies during the pandemic is related to the Secretary of Social Assistance's professional profile. Before becoming Secretary, she worked in a charitable organization and was very used to fundraising. The Secretariat was able to take advantage of this to collect millions of reais in private donations at the beginning of the pandemic, making it possible to provide meals to the vulnerable. The Secretariat also used donations to keep shelters open. The Secretariat's decision to turn to private donations caused some discomfort in the social assistance policy community. Although there is a provision in the SUAS for "contingent benefits" to address temporary vulnerabilities, these benefits are meant to be temporary. With the pandemic, some in the policy community became apprehensive about returning to charity policies, which had been common before the 1988 Constitution made social assistance a right.
The state of São Paulo also adopted a cash transfer program for vulnerable groups. The program, however, was not coordinated with the Education Department's transfers for students or other state cash transfer programs already in place (i.e., Renda Cidadã and Ação Jovem). By the end of 2021, the state had integrated a number of programs into a new social policy strategy (Bolsa do Povo). As in the past, however, policy discussions did not include social assistance policy managerial personnel and were restricted to managerial staff working closely with the governor (interviewees 8 and 9).
Manaus' and São Paulo's Municipal Departments had a greater role in providing, adapting, and expanding social services and shelters; distributing food products; and approving cash transfer programs. Municipal Departments did not coordinate with state and federal cash transfer programs in either case.

Education
From 2012 to 2016, with Rossieli Soares as Secretary of Education, the state of Amazonas embraced a policy path of expanding access to and the quality of education. Within this context, and since it was difficult to hire trained teachers to teach elementary and secondary courses in distant and remote areas (especially ribeirinhas and Indigenous schools), the state created the Media Center (Centro de Mídias). The program, which reached around 30,000 students every year, required the expansion of infrastructure and the development of materials for teachers and students. Since 2016, however, a series of Secretaries of Education have held office, some for only a few months at a time, creating policy discontinuity.
In São Paulo, political leadership in the education department has changed multiple times, and education policy has shifted as a result, even though the same party (Partido da Social Democracia Brasileira, PSDB) has governed since 1995. At the municipal level, different parties have been in control, including left and right-center parties, but during the pandemic, a mayor from PSDB was in office. Given this alignment with the state government, responses at the beginning of the pandemic were coordinated.
Just a few weeks after the pandemic forced schools to close, Amazonas created a program called Class at Home (Aula em Casa). The state was able to respond rapidly to school closures because of the Media Center, which had already prepared officials for distance learning. The speedy creation of Class at Home also included a partnership with the city of Manaus and cooperation with other governments (interviewees 5 and 7). This was key; if the Manaus Department of Education had waited for national guidelines and assistance, "nothing would have been done" since "the MEC [federal Ministry of Education] stopped" (interviewee 7).
By the time the pandemic began, Rossieli Soares had been the state of São Paulo's Secretary of Education for about a year. (He had also briefly served as federal Ministry of Education before moving to São Paulo.) Rossieli brought the Media Center and the idea of distance learning with him to his new post. With cooperation from the municipal Education Department, the state's Education Department created its own Media Center and developed content to broadcast online and on TV. Students could access online content without using up their mobile data, and SIM cards were distributed to vulnerable students to expand their studying hours. The São Paulo Education Department also elaborated protocols for opening schools and returning to class with the support of healthcare specialists. State schools re-opened for the most vulnerable at the end of 2020 (interviewees 13, 14, 15, and 16). State and municipal Education Departments overcame challenges related to student access to equipment and devices by distributing materials in person.
It is important to say that Amazonas and São Paulo's ability to adapt to distance learning was exceptional. Given the lack of national coordination, access to education through remote learning was not ensured across the country (interviewees 13, 14, 15, and 16).
Regarding schools' free meals programs (a key piece of the federal government's education policy, especially for the most vulnerable students), the government did not adapt its project grant or distribute pandemic guidelines. This led to fragmented policies (food distribution and money transfer) and unclear eligibility criteria (interviewees 13, 14, 15, and 16).
As suggested above, interviewees confirmed that subnational responses were not shaped by national guidelines and resources. Rather, state secretaries and CONSED staff shared information and discussed policies among themselves in both more formal settings and informally.

Subnational policies within a context of uncoordinated national responses
National coordination began to decline when Bolsonaro came to power in 2019. The pandemic-an external factor-exacerbated, but did not alter, this lack of coordination. Even though subnational governments created and adapted policies to respond to the pandemic, the lack of national coordination has produced a fend-for-yourself federalism resulting in fragmented subnational policies in these two policy areas.
Analysis of the two cases shows that diminishing coordination affected both Brazil's education and social assistance policy areas, although it was more significant for social assistance. This was in part because social assistance had been more affected by decreasing fiscal resources since 2016, and even more so since 2019; existing national regulations require that states and municipalities spend 25% of revenue on education, which provided some minimum guarantees for this policy area. More than this, although the existing information system (CadÚnico) and SUAS' capacities were important in guiding subnational responses, the fragilities of SUAS were laid bare. As well, the federal government's decision not to mobilize existing public social assistance facilities when Emergency Aid was implemented resulted in a lack of coordination between social services and cash transfer programs in both Amazonas and São Paulo.
The analysis of both cases demonstrates that, within the context of a lack of national coordination, policy legacies and previous capacity, in part the result of national-level coordination, are important to subnational responses, although they do not fully explain them. In social assistance, short-term emergency relief actions, such as short-term meal programs, have prevailed. Interviewees reported that this was essential for pandemic response, but it falls far short when addressing the more deeply entrenched and complex socioeconomic inequalities the pandemic has exacerbated. The policies municipal and state governments adopted included a variety of cash transfer programs, some with quite limited benefits, but they were not coordinated with the programs created by other governments, more notably the federal government's Emergency Aid. Shared responsibility but no coordination between the different levels of government created a situation where mayors and governors sought to claim credit for tackling social issues and blamed the federal government for not making the same effort. In both the state and city of São Paulo, political leaderships were not willing to use the existent capacity in this policy area to adopt and implement more sustainable policies. In the case of the state of São Paulo, leadership's career trajectory eased access to private donors and philanthropic resources, which shaped responses.
In education, subnational governments responded to the pandemic with policies that are more suitable for tackling inequalities in the medium term and long term, and municipal and state policies were more coordinated. Amazonas' response was possible due to previous capacity and in particular the Media Center. The Media Center, expanded during Rossieli Soares' mandate as state Secretary of Education, was adapted in the context of the pandemic by managers from the state Education Department. Rossieli Soares introduced the idea of the Media Center in São Paulo and mobilized actors-bureaucrats, other governments, and non-governmental organizations-and their resources to implement the Media Center and other initiatives during the pandemic. This means that, in both cases, local actors were key in activating previous capacity or mobilizing and convincing other actors about specific policy ideas. Furthermore, the previous existence of an institutionalized forum, CONSED, enabled formal and informal horizontal sharing of information and policies between bureaucrats and staff members of different subnational governments.

Conclusions
The pandemic influenced changes in both education and social assistance policy in São Paulo, Amazonas, and elsewhere in Brazil, to tackle some of the inequalities the COVID-19 pandemic has exacerbated. The context of weak national coordination has meant that fend-for-yourself federalism and fragmented subnational policies have prevailed. Since 2019, the federal government has unilaterally weakened its role in producing guidelines and standards and has eliminated some project grants, reducing the role of national coordinating mechanisms that had been in place since the 1990s. The federal government's decision not to offer technical or financial assistance to subnational governments and not to mobilize existing central coordination mechanisms during the pandemic, which goes beyond policy drift (Streeck & Thelen, 2005), led to un-coordinated responses and reinforced heterogeneity and inequality among subnational policies.
Even though service provision decentralization allowed subnational governments to adapt policies to local contexts and better tackle inequalities during the pandemic, they were not able to fill in the gap left by absent national coordination. In an unequal country like Brazil, deliberately uncoordinated policies tend to increase disparities among localities and social groups. In the context of the pandemic, where the living conditions of historically vulnerable groups have worsened, this political decision has created additional challenges for subnational responses that seek to tackle inequalities.
The degree of subnational coordination and its capacity to tackle inequalities over the course of many months depend on policy area legacies, previous capacity, and the role of actors within institutional contexts. Partisanship does not explain subnational responses to the pandemic in Brazil; instead, policy legacies and previous capacity explain the pandemic response. Even though, in social assistance, the SUAS, CadÚnico, and previous municipal and state policies had built subnational capacity to deal with some of the challenges that emerged during the pandemic, a shared responsibility model created a situation where different actors could claim credit for programs. Local actors were pushed to act, but relief responses were fragmented, uncoordinated, and only met recipients immediate needs. In education, funding and a more decentralized path allowed subnational governments to adopt more sustainable responses. However, these depended on the role of specific actors who activated subnational capacity and mobilized other actors around specific policy ideas.
This study highlights the importance of analyzing variations in social policy at the subnational level, including both national and subnational coordination mechanisms, and of considering the intersections between institutional contexts and agency. Future comparative studies are key to better grasping subnational policies, especially welfare policies; their heterogeneity and effects on inequalities; and the factors that explain their existence and content. This study also calls attention to the importance of further exploring institutional structures' activation, resilience, and capacity for change, and formal and informal relationships built during critical juncture periods.