Analysis of census data about income differences by color and their links with education. According to Laura Randall,
Some of the findings of the article are that the difference in income between self-identified browns and blacks increased from 1980 to 1991, but fell from 1991 to 2000; by that year, they were close enough so that blacks and browns, as regards income, could be considered to be a single group; but they could not be considered as only one group before then. The factors explaining the difference in the income of the two groups are shown, as they are for the difference in income between whites and the “blackplusbrown” group for the 2000 census. Detailed analysis of the role of years of schooling is presented.
I would like to respond to Sergei Soares’ response to my comments, because I think that the logic in back of our disagreement is of general interest. My answers are presented in the order used by Sergei Soares.
• When a larger data set is available than that used in earlier studies, it may be worthwhile to use the larger set, if only to verify that the results do not differ significantly. This comparison could be made if the control variables in the two studies being compared are identical. None of the data sets used in earlier studies that are mentioned by Soares include all of the control variables that are in my article. For example, Table 9 in my article indicates that in 2000, the coefficient for “black plus brown” in an ordinary least squares regression for total income is -0.1287, compared to .1589 for city, and -.1308 for disability. Table 11 indicates that the coefficient of city in the explained difference in income in a Oaxaca-Blinder decomposition is -0.1, which is 1.91 percent of the raw difference in predicted income of “blacks plus browns” and whites. The coefficient of disability is 0.0, indicating that there is no difference in how disability impacts the income of “blacks plus browns” and the income of whites.
Regarding Soares’ comment that the 25% of total income that is received from social security is more or less proportional to wages, and that the amount of income that comes from transfers and the badly measured income from capital is small, so that the difference between salaries and total income is not very relevant for our understanding of discrimination: this is a statement that I disagree with, because the more precise our knowledge of total income, the better our understanding of the impact of discrimination on total income.
• The excellent articles listed by Sergei Soares in his second point either have different dependent variables or different control variables from those I used. This makes it technically inaccurate to say that nothing new was learned. Part of the difficulty is that Soares focuses on black/white differences, as do the articles he mentions, while my article first focused on black/brown and then presented black/white differences. A “typical” reader might not be aware of the changes in black/brown differences because the IBGE website summary tables on the impact of cor/raça present black/white differences. IBGE directs users of its website who wish to learn the black/brown differences to go to the IPUMS website, where they would first have to be authorized to use the site, and then download the censuses, read them into a statistical program, and then use the coding that would enable tables to be constructed, and statistical tests to be performed. Soares later in his comments writes “”sei quão dificil é trabalhar com o Censo”… thus, it is possible that many readers of my article would not know that black/brown differences were larger and significant in earlier times.
Soares states that the following 5 articles use the census. This statement is accurate for 4 of the articles. The articles, the dependent variables, and the control variables are as follows:
Lovell, P. (1994), “Race, Gender and Development in Brazil”, Latin American Research Review, Vol. 29, No. 3, pp. 1-35.
Lovell examines the “differential gains made by white and Afro-Brazilian women and men in the urban workforce. Using sample data from the 1960 and 1980 demographic censuses (the most recent ones available), I have estimated the magnitude of racial differences by gender in place of residence, education, occupational distribution, and wages…. The sample is further restricted to the Northern and Southern regions of Brazil.” Her analyses of racial and gender discrimination are “restricted to women and men between the ages of 18 and 29 who worked 40 hours a week or more. Only women with no children were included … equations were estimated only for 1980”. The wage regression estimated includes race, sex, experience, experience squared, years of schooling (dummies for groups of years), region (North-Northeast), migrant status, and marital status. Lovell uses the Oaxaca-Blinder composition analysis.
Lovell, P. (2000), “Race, Gender and Regional Labor Market Inequalities in Brazil”, Review of Social Economy, Vo. 58, No. 3, pp. 277-293.
“Abstract. This study investigates the relationship between unequal regional development and racial and gender wage inequality in Brazil. Using sample data from the 1991 Brazilian census, I estimated monthly wages for a white, brown and black women and men working in the states of São Paulo and Bahia. The findings suggest that while women and Afro-Brazilians in Brazil’s most developed region of São Paulo had the advantages of higher levels of state sponsored work benefits and more equitable occupational and wage distribution, they nevertheless experienced the greatest discrimination. In contrast, the less developed state of Bahia where racial and gender gaps in education, occupation and wages were the most severe, wage discrimination was lowest.”
Lovell uses three color categories (white, pardo, and preto) to analyze single years, but “whenever data are compared at two points in time I combine pardos and pretos into a single Afro-Brazilian category.” Urban São Paulo and Bahia are included in her study.
Lovell, P., and C.H. Wood, (1998), “Skin Color, racial identity and life chances in Brazil”, Latin American Perspectives, Issue 100, Vol. 56, No. 3, pp. 90-109.
This article uses demographic projections to estimate how many Brazilians changed their declaration of color category between 1950 and 1980 for people 30 years of age and older.
These figures indicate that approximately 38 percent of men and 39 percent of women who declared themselves black no longer did so in 1980. There was a 9 percent decrease of men and 6 percent decrease of women in the white category. Lovell therefore collapses blacks and browns into a single Afro-Brazilian category, although some of those in this category previously were categorized as whites.
Telles (2002) reported that regarding the relationship between the color reported by interviewers and by those interviewed in a 1995 national survey, “Overall, classification as white, brown or black is consistent 79 per cent of the time. However, persons at the light end of the colour continuum tend to be consistently classified, whereas ambiguity is greater for those at the darker end. Based on statistical estimation, the findings also reveal that consistency varies from 20 to 100 per cent depending on one’s education, age, sex and local racial composition. Inconsistencies are in the direction of both “whitening” and “darkening”, depending on whether the reference is interviewer or respondent. For example, interviewers “whitened” the classification of higher educated persons who self-identified as brown, especially in mostly non-white regions.”1 Thus, both the interviewer’s and the respondent’s characteristics influence the color designation of the person interviewed.2
The use of terms denoting color depends on the social context in which the designation of color takes place. If this is understood, then the use of multiple categories at various periods of time rather than a black/white dichotomy is reasonable.
Future studies may estimate differences between determinants of income, wages, and other variables according to whether a person has reclassified him/herself according to color, and how.
Since the terms to denote color in the census are not those most frequently used in popular discourse, it would be interesting to see if the next census provides a larger number of terms, or alternate terms, to denote skin color, and compare the results of analysis depending on which of these terms is used.
Lovell, Peggy A., (2006). “Race, Gender, and Work in São Paulo, Brazil, 1960-2000”, Latin American Research Review, vol. 41. No. 3.
Abstract: This study relies on Brazilian census data from 1960–2000 to analyze long-term trends in racial and gender wage disparities in the urban labor market of São Paulo, one of Latin America’s most dynamic economies. Afro-Brazilians and women have made remarkable progress over the past four decades in securing hard-won legal rights and in gaining access to the highest levels of schooling, entrance into higher paying occupations, and narrowing the intraethnic gender wage gap. Despite such progress, Afro-Brazilians and women are paid less than similarly qualified white men, and wage discrimination is increasing. Placing the interplay of race and gender at the center of this analysis shows how the work- place barriers people confront on the basis of skin color and sex play a fundamental role in shaping social and economic inequality in contemporary Brazil.
Lovell notes that “My sample is limited to wage earning employees aged 18–64 in São Paulo’s urban labor market… Because my sample omits workers who are self-employed or unemployed, my conclusions do not explain racial or gender inequality in access to jobs. My results instead pertain only to the factors that affect wages among current employees.”
Silva, N.D.V., A.L. Kassouf, (2000). “Mercados de Tablho Formal e Informal: umas analise da discriminação e da segmentação”, Nova Economia, Belo Horizonte, Vol. 10, No.1
This article uses data from PNAD 1995, not the census.
• Soares’ states that “Não acho que valha apena fazer mais decomposição OB de diferenciais raciais de rendimento”.
The articles Soares praises as contributing to the literature (Arcand e D’Hombres; the M.A. thesis of Anna Crespo, and the article of Mauricio Reis e Anna Crespo) are useful and contribute to our knowledge of cor/raça. They decompose the differences in black and white income for subgroups of the population. The point is that many of the new contributions that Soares’ finds of interest have been and I hope will be made for subgroups of the population, whether by quantiles, region, occupation, education or other characteristic. These contributions would make excellent monographs, and be especially useful if descriptions and analyses of the industries, regions, etc., were to be included, as well as the econometric estimates, so that we understand the facts and historic processes needed for an adequate explanation of the econometric results. Even if we have an adequate general knowledge of cor/raca, the details for the subgroups have been and will be of continuing interest.
• Regarding the use of the Oaxaca-Blinder technique in the future: Surely this use is warranted for analysis of data that will be provided in the next census, if only to obtain results that could be compared to earlier studies. However, the form of Oaxaca-Blinder technique used may differ from that used in earlier studies, which were carried out before computers became relatively inexpensive and more powerful, enabling computations to be carried out more rapidly than on earlier computers, and made easier by the increasing number and types of analyses offered in standard statistical packages. The matching comparisons technique used by Ñopo modifies the traditional Oaxaca-Blinder computations. Its application to Brazil is presented in Luana Marques Garcia, Hugo Ñopo and Paola Salardi, “Gender and Racial Wage Gaps in Brazil, 1996-2006: Evidence Using a Matching Comparisons Approach”, IADB June 2009 pubWP-681.pdf . This article uses PNAD data. Its outstanding presentation of methodology, review of the literature, and summary of earlier work may lead researchers to use both the traditional Oaxaca Blinder analyses and the matching comparisons technique, and compare the results.
I thank Sergei Soares for his detailed response, and look forward to the new studies that our discussion may encourage.
Notes
1. Telles, Edward E. 2002. ‘Racial ambiguity among the Brazilian population’, Ethnic and Racial Studies, 25:3, 415 – 441. To link to this article: DOI: 10.1080/01419870252932133
URL: http://dx.doi.org/10.1080/01419870252932133 . Also see Bailey, Stanley r. and Edward E. Telles,(2006). “Multiracial versus collective black categories Examining census classification debates in Brazil”, Ethnicities, Vol. 6(1):74–101
2. Telles, Edward E. 2004. Race in Another America: The Significance of Skin Color in Brazil. Princeton University Press, pages 88-91.
Tres pontos sobre a resposta da Laura Randall aos meus comentários.
De fato, a totalidade dos estudos usando Oaxaca-Blinder que conheço decompoem salários e não renda total. No entanto, isso não é muito relevante. Aproximadamente 75% da renda medida em Pnad ou Censo é renda do trabalho. A maior parte dos outros 25% é renda da seguridade social que é mais ou menos proporcional à renda do trabalho quando na ativa. Ou seja, a diferença entre renda do trabalho e renda total se encontra principalmente nos 1-2% que são programas de transferencia de renda não contributivos como o Bolsa Família e também na muito mal medida renda do capital.
Segundo ponto. Também é verdade que a maior parte dos estudos comparando rendimentos do capital por cor usa a Pnad, mas cinco estudos usam o Censo, e o de Peggy Lovell de 2006 usa o Censo de 2000.
Lovell, P., (1994), “Race, Gender and Development in Brazil”, Latin American Research Review, Vol. 29, No. 3, pp. 1-35.
Lovell, P., (2000), “Race, Gender and Regional Labour Market Inequalities in Brazil”, Review of Social Economy, Vol. 58, No. 3, pp. 277-293.
Lovell, P., and C.H. Wood, (1998), “Skin colour, racial identity and life chances in Brazil”, Latin American Perspectives, Issue 100, Vol. 25, No. 3, pp. 90-109.
Lovell, Peggy A., (2006), “Race, Gender, and Work in São Paolo, Brazil, 1960-2000”, Latin American Research Review, Vol. 41, No. 3..
Silva, N. D. V., A. L. Kassouf, (2000), “Mercados de Trabalho Formal e Informal: uma analise da discriminação e da segmentação”, Nova Economia, Belo Horizonte, Vol. 10, No. 1.
Meu terceiro ponto é mais geral. Não acho que valha a pena fazer mais decomposição OB de diferenciais raciais de rendimento. Contei e há, pelo menos, 16 estudos que usam Oaxaca-Blinder para decompor rendimentos de acodo com cor/raça. Arcand e D’Hombres (2004 – “Racial Discrimination in the Brazilian Labour Market: Wage, Employment and Segregation Effects”, Journal of International Development, Vol. 16, pp. 1053-1066.) usam quantis para decompor as diferenças. Anna Crespo na sua tese de mestrado na PUC-RJ usa OB por UF em um trabalho tão detalhado que tem dezenas de páginas so de tabelas. Meu ponto é que esta literatura entrou ha muito tempo em rendimentos decrescentes e hoje os redimentos são quase nulos.
Para não dizer que são completamente nulos, posso citar o trabalho de Maurício Reis e Anna Crespo (2005 – “Race Discrimination in Brazil: An Analysis of the Age, Period and Cohort Effects”, IPEA, Rio de Janeiro, Texto para Discussao, No.1114.) que mostra algo de substantivamente novo – que há uma redução na discriminação para as coortes mais recentes.
Quero enfatizar que o trabalho da Laura Randall é empiricamente bem feito e bem escrito – sei quão dificil é trabalhar com o Censo, mas não há nada de novo a ser aprendido fazendo mais OB com diferenciais raciais de rendimento.
Minha opinião é que hoje os pesquisadores no tema de relações raciais no Brasil tem que passar a outras perguntas como qual é a relação entre cor e classe no Brasil e quais são os mecanismos de acensão social de negros no Brasil. OB foi importante no seu dia, mas não tem mais nada a contribuir.
Sergei Soares did not notice that the regressions in my article are for total income, not for salary, and that Census data were used. The Censuses provide a larger data set than do the various PNADs. The PNADs provided the data for many of the “10-20 trabalhos nessa linha”. I believe that these features and the detailed regression and Oaxaca decomposition results provide new information.
O trabalho, embora não mal-feito, não tem nada de novo. Regressões de salário com dummy de raça temos desde o trabalho do Nelson Valle Silva e Carlos Hasenbalg nos anos 70. Oaxaca tem desde o meu em 2000, sendo que desde lá tivemos uns 10-20 trabalhos nessa linha com controle de Heckman, regressão quantílica, e assim por diante. O trabalho não está ruim, mas não tem nada que não tenha sido publicado há pelo menos uma década.