Despite having no legal segregation or disenfranchisement based on race, Brazil’s political economy is still deeply tied to the history of slavery. This is both because of the plantation economies that supported oligarchic white family rule in the Northeast for generations, and because there had never been substantive efforts to mitigate the accumulated effects of slavery until the rise of the PT in the 2000s. Because universal primary care began to expand in Brazil in the late 1990s, becoming a focus for the Lula government after 2002, and focused specifically on building centers first in the poorest neighborhoods before expanding, one would hope not to find an association between race and the implementation of the primary care reform. Yet health outcomes are another story and we should expect white populations to have better health outcomes for quite some time.
The IBGE has collected data on race in the Brazilian states only since 2002. In most states the breakdown does not shift much over this period, but there is a gen- eral trend toward greater numbers of inhabitants identifying as black or mixed, across most states. Yet it appears that this shift is not primarily because of changing demo- graphics, but because of changes in the level of social stigma attached to blackness, with mixed race respondents who previously identified as white beginning to claim a mixed heritage. Because my purpose is primarily to control for the political legacies of racial oppression, small changes in racial composition over time are not especially meaningful. I therefore take the average white population of the past decade and apply it to the entire democratic period. Assessing whiteness as a residual category is more appropriate than developing an arbitrary theory about which particular racial
categories and subcategories matter the most for outcomes. Figure 3.1 presents the data, grouped alphabetically by the five regions in order to permit the reader a sense of the geographic shift.
Figure 3.1: Race in the Brazilian States
3.8 Democracy
Because most of my data are only available since the onset of the democratic transitions, a binary indicator for democracy provides little traction. Cumulative weights for democracy are the alternative, yet in the case of Spain are simply too highly correlated with other important variables that also increase steadily over time. Following the findings of Huber and Stephens (2012), I apply a 20-year democracy lag in Brazil. While governors were elected in 1982, not only was political competi- tion still constrained, but the literacy requirements that had disenfranchised half the
population even before the onset of military rule were not eliminated until after 1985. Therefore, in addition to no variation across regions, there are very few years still in the post-20-year lag. Ultimately, comparative historical evidence can illuminate the role of democracy in social and policy outcomes for regions within a given country, but it is hard to gain traction statistically after the onset of democracy.
3.9 Income Inequality
State and AC income ginis are used, but a note of caution is in order because the figures come from household survey data. The sample sizes are almost certainly smaller than they should be for truly reliable estimates and so variation from year to year, as with infant mortality, is likely excessive. Still, differences between units and broad shifts over time capture an important empirical reality. In Brazil, several years are missing (1991, 1994, and 2000) and so are interpolated. In both cases gi- nis are post-tax and transfer from household surveys—in Spain provided by a team of economists who have been working with the microdata data from the Encuesta de Presupuestos Familiares (EPF, Household Budget Survey) for many years (Go- erlich 2012; Goerlich and Villar 2009) and in Brazil from the Instituto de Pesquisa Econˆomica Aplicada (IPEA, Institute for Applied Economic Research).
In the case of Brazil, where improvements in income inequality have been ob- served overall, it is important to understand the difference between a national gini and subnational ginis. It is entirely possible for income inequality patterns within subnational units to vary considerably from the national gini because the unit for spacial comparison is completely different. In Brazil the subnational evidence sug- gests that we should not be overly optimistic about the national trend downward in income inequality. What we see at the state level is that income inequality increased in most states during the 1990s and then began to fall. Yet in the poorest and most unequal states, levels of inequality are still very similar to what they were in 1981
when the first indicators become available. In the South and Southeast, where income inequality was lower to begin with and incomes far higher, income inequality has im- proved the most. Figures A.9 and A.10 clearly show this pattern in the Northeast and in the South. What are the long term implications for equity if the worst-off places in the country are becoming pockets of persistent inequality, while the whitest, richest, and most egalitarian become more and more so?
3.10 Income and Growth
State and AC GDP per capita are used as a measure of income and, as with growth rates, control for local inflation. The Spanish data are constructed using GDP, census data, and the Consumer Price Index for each AC all published by the INE, and in Brazil are provided in constant reais directly by the IPEA.
Ideally we would test for the impact of poverty in both cases. In Spain, data are only available beginning in 2004 and for some ACs have sample size problems. Regardless, because primary care reform was mostly complete by 2004, the data are not useful for the Spanish model. In Brazil poverty data for the regions are provided by the IPEA and are calculated as double the rate of extreme poverty, based on a state-specific poverty line that quantifies the cost of a basic minimum basket of necessary purchases following WHO norms. The same years are missing as for the income inequality data (1991, 1994, and 2000). In 1986 the poverty data were unreliable for all states, nearly half the rates of the years on either side and dropping to zero in some cases, so I also interpolate 1986 for all states.