As it would be expected of a rather ample concept, HD is affected by and affects a long list of socioeconomic factors. The two-way relationship between economic progress and HD was charted by Ranis, Stewart & Ramirez (2000), Ranis & Stewart (2001), and Suri et al. (2011). Given past EG, families, governments and the civil society make allocative decisions on consumption (e.g., in better nutrition) and investment (e.g., saving for college) that are, to a greater or lesser degree, HD-enhancing. Naturally, EG is itself modified by family, government and civil society actions. The emerging picture is that of a deeply interconnected system of direct, indirect and feedback links shaping HD. With this complexity level, no automatic causation between income rise and the latitude of human choices should be taken for granted.

For the sake of parsimony, take the two sets of players that most directly affect HD. Income levels import for a country’s education and health indicators (e.g., Baldacci et al., 2008; Suri et al., 2011). However, families’ resources are not well depicted by country, state or local averages; income inequality has to be accounted for, as well as the extent to which poverty has been alleviated. In spite of a higher propensity to consume HD-related items, low-income families are doomed to a limited acquisition of such items. Notwithstanding, after controlling for income inequality and other quality of life aspects, income level tends to lose statistical significance as a predictor of morbidity and mortality (Kenny 2005). Moreover, in the decision dynamics within families, gender and schooling of household heads - those in charge of finances, often the family member who accrues the largest income share - make a difference. Women-headed households tend to spend more in HD-enhancing goods and services, such as groceries, water and education (Ranis, Stewart & Ramirez 2000); women’s education positively correlates with the probability of controlling domestic allocative choices and is itself a driving force for ameliorating nutrition and child survival rates (Schultz 2002).

Beyond improved allocation of available resources, the very nature of these resources is crucial. Departing from an income-centric perspective, Ranis & Stewart (2001, 6) defend that social capital, “the available horizontal links among actors”, may amplify the benefits from EG - or mute the consequences of a downtrend. Also, any reasonable account to the plunge in mortality rates over the 1900s would have to include knowledge, science and technology, argue Cutler, Deaton and Lleras-Muney (2006). Understanding of available technologies may cheapen HD inputs (e.g., preventive medical care, adequate nutrition and quality education), and the latter may reinforce each other through symbiotic connections. Consider, for instance, how vaccines and some inexpensive interventions (e.g., professionally training midwives) dramatically reduce health care costs (Ranis & Stewart 2001; Kenny 2005; 2009), and the potential boost in schooling, nutrition and self-esteem brought by conditional cash transfer programs such as Oportunidades (formerly known as Progresa) in Mexico and Programa Bolsa Família, in Brazil. Besides, there is evidence of a robust and bilateral association between education and health, in part supported by the fact that formally educated people are more likely to assimilate medical and sanitation technologies (Kenny 2005; 2009; for links between education and health, see Weare 2002; Oliveira 2005; Cutler & Lleras-Muney 2006; Feinstein et al., 2006; and Brunello et al., 2011).

Factors touched upon in previous paragraphs tell a story about a multi-causal phenomenon, whose determinants are of varied natures and find themselves intertwined by complex bonds. Put this way it is not surprising that income growth is only one facet of HD and that it may not guarantee the latter. However, there is more to it. Initial levels, of both income and other quality of life indicators, also dictate the speed of HD growth. The marginal effect of income on HD tends to decline (Kenny 2005), and an overall pattern of slow but consistent convergence has been found both cross-country and among Brazilian municipalities (Karruz & Vale 2015). Bounded indicators certainly play a role in reducing disparities - after all, no place can go above 100 per cent literacy rate, and present life expectation will not top the vicinity of 100 years (Kenny 2005; Noorbakhsh 2006).

Most of the literature on convergence deals with trajectories of per capita income among countries (e.g., Barseghyan & Dicecio 2011; Dhongde & Miao 2013), states (e.g., Galeano 2014) and municipalities (e.g., Coelho & Figueiredo 2007; Ferreira & Cruz 2008; Ribeiro & Almeida 2012). A smaller group of scholars has applied the notion of convergence to report HD trends, and studies of this nature typically analyse the country level (Kenny 2005; Noorbakhsh 2006; Kónya & Guisan 2008; Mayer-Foulkes 2010; Kónya 2011; Jordá & Sarabia 2015; 2015).

Convergence studies share the concern about spatial inequalities and the speed at which less advanced regions seem to catch up towards the highest development levels experienced by other regions. Most of these studies cite the seminal works of Robert Solow (1956) and Trevor Swan (1956). These authors proposed neo-classical growth models predicting that, given perfectly free markets, capital investments would travel the world after the highest return rates. With capital’s diminishing returns, best investment opportunities would concentrate in the least developed areas, where the stock of capital of capital is lower. According to this view, regional disparities would disappear, and income convergence would be expected. An alternative perspective foresees an entirely different outcome. As per the cumulative circular causation principle, capital and skilled labour would cluster in certain areas due to economies of scale and agglomeration, possibly generating a pattern of lingering inequality. This view has been associated with the work of Myrdal (1968), Dixon and Thirlwall (1975) and Kaldor (1978), explain Marques and Fochezatto (2012). Note that income convergence accounts provided above regard markets or territorial units much more so than a convergence of families’ income.

As an extrapolation from the income studies, HD convergence claims that diminishing marginal returns apply to non-economic dimensions as well. Here, the insight is that basic education and health services (e.g., primary education, immunisation) tend to cost considerably less than their advanced counterparts (e.g., higher education, open-heart surgery). With the same dollar amount, a larger quality of life return would accrue in less developed areas, which are still struggling to meet basic and less expensive demands. HD convergence could then take place (Kenny 2005; Noorbakhsh 2006).

Empirical research has found slow HD convergence between countries (Noorbakhsh 2006; Kónya & Guisan 2008; Jordá & Sarabia 2015). Considering each HD dimension separately, Mayer-Foulkes (2010) observed a series of superposed transitions marked by initial divergence followed by periods of convergence, with urbanisation playing a central role in promoting the latter. Social policies are arguably more straightforward to implement in urban than rural areas (Kenny 2005), and urban dwellers are more likely to send their children to school, due to the typically facilitated access and lower private costs of education (Gupta, Verhoeven & Tiongson 2002).² Speed of convergence seems to vary across dimensions, and has been more accelerated in schooling (Neumayer 2003; Mayer-Foulkes 2010; Jordá & Sarabia 2015). In the Brazilian context, municipality-level HD has been predominantly driven by the schooling dimension (PNUD, IPEA, & FJP 2013).

Primary education and health care, two key HD-inputs, are policy areas in which Brazilian municipalities take centre stage - and, thus, where service heterogeneity is prone to arise. Next section reviews the literature on the HD implications of public spending in these areas. It also briefly describes the social services provision landscape in Brazil, touching upon topics of regulation and funding.

Governments play a crucial role in fostering HD as both education and health are considered public goods. Total budget and apportionment by function, as well as the distribution of funds between various HD-related programs, are all impinging upon HD results (Ranis & Stewart 2001).

Based on panel data from 118 developing countries covering years 1975-2000, and after applying several techniques to avoid measurement error and endogeneity,³Baldacci et al. (2008) found that expenditures on education positively associate with accumulations of education capital (primary and secondary gross enrollment rate). Such connection works through both contemporaneous and lagged effects, the latter spanning 10 years. At their turn, positive effects of health spending do not seem cumulative, being detected only a contemporaneous (five-year period) influence over the under-five child mortality rate. Nevertheless, such positive effects of education and health spending are contingent on the quality of governance, portrayed by a combination of corruption and accountability indicators. Poor governance reduces the effects of education spending and nullifies those from health spending, report Baldacci et al. (2008).

Suri et al. (2011) found similar results as they observed panel data ranging from 1960 to 2001 on around 100 developing countries. Their dependent variable was a composite index representing the portion of the variation of three HD indicators (infant mortality, life expectancy, and secondary school enrollment) that is not attributable to a country’s income growth rate alone. Per capita public expenditure on education and health both showed a positive effect on HD, whereas income inequality displays a strong adverse effect.

Public education and health care are highly decentralised, universal services in Brazil. Public policy decentralisation gained strength in the 1990s, in good part because of the 1988 Constitution, which was built around a political commitment to empower local communities and to expand municipalities’ taxation power. Later in that decade, new measures were taken, but now emanating from a different set of motivations. Brazilian municipalities are extremely unequal - in population size and composition, economic dynamics, policy needs, managerial capability - and started in very different positions as far as their participation in supplying social policies. To help levelling the field and to overcome local fiscal constraints, the federal government passed constitutional amendments to establish new policy funding designs (Souza 2004), keeping service delivery (policy-making) local, but concentrating policy decision-making at the central level, especially in health care and education (Arretche 2012).

A system of incentives (e.g., new funding availability) and sanctions (e.g., risk of resource loss or impossibility to tap into new sources), combined with strong regulation and service protocols, are arguably the critical design features behind the successful municipalization in primary education and basic public health (Arretche 2002; 2012; Souza 2004). Since Sistema Único de Saúde (SUS, the Unified National Health System) was established in 1988, municipalities have opted to become responsible for health services provision. Many have also adhered to the 1998-launched Fundef (Fundo de Manutenção e Desenvolvimento do Ensino Fundamental e de Valorização do Magistério, loosely translated as Fund for Primary Education Maintenance and Development and Recognition of the Teaching Profession). In 2000, 99.0 per cent of municipalities had taken over health services supply, whereas municipal schools accounted for 51.4 per cent of primary education enrollment - against 40.7 per cent in 1997 (Souza 2004).⁴

Some key legislation deepened decentralization in the following years. In 2000, Amendment 29 regulated local revenues to be spent in health: a minimum of 7 per cent that year, with yearly increments, reaching 15 per cent by 2004 (Faveret 2003). In 2007, Fundef was replaced by another intergovernmental fund, Fundeb.⁵ Municipalities’ contribution to the pooled resources increased from 15 to 20 per cent of revenues, as childcare/pre-school and secondary education were included in the Fundeb per pupil funding calculations. Note that since the 1988 Constitution, municipalities were obliged to spend at least 25 per cent of collected taxes on education-related services (art. 212).

In the health front, required minimum investment levels combined with conditioning transfers of federal resources to the supply of specific services raised ambulatory and family health care, as well as total health expenditures, and lowered funding inequalities across the country (Vazquez 2014). In the education domain, Fundef led to the expansion of enrollment and reduction in intra-state per pupil spending inequality (Vazquez 2014). Nonetheless, findings have been ambivalent on the association between municipalization and educational outcomes - for instance, student performance at standardised tests (Leme, Paredes & Souza 2009; Ceneviva 2012). Taking Prova Brasil⁶ scores as the dependent variable, Domingos, Silva and Souza (2013) found no statistical effect of municipality spending, while Amaral and Menezes-Filho (2008) observed a positive and significant effect only for the communities with top performance. One potential explanation formulated by the latter authors is that a certain level of preexisting managerial capability is required to transform additional resources into quality education. Along the same lines, Diaz (2012) was unable to detect an impact of local expenditures in education on IDEB (Índice de Desenvolvimento da Educação Básica, the Basic Education Development Index), a composite measure that combines both student promotion and learning rates. On the other hand, Gonçalves and França (2010), Amaral (2011) and Kroth and Gonçalves (2014) observed a positive influence of local spending on student performance. From a rather different viewpoint, Soares and Alves (2013) estimated school and municipality effects on Prova Brasil scores as the residuals of a hierarchical linear model. These effect estimates were poorly correlated with both per-student costs and the municipality socioeconomic status. Such a result indicates the existence of differences among local school systems (potentially regarding more or less adjusted management policies and instructional methods) that should not be confused with other, more visible cross-municipality disparities.

Decentralisation of public policy created a heightened need for professionalisation of local public management. To learn about both personnel technical qualification and potential obstacles to the institutional consolidation of local bureaucracies, Veloso et al. (2011) surveyed public managers from 194 city halls of seven Brazilian states. Inspired by the Government Performance Project (Ingraham, Joyce & Donahue 2003; Ingraham 2007), which assessed the public management capacity of federal agencies, state governments, counties and large cities in the United States, the survey addressed four management capacity components, or subsystems: financial, capital, people, and information technology. Veloso et al. (2011) define management capacity as the competence of a local Executive to meet its goals and legal obligations, according to the public interest, efficiently and efficaciously, deploying available resources through adequate management instruments; the concept encompasses the ability to coordinate efforts with other governments and partners too. Some of Veloso et al. (2011) findings include a high level of financial management outsourcing (which the authors deem a potential threat to sustained good fiscal performance); poor technical qualification of the teams in charge of capital investments; and ample disparities in people management strategies.

Fundeb and SUS-related intergovernmental transfers are only two of several funding sources for local governments. Soares & Melo 2016 group local revenues as follows:

• Those collected directly by municipalities (own revenues), mostly derived from property taxes and taxation of professional services;

• Constitutionally mandated transfers, such as Fundo de Participação dos Municípios (FPM, which redistributes the federal income tax and the tax on manufactured products), municipalities’ share of the state-level value-added and vehicle taxes (cotas-partes), and Fundeb;

• SUS-related transfers;

• Other legally mandated transfers, including some earmarked resources (e.g., from the Fundo Nacional de Desenvolvimento da Educação, the National Fund for Education Development), and oil royalties;

• Discretionary transfers from federal, state or other municipalities’ budgets.⁷

Soares & Melo 2016 classify these five main sources according to their associated level of fiscal autonomy and potential to reduce inequalities. Own revenues enjoy high autonomy and low redistributive potential since they reproduce in public coffers the vigour of each local urban economy. That is a particularly relevant point since the present work focuses on communities whose economy has grown less than expected. If performance on social indicators exceeds expectation, and if national norms concerning fiscal federalism are homogeneously enforced, two of the possible accounts would be that these municipalities are levying more of their citizens and/or better spending their revenues. In a scenario of an economic downturn, the more dependent own revenues are on the household income levels (e.g., the more reliant on services versus property tax), the more acute would the tax burden be on family budgets, all else constant, ultimately threatening families’ HD-enhancing consumption. As per the second explanation, the increased share of municipal taxes could imply stronger state capacities (e.g., those involved in keeping taxpayer registries up-to-date, collecting outstanding balance). From this viewpoint, Soares & Melo 2016 use own collections as a proxy for both municipality wealth and technical capabilities; Marenco, Strohschoen and Joner (2017) observed that in municipalities where more than 15 per cent of public servants are college graduates, the odds that the property tax’ share of total revenues is larger than one-fifth are 80 per cent higher.

Among the constitutionally mandated flows, the FPM was designed as a redistributive fund, whereas state-level cotas-partes apportion sums proportionate to each municipality’s contribution to the total collected; like Fundef, Fundeb was meant to be redistributive, especially within states and, in contrast to FPM and cotas-partes, is earmarked. SUS transfers and most of the other legally imposed ones are redistributive by nature; observe that source d) comprises both earmarked and freely allocated funds. Finally, because discretionary transfers are intended to support specific projects pre-approved by the ceding government (such as building bridges and water and sewer systems, purchasing ambulances), they offer little autonomy, and their redistributive potential is yet to be determined. Available evidence indicates that federal discretionary transfers favour municipalities with a higher proportion of own collection (Soares & Melo 2016); these tend to be the bigger cities.⁸

Nunes (2015) explores the patterns and rationale of federal executive, discretionary transfers in countries with varying levels of centralization. Data from Latin America indicate that the more decentralized the nation, the more likely that these transfers favor municipalities governed by co-partisan mayors, even after controlling for a series of alternative explanations, including targeting poverty alleviation. Drawing findings from a regression discontinuity design applied to the Brazilian case, Brollo and Nannicini (2012, p.759) had also found a heavy political motivation to these transfers, notably “aimed at penalizing unaligned municipalities in which mayors belonging to the opposition coalition won by a narrow margin.”

Nonetheless, there seems to exist other goals guiding allocation of discretionary transfers. Bueno (2017) identified a bypassing strategy, by which central government distributes resources to municipalities with unaligned mayors through transfers to nonstate public service providers.⁹ Such transfers are even less visible to voters than are those to mayors, the former incurring lesser of a “credit highjacking” risk. Beyond partisan alignment, incumbents (presidents, governors) care about boosting voters’ welfare, gaining legislative support, mobilising swing voters, and even exchanging transfers for campaign contributions, explains Bueno (2017).

Efficiency in the application of intergovernmental transfers is most certainly a concern in a system where average municipality’s share of own revenues is no larger than 12 per cent of the total budget, warn Costa and Castelar (2015, based on 1999 to 2009 data). These authors set out to test the flypaper hypothesis,¹⁰ which predicts unconditional grants to “increase public expenditures by more than an equivalent increase in income from other sources,” saving the popularity cost of taxing one’s own constituency and opening space for less rigorous fiscal policies. Costa & Castelar (2015) found no evidence of the flypaper effect in Brazilian municipalities, even after analysing different expenditure quantiles.

While the flypaper hypothesis posits lower fiscal responsibility the larger the share of transfers, one has to consider another narrative, valid to discretionary transfers in particular. To be granted, these transfers require grantee municipalities to elaborate a work plan for the intervention being funded, many times involving an engineering project, an environmental permit and other documents necessary for public works, thus potentially increasing room for accountability.¹¹ Beyond that, Bertholini, Pereira and Rennó (2018, 9) point that discretionary transfers may be a somewhat efficient form of pork because interested politicians will closely monitor project execution, making sure that “promises will be accomplished to credibly claim credit for them.” These authors found that transfers originated from legislator-authored amendments to Brazilian central government’s annual budgets can generate social and economic inclusion; however, effects tend to dissipate over time, potentially due to policy investments’ diminishing returns and inflow concentration into specific locales.

Programa Bolsa Família (PBF) is a federal conditional cash transfer initiated in 2004 that benefits around 14 million families across the country. Devised during President Luiz Inácio Lula da Silva’s first term, PBF replaced and integrated a series of other transfers. As a conditional cash transfer, PBF seeks to a) reduce current poverty and inequality, by providing a minimal income inflow to low-income families; and b) break the inter-generational transmission of poverty, by establishing incentives for these families to invest in human capital. Transfers are conditioned on beneficiary compliance with requirements (conditionalities), such as school attendance, immunisation, prenatal visits, and participation in nutritional programs.

PBF eligible families are identified according to household per capita income (no more than R$ 170 monthly, approximately US$ 50 in 2017 values) and household composition. A portion of the benefit varies according to the number of children and adolescents, pregnant or nursing women, and the per capita income of extremely low-income families (those living with less than R$ 85, even after receiving the regular PBF transfers). In December 2017, the average PBF benefit was about R$ 180 per family (MDS 2018).

A single registry (Cadastro Único) contains information about current and potential PBF beneficiary families. Municipalities inscribe families (who actively register or are reached out), feeding Cadastro Único with self-declared data about their means, household composition, etc.; this registry is periodically updated and checked against official records, like those related to job participation. Proper maintenance of the database is critical in the reduction of gaps and duplications and in avoiding fraud.

Throughout the last decade, several impact studies have attested that PBF has helped improving living conditions. In particular, income transfers account for 24 per cent of the per capita income inequality reduction between 1995 and 2006, estimated Silveira Neto (2010). After surveying numerous impact studies, Jannuzzi and Pinto (2013) ascertained favourable effects of PBF over a range of socioeconomic indicators, such as increased school attendance, improved performance at large-scale tests, reduced grade retention and child labour. There has also been detected a decline in acute malnutrition and infant mortality, and a rise in immunisation rates, pre-natal care and birth weight. As far as the job market, evidence dismisses the hypothesis of a lowered labour supply on the part of beneficiaries, except for women with children. Furthermore, women - who comprise the almost totality of family members entitled to withdrawal the cash transfers - show elevated decision power in household allocation choices (e.g., regarding the purchase of durable goods, medicines, and clothing, as compared to nonbeneficiaries), which helps to reduce intra-family gender inequalities through empowering.

The above search for influences over HD has spread to different disciplines (Economics, Political Science, Education, Public Health), encompassing a great variety of references and causal claims. To organise the stated arguments and assist modelling, Chart 1 presents an inventory of such influences. Given the plethora of factors interfering with one’s health and schooling, not to mention the interactions between such factors, this list is not exhaustive; on the contrary, it aims to register only the main associations explored by the reviewed literature.

Chart 1
Influences over HD that have been identified in the reviewed literature

The main argument tested in the analyses ahead is that public policy from central and local governments can have substantive effects on HD. In particular, variation in discretionary transfers from the federal and state governments to municipalities, and the HD-oriented spending of local administrations, are expected to display a positive effect on HD, independently of the local economy’s vigor. In this sense, social spending could help counteract income fluctuations’ negative effects on living conditions.

The empirical analysis ahead draws information from four sources. Data on 2000 and 2010 MHDI, its dimension-specific indexes¹² and other demographic and socioeconomic indicators were extracted from Atlas do Desenvolvimento Humano no Brasil - an online database built by the Brazilian United Nations Development Programme office and partners, assembling HD-related data for multiple subnational territorial units. Local budgetary information was retrieved from Finbra – Finanças do Brasil, which publicises revenues and implemented yearly budget values as per declared by municipalities to the Secretaria do Tesouro Nacional, a division of the Brazilian Finance Ministry. Other budgetary indicators come from Rio de Janeiro’s Federation of Industries, Sistema Firjan. The fourth source is the website of the Ministério do Desenvolvimento Social (Ministry of Social Development), from which data on PBF were downloaded through the platform Matriz de Informação Social (the Social Information Matrix).

All these pieces of data were combined and organised as a panel whose unit of analysis is municipality-year, covering the 5.565 existing municipalities in 2010. Years include 2000 and 2010, except for budgetary and PBF variables - for which decade averages were calculated based on yearly values.

Two major analytical steps are undertaken in this work. First, identification of the municipalities with slower-than-expected income growth and, within these, those with either above- or below-expected trends in schooling and longevity.

Grouping municipalities according to changes in their social indicators is a variation of the approach used by Ranis, Stewart and Ramirez (2000) and Suri et al. (2011). Both studies classify the trajectory of developing countries in the second half of the twentieth century into four categories: (i) virtuous cycle, for those showing above-average advancements in both HD and EG; (ii) vicious cycle, for countries performing below average in both dimensions; (iii) EG-lopsided, for cases with worse-than-average changes in HD but above-average EG; and (iv) HD-lopsided, the symmetrical category for (iii). Empirical evidence indicates that “policies to improve HD must precede or at least complement growth-oriented policies if growth is to be accelerated and sustainable” (Suri et al., 2011, 519), a finding with direct implications over policy sequencing.

The present study focuses on categories (ii) and (iv), where EG is less than expected. Specifically, it aims at finding out what sets (ii) and (iv) apart. Differently from Ranis, Stewart and Ramirez (2000) and Suri et al. (2011), here the HD dimension is broken down into the longevity and schooling components, each with its own (above or below expectation) trajectory over the 2000s. This approach allows for more granularity in the investigation on potential influences of governments’ spending on health and education standards.

A well-documented pattern of HD-convergence, discussed earlier, provides the theoretical grounds for estimating annual growth rates for each MHDI dimension. Equation (I), a simple convergence model adapted from Jordá and Sarabia (2015), was applied to calculate unconditional convergence rates and the respective predicted growth rate for municipality i. Y represents each of the MHDI dimensions.

Figure 1, below, portrays scatter plots of the convergence analysis’ dependent and independent variables, as well as of the general MHDI. A negative slope is visible in each panel, pointing out that the larger the 2000 value, the smaller the expected 2000-2010 growth, all else constant.

Figure 1
Growth rates by initial value: MHDI and its dimensions (2000-2010)
Note: N = 5,565 municipalities in all four panels (1.a to 1.d).Source: The author, based on raw data from Atlas do Desenvolvimento Humano no Brasil.

All four indicators analysed show a convergence pattern, as per the estimated slope coefficients (Table 1). Convergence seems most intense in schooling and least in income; still, initial income dimension levels solely explain 53.9 per cent of variation around the mean. All else constant, a one per cent increase in 2000 Income MHDI associates with a 0.029 per cent decrease in its expected annualised 2010-2000 ratio.

Table 1
Regression estimates from the convergence models (2000-2010)

Municipalities with 2000-2010 observed income growth rates smaller than the lower limit value of a 95 per cent confidence interval around the predicted dependent variable are kept in the working sample, all other ones being removed from it. Following the same tactics, are then excluded localities whose observed longevity (schooling) growth rate falls within the estimated confidence interval for the respective dimension. The sample is reduced to 2,482 municipalities, each classified in one of the four groups described in Chart 2.

Chart 2
Grouping of municipalities

Group 1 congregates cases where all three dimensions grew less than foreseen and are nicknamed double social loss group. Group 4, on the other extreme, is the double social gain one. Groups 2 and 3 combine some social gain with some social loss. Because no arbitrary judgment regarding the relative values of longevity and schooling are made, it is not possible to order the four groups based on their desirability, although 1 and 4 are the least and most favourable clusters.

The second analytical step gets to the core goal of this work: to map characteristics of family and policy environments, as well as of the demand for social services, that seem to associate with an elevated chance to escape the double social loss situation. To that pursuit, a multinomial logit model is calculated as in Equation (II).

Multiple equations (J - 1) are estimated, being J here equal four, the number of groups. The dependent variable is the log-odds of falling in group j relative to J, the referent one - set to be the double social loss group. A constant (α) and a series of linear slope coefficients (β) are predicted in each equation. Vector X contains the independent variables (see Chart 3), many of which are paired with a respective measure of change within the 2000s (to account, where possible, for variation within the relatively long ten-year time window). X also includes a set of dummies for states, to control for unobserved cross-state heterogeneity that is constant over the 2000-2010 period;¹³ ε is the stochastic term. Robust standard errors are estimated to tackle heteroscedasticity. By exponentiating the estimated parameters, one can project the influence of each explanatory variable in the risk of pertaining to a given group relative to the referent one, all else constant. The relative risk ratio (RRR) consists of a ratio of two probabilities, in this case, the probability to belonging to group j (π_ij) over the probability to belonging to the double social loss one (π_iJ); is interpreted as a multiplicative effect on RRR.

Chart 3
Explanatory variables in the multinomial logit model

Chart 3 (Cont.)
Explanatory variables in the multinomial logit model

All else constant, the expectation is that municipalities with a lower income or a higher Gini coefficient (a measure of per capita household income concentration) will suffer more from a slow economic scenario since many families will have insufficient resources to cater for their education and health care needs. The higher the per cent of children (up to 14 years old) living in households without a middle school graduate, the lower the expected advancements in health and education. Such variable reports the share of children in households where the cultural capital is underprivileged, with potential implications on adults’ employability and resourcefulness - e.g., in helping the schooling of children and fostering their health.

The second set of covariates concerns the potential of local governments to execute public policies, through the lenses of resource availability and application. It controls for per capita total revenues in 2000, as well as by its average year-to-year per cent growth rate. Importantly, given the constitutional floor percentages to be engaged in education and health, the higher the total revenue, the higher the nominal values deployed in such functions. Total revenues are then broken down considering: a) how much originates from own collection, and b) how much comes from the federal and state governments in the form of discretionary transfers. The hypothesis is that a) positively correlates with town hall’s capability to levy taxes (e.g., via higher tax burden, more efficient collection), which could indicate a highly skilled bureaucracy (Marenco, Strohschoen & Joner 2017) - similarly capable of executing well-adjusted social policies. As far as b), the hypothesis is also in the positive side: while discretionary transfers are not necessarily redistributive, they require a good level of bureaucratic resourcefulness from local governments, and here too an expectation regarding overall administrative capability is tenable; not the least, the very projects funded by discretionary transfers may be HD-enhancing (Bertholini, Pereira & Rennó 2018). From the expenditures perspective, local investment in education, culture, health and sanitation should positively influence social indicators, both directly by the primary function and indirectly, through the effects of one’s good education on health status (and vice versa).

Alongside the detailed budgetary variables, the Índice Firjan de Gestão Fiscal (IFGF, the Firjan Fiscal Management Index), calculated by the Federation of Industries of the State of Rio de Janeiro, is taken both as a measure of resource availability and as a proxy for state capacity (e.g., to collect taxes, formulate efficient policies). IFGF was conceived to portray challenges faced by local administrations across the country, especially those regarding resource insufficiency and the burden of debt. It comprises five indicators, all ranging from zero to one (the greater, the better the fiscal management), just like the composite IFGF itself. IFGF touches upon the following dimensions: capacity to obtain revenue (own collection); capacity to make investments; the degree of budgetary rigidity (proportion of revenues used to pay personnel); fund availability to meet short-term obligations; and the cost of long-term debt (Sistema Firjan 2015).

Additionally, the governments’ set of variables includes two measures regarding PBF. One concerns program reach (in relation to the number of target families); by fighting extreme poverty and imposing education and health care conditions to participation, the program is expected to improve HD among the most vulnerable.¹⁴ The other corresponds to the Índice de Gestão Descentralizada (IGD, the Decentralized Management Index), which should help isolate the influence of PBF coverage from occasional implementation problems (of dated information in the Cadastro Único and faulty conditionality monitoring). IGD was designed to support a performance-based financial incentive paid by PBF to the municipalities. The goal was to encourage local governments to keep Cadastro Único up-to-date and to oversee the meeting of program conditionalities, explain Soares (2012) and Paiva et al. (2016).¹⁵ IGD ranges from zero to one; the closer to one, the better the PBF management at the local level. Furthermore, as discussed above, beyond the sheer amount of available resources, aspects of quality in their application make a great deal of difference in fostering HD. Quality may refer to more or less adequate choices made among policy alternatives, which entail varying levels of efficiency, efficacy, effectiveness and social justice. It may also regard operational capability to implement the chosen policies, bringing to life expectations written down in intervention proposals and laws. IGD is included as a control variable that also accounts for at least part of the variation in local administrations’ competence to devise and execute quality actions.

As far as the demand for education and health services, a younger population imposes a stronger pressure on the primary and secondary schooling system. It may also signal that ageing is less likely due to, for instance, urban violence. Conversely, a larger per cent of 6-17 year-olds may represent a type of demand that is “easier” to serve, since Schooling MHDI indicators of that age bracket mostly regard enrollment, not graduation.¹⁶ Furthermore, a younger demographic profile may also indicate reduced infant mortality and a prevalence of (cheaper) prophylactic (versus more expensive therapeutic) treatment needs. Similarly, a smaller share of senior citizens demands less of the health care services, especially those to cope with chronic conditions (e.g., high blood pressure, diabetes) and of higher complexity. Because the relationship between per cent of 6-17 year-olds and social progress could defensibly go either way, no specific hypothesis is defined for this covariate.

Moving to the variables that control for the number of inhabitants and their spatial distribution, the more urbanized the municipality, the more likely it is to perform well on HD, as previously discussed. With respect to the sheer size of population, its influence could also go both ways: variety and complexity of demands tend to be lower in smaller localities, however so too tend to be the availability of skilled government staff and response procedures to emerging needs. The controls for initial HD level are expected to return positive coefficients, given the convergence trends explored earlier; to avoid having a proxy of the explained phenomenon among the independent variables, no “change” analogues were defined for these two covariates.

Multinomial models rest on the independence of irrelevant alternatives (IIA) assumption, which requires that “adding or deleting outcomes does not affect the odds among the remaining outcomes” (Long & Freese 2001, p.188). To meet IIA, outcome alternatives must be dissimilar and not merely substitutes for one another. This is a somewhat vague advice, and empirical tests developed for IIA will not offer much more guidance. Long and Freese (2001) discourage the use of such tests (e.g., Hausman and the Small-Hsiao) because they often reach conflicting results. In this study, given the fact that the four groups do not cover the entire range of possibilities - they do not include municipalities that fall within confidence intervals - it is argued that these groups are not necessary substitutes.

Another assumption of the multinomial models is that the outcome levels (groups) have no natural ordering. On a HD scale, it is clear that double social loss fares more poorly than double social gain, however relative desirability between the two intermediate groups is much less evident. To test whether the two latter groups should be meshed together - actually, to assess if any pair of groups should be combined - Wald tests were conducted; results (not shown) indicate that none of the four groups should be fused, at the 0.1 per cent level of significance.

The identification strategy for the association between each covariate and the probability that the municipality will fall in a certain group relies on: a) differential levels of the independent variables as a source of variation among municipalities; and b) an extensive set of controls covering demand for HD and initial HD levels. As this work’s ambition is to explore associations, not to explain them, such identification strategy is deemed compatible with the present research goals.

A challenge to internal validity is the fact that, concomitantly to the resources and expenses visible in local budgets, state and federal governments are acting through direct investments that affect HD (e.g., expanding the high school system, building/ improving interstate highways). For such investments not to bias estimated parameters, it would require to admit that the former are homogeneously or randomly distributed; unobserved time-invariant cross-state variability is controlled for by a set of dummies for states.

For the multinomial estimation, the number of observations is constrained by Finbra – Finanças do Brasil and IFGF data availability, which forces the exclusion of 137 (5.52 per cent) localities from the initial sample (n = 2,482).¹⁷Table 2 presents the mean value in each group for covariates in the multinomial model and other selected variables, considering the working sample (n = 2,345).¹⁸

Table 2
Means of explanatory and other selected variables, by group

Table 2(Cont.)
Means of explanatory and other selected variables, by group

The “change” Gini variable depicts (in Table 2) a scenario of decreasing per capita household income inequality over the 2000s. Noteworthy, per cent of children in households with no middle school graduates fell 20 per cent points or more in all four groups. Budgetary indicators tended to change little from one year to the next, except for the per capita total revenues, which grew on average 7.24 per cent each year, net of inflation.¹⁹ Per cent of own collection revenue and per cent spent in health and sanitation increased by 0.26 and 0.59 per cent point yearly, on average. On the other hand, the share of discretionary transfers and expenditures on education and culture tended to shrink: 0.21 and 0.24 per cent point a year. Initial IFGF components tended to be smaller (greater) in group 1 (group 4) than in the other groups.

In the working sample, age profile seems to have grown older in the 2000s, with a 5.65 per cent point average decline in the per cent of the 6-17-year-old population over the decade. Urbanisation advanced 4.82 per cent points in ten years, on average, while the total population increased 10.19 per cent between 2000 and 2010, on average. Initial Schooling MHDI levels in 2000 were less than half those of the longevity dimensions.