Racial inequality and the Brazilian labor market

MISSING IMAGE

Material Information

Title:
Racial inequality and the Brazilian labor market
Physical Description:
vii, 183 leaves : ill. ; 28 cm.
Language:
English
Creator:
Lovell, Peggy A., 1958-
Publication Date:

Subjects

Subjects / Keywords:
Race discrimination -- Brazil   ( lcsh )
Discrimination in employment -- Brazil   ( lcsh )
Blacks -- Employment -- Brazil   ( lcsh )
Genre:
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Thesis:
Thesis (Ph. D.)--University of Florida, 1989.
Bibliography:
Includes bibliographical references (leaves 170-182).
Statement of Responsibility:
by Peggy A. Lovell.
General Note:
Typescript.
General Note:
Vita.

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001500393
notis - AHB3121
oclc - 21381671
System ID:
AA00003770:00001

Full Text













RACIAL INEALITY AND TE
BRAZIIJAN IABOR MARKET












By

PEGGY A. IDVEIL


A DISSERTATION PHES O THE G ADU ATE SCHOOL OF THE
UNIVERSITY OF FIDRIDA IN PARTIAL FULFILLMENT OF THE
REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA


1989













AC24OWIEDGEMENTS


That I ended up in graduate school at all is due to an

intriguing article written by Charles Wood that I read as an

undergraduate. His ideas on inequality and on the reciprocal

relationship between the process of social change and

development influenced my approach to the study of racial

inequality in Brazil. For this, for his generous support, and most

of all for his friendship, I would like to thank him.

In the course of graduate school I have benefitted from the

opinions and support of many friends and professors. The guidance

offered by members of my oumnittee, Joseph Vandiver, John

Henretta, Leonard Beeghley, and Maxine Margolis, has inspired me

to pursue this research topic beyond graduate school.

I am grateful to my fellow students in the Department of

Sociology and the Center for Latin American Studies who

contributed more to this effort than they realize. Special thanks

are due to Jeffrey Dwyer for his constant support and friendship.

This research would not have been possible without him. The

logistics of this endeavor would have been much more difficult

without the technical expertise provided by Fred Burch and Debi

Van Ausdale. I would also like to thank Graham Webster for his

support.

Finally, I owe my deepest gratitude to my family. Martha

Lovell, my mother, and Raye Bryant, my grandmother, have been a
ii













source of strength and encouragement throughout this process. My

son Graham helped me keep things in perspective by talking about

his inventions and playing football with me. For never once

czxplaining about frozen dinners, and for loving me for reasons

that had nothing whatsoever to do with this work, I dedicate this

work to him.














TABLE OF OCMtENIS


ACKNOWLEDGEMENTS. . .

A E CT. . . .

CHAPTERS

1 INnIrXCTION. ...............

Overview of Race Relations in Brazil
Slavery, Abolition, and Regional
Development. . .
Color in Brazil . .
New Directions. ...
Research Design. . .
Notes. .. .. .. .. .


.ii
.ii


S. .vi






. 2

. 4

S. .12
. .15
S. .17


2 THDRIES OF MAGE DISCRIMINATION . 18

Neoclassical Perspectives on Discrimination. .19
Econcnic Sectors and labor Markets .21
Segnted abor Markets . 22
Split Labor Markets . ... 25
Informal/Formal Sectors . .26
Human Capital. .......... ...... .27
Methodological Caveats . ... 29
Measuring Discrimination . .31
Notes. .......................34

3 DATA AND VARBIES. ... .. .... 35


The Validity of the Census Question on
Sample Restrictions. . .
Variables Used in the Analysis ..
Summary and Discussion .
Notes. .................


Race.
. .
. .
. .
* *


4 MEASURING WGE INEQUAITTY .. .
The oaplete Earnings Function .
The Interaction Model. . .
Human Capital and Background Characteristic
Correlations with Earnings. .
The Earnings Function by Color .
A Summary Measure of Discrimination. .
Summary and Discussion . .
Notes. ...................


.36
.38
.39
.49
.50

.51
.52
.57

.59
.61
.67
.73
.74













5 ENINGS AND REGIONAL INEQULI. ...


Regional Ocqparisons . .76
Reformulation of the Earnings Function. .. 77
The Earnings Function by Region 78
The Earnings Function by Race and Region. .83
Decumposition of the Wage-Gap by Region .93
Summary and Discussion . 97
Notes. .... ............. ...99

6 INDUSI IAL SECIR EFFECT CN EARNINGS. 100

Industrial Sectors . 101
Sectoral Analysis of Wage Discrimination 103
Industrial Labor Market Characteristics 105
Racial Distribution by Sector . 106
he Earnings Function by Industrial Sector. 107
The Earnings Function by Race and
Industrial Sector. .. . 111
Decxuposition of the Wage-Gap . 120
Suwiary and Discussion . 122
Notes ................ .. ... 124

7 INMRA-00CPATIONAL ANALYSIS ....... 125


Occupational Discrimination. ...
Wage Discrimination: Intra-Ocupational
Analysis. . ..
The Earnings Function by Occupation ..
The Earnings Function by Race and
Occupation . .
Decznposition of the Wage-Gap ...
Summary and Discussion . .


. 125

. 129
. 134


139
143
144


8 SUMMARY AND DISCUSSION. . 146

APPENDICES

A 1980 BRAZILTAN CENSUS OCXUPATINAL CATEGORIES 166

B 1980 RAZILIAN CENSUS INDUSTRIAL CATEGORIES 167

REFERENCES .................. ....... 170

BIOGRAPIHCAL SKETi ..... ...... 183


. .75















Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy


RACIAL INEQUALITY AND E BRAZILTAN IABOR MAKE

By

Peggy A. Lovell

August 1989


Chair: Charles H. Wood
Major Department: Sociology

The widely held assuqtion that racial discrimination does

not exist in Brazil has been questioned by recent studies of

social mobility, the determinants of urban wages, and child

mortality. This analysis extends the scope of previous research by

using average marthly earnings as the dependent variable in the

study of racial inequality among male workers in metropolitan

Brazil. Estimates of earnings from the 1980 demographic census

show that the average wage of the nonwhite population was nearly

half that of the white population. Using regression

standardization to calculate the proportion of the earnings

differential due to labor market discrimination, the analyses

showed that, for blacks and mulattoes respectively, 25 and 32

percent of the earnings gap could be attributed to discriminatory

labor market practices. Disaggregated analyses of labor markets

further demonstrated that nonwhites receive a different treatment









in the labor market, that there are crucial differences between

blacks and mulattoes, and that the degree of wage discrimination

varies by region, industrial sector, and occupational position.


vii















CHAPTER 1
INIRODJCTION



Gilberto Freyre, one of Brazil's most eminent sociologists, once

wrote that race relations in Brazil are "probably the nearest approach

to paradise to be found anywhere in the world" (1959:9). Freyre's

comment echoed a omm-n and still popular theme, that in Brazil there

is no discrimination on the basis of skin color. The racial democracy

thesis implies not that blacks and whites are equal in social standing,

but rather that individuals have equal opportunityy for social

advancement regardless of skin color. Moreover, once Afro-Brazilians

overcame the legacy of slavery by acquiring sufficient levels of

education and income, the thesis predicted that nonwhites would find no

barriers to social or economic mobility. Yet, one hundred years after

the abolition of slavery, Brazil remains a country of marked racial

inequality.

In 1960, the average income of the nonwhite population in Brazil

was nearly half that of the white population (Silva 1978). Estimates

for 1980 indicate that whites continue to earn more than nonwhites

even after controlling for differences in human capital investments.

In the Brazilian sociological literature, wage differentials between

whites and blacks have conventionally been interpreted as the result

of an incomplete process of social mobility. Hence, differences in

socioeconomic achievement, rather than discrimination, account for












racial income differentials. In contrast to the mobility argument, a

new wave of research on racial inequality in Brazil attributes wage

differences by race to labor market discrimination rather than unequal

levels of education or skills (Silva 1978, 1985; Hasenbalg 1985;

Hasenbalg and Silva 1987; lovell Webster and Dwyer 1988a, 1988b).

The purpose of this dissertation is to determine the extent to

which wage inequality in Brazil is attributable to labor market

discrimination as aroed to unequal individual and demographic

attributes. Using sample data on metropolitan areas from the 1980

census of Brazil, the objectives are to (1) generate a profile of

sociioecoa nic differences between the white, mulatto, and black male

workforce; (2) identify the market mehanisms that lead to racial wage

differences by region, industry, and occupation; and (3) estimate the

proportion of the wage gap between whites and nonwhites that is due to

discrimination in the labor market, as opposed to effects associated

with different levels of human capital endowmnts.


Overview of Race Relations in Brazil

Contemporary Brazil has the world's largest population of African

descent outside of Africa. More than 53 million, or 45 percent, of

Brazil's 120 million people are black or of mixed race (Wood and

Carvalho 1988). The first black slaves arrived in 1538, initiating a

300 year period during which there was a steady flow of Africans to

Brazil. Over the centuries, Brazil received roughly 38 percent of all

the slaves taken to the New World (Rocha 1988). Slavery transformed

Brazil fran a land inhabited by a few million Indians and a tiny number












of Portuguese colonists into one of the world's most heteragenous

populations.

Brazil oc~pies a special place in the study of race relations in

the Americas. Unlike the United States, segregation of the races after

the abolition of slavery was neither legally mandated nor socially

practiced. The passage from slave-holding society to a free one

occurred without the racial violence, such as the riots and lynchings

that marked the Unites States experience. Intermarriage, if not the

rule, was certainly acann. Thus, in comparative terms, it appeared

that, on the surface at least, Brazil was justifiably famous for its

reputation as a multi-racial society free of racist sentiment.

It was through the work of the Brazilian social scientist Gilberto

Freyre that the idea of harmonious race relations gained wider

currency. Freyre attributed the absence of racial prejudice and

discrimination, and, hence, the existence of equal opportunities for

whites and rnnwhites, to the unique pattern of race relations which

developed in Brazil. Unlike the North American continent, which

attracted colonists who came to the New World as family units,

colonists in Brazil were single males. As a result of the relative

absence of white women, unions were camon between Portuguese and

either black slave or Indian women. Intermarriage, as Freyre's

reasoning suggests, was the cornerstone of "racial democracy."

The consequence of the way in which the Portuguese colony was

populated has been the whitening, or branueamento, of Brazil's

population. At ane time, branqueamento was also a political doctrine.

Born of the belief in the racial inferiority of the black population in











the late nineteenth and early twentieth centuries, the whitening ideal

found official expression in immigration laws designed to stimulate

European migration to Brazil (Ferreira da Silva 1989). Brazilian

intellectuals and politicians endorsed the idea that drowning black

blood in imported ropean white blood would make the population

whiter and therefore "better" (Skidmore 1974).


Slavery. Abolition. and Recional Develoment

Conte-porary patterns of racial inequality in Brazil are the

legacy of Portuguese expansion in the search for markets, raw

materials, land, and labor. Portuguese colonists brought over

approximately 3.6 million Africans during the three and a half

centuries of slave trade in Brazil.1 The growth and allocation of

slave labor in Brazil was intimately associated with spatial features

of economic development. The demand for labor on the sugar plantations

in the northeast was the initial impetus for the slave trade. Black

slaves were concentrated in the sugar producing area, Zon da Mata,

along the northeastern coast until the profitability of sugar declined.

Competition from a growing supply of West Indian sugar led to a fall in

the world price, and the once prosperous northeast was plunged into

recession. A lasting social cosequence of this heritage is that, to

this day, the Northeast remains the most populous and poverty stricken

area of Brazil.

A series of boon and bust cycles in the Brazilian economy were

associated with the forcible internal migration of slave labor (Silva

1985). As sugar declined in economic importance, black slaves were












partially shifted to the center-south regions of the country when gold

and diamrnd mines were discovered in Minas Gerais. The growing mining

industry spurred an increased slave trade in the second half of the

seventeenth century, following the period of relative stagnation due to

the crisis in the sugar industry. In the nineteenth century, gold

mining sWsequently declined, and the slave population was again forced

to migrate. Many slaves were sent back to a recovering sugar industry

in the Northeast. A smaller faction were sent on to the rapidly growing

coffee industry in the Southeast. Of all Brazil's export comeodities,

none had a more lasting impact on the contemporary structure of

regional and racial inequality than did coffee.

The abolition of slavery in Brazil occurred in 1888 during a major

upswing in the coffee cycle. Rather than incorporate the newly freed

slaves into the wage labor market, the coffee growers, with the backing

of the Brazilian government, turned to European and Asian immigrants as

a substitute labor force (Gebara 1986). Eager for a stable labor

supply, the fazendeiros used export proceeds to underwrite the costs of

importing European and Asian immigrants to the southeast, particularly

Sao Paulo (Merrick and Graham 1979).

That there was a clear preference for the lighter-skinned European

and Asian worker over the dark-skinned freed slave cannot be denied

(Andrews 1988). It has been argued that the preference for immigrant

labor was not because they possessed better skills since nearly all

migrant workers learned their skills on the job (Hall 1969). Rather,

blacks were excluded from wage work because immigrants were considered

to possess the proper work ethic and were thought to be more docile and












amenable compared to the runanagable and potentially volatile ex-

slaves (Ferandes 1965). As a result, the immigrant posed serious

competition to the newly freed slave, pushing the latter into a

marginal position in society at precisely the time when the key sectors

of the Brazilian eoanmy were undergoing the transition to a capitalist

mode of production based on wage labor (Silva 1978:45).

European and Asian immigrants in the Southeast played ai crucial

entrepreneurial role in the growing urban-industrial eoanmy of Sao

Paulo. The immense profits generated from coffee exports promoted the

emergence of banking, manufacturing and light industry in the south,

creating a growth pole that attracted additional investments and

numerous opportunities for social advancement (Merrick and Graham

1979). Over time, the consequence has been sharp geographic disparities

that coincide with the unequal racial distribution of the Brazilian

population.

The two most striking regional contrasts are the poor and

underdeveloped Northeast, where a disproportionate rnmber of blacks and

mulattoes live; and the industrialized and developed Center-South,

where the greatest part of the white population is concentrated. Since

the days of slavery, the educational and economic opportunities have

been much scarcer in the North than in the South. Although there have

been recent governent-spnsored attempts to industrialize the

Northeast, investments have been limited to petrochemical industries in

Salvador, Bahia, with little benefit accruing to the rest of the

Northeast (Faria 1982; Guimaraes 1987). Meanwhile, the South of Brazil












has maintained impressive rates of ecrxnic growth and remains the

locus of manufacturing and mechanized agriculture.

The fact that most blacks and mulattoes reside in the Northeast

has important implications for this study. Among other things, it means

that much of the disparity between white and necnhites observed in the

aggregate is associated with geographic differences in the level of

regional development. Analyses of wage discrimination must therefore

control for place of residence in Brazil.


Color in Brazil

As a result of the widespread miscegenation which has occurred

over the centuries, the racial Iocpositicn of the Brazilian populace is

complex. Rather than a simple black/white dichotomy as in the United

States, in Brazil a multicategory system identifies shades of skin

color, physical characteristics and social standing (Harris 1964;

Azevedo 1953). For example, the word brn (white) can refer to a

predominantly white person no matter what his or her social standing,

or brano may refer to somene of mixed heritage with a high social

status. Neo (black), on the other hand, may mean a person of

noticeably Negroid features, or any individual of low social rank.

Sometimes the term Dqo is also applied to close friends as a sign of

intimacy and endearment. In the United States, by contrast, the

black/white color line has been historically defined and rigidly

enforced. Anyone with known African ancestry is classified as "black"

regardless of their appearance.












In the post-abolitionist period, Brazil and the United States

developed different systems of racial classification. People in the

United States recognize only two racial groups (whites and blacks) and

classify people into one or the other according to the rule of hypo-

descent (Harris 1964). Accordingly, individuals known to have any black

ancestry are classified as black. In Brazil, on the other hand, the

classification rules are not as rigid. A key difference in the

Brazilian case is the socially recognized intermediate category,

mulatto.

Degler in his classic book Neither Black nor White, argued that

the mulatto occupies an intermediate position between the polar

extremes. As such, nulattoes have wider possibilities for social

mobility than blacks. Degler referred to this phenomenon as the

"mulatto escape hatch." As far as the pattern of race relations was

concerned, Degler argued, the "mulatto escape hatch" did not prevent

racial discrimination in Brazil, so much as blur, and thereby soften,

the line between black and white. As a consequence, racial

discrimination in Brazil was often more subtle compared, say, to the

United States (Degler 1971:225).

One of the dominant themes in the Brazilian literature on race

relations is the idea that racial inequalities are not the result of

discrimination so ruch as unequal socioeconxnic achievement. A version

of the "class over race" argument, known as the "Bahian school" of

thought, was put forth by social scientists Donald Pierson (1942) and

Thales de Azevedo (1953). Consistent with the racial democracy thesis,

both analysts agreed that factors such as education, occupation, and












wealth were more important than race in the determination of social

rank and mobility. Pierson and Azevedo claimed that even though racial

prejudice existed in Brazil, derogatory stereotypes against blacks did

not translate into discriminatory behavior. Hence, an independence

between prejudice and discrimination was assumed. Indeed, the two

concepts were necessarily independt inasmuh as the racial democracy

thesis contended that race was of limited ixpartance as a determinant

of social mobility. Harris (1964) holds essentially the same

perspective. He notes that racial prejudice ard stereotypes do exist in

contemporary Brazil, but he believes these to be of relatively little

consequence. The real factor, aoording to Harris (1964:63-64), is that

of class:

It is one's class and not one's race which
determines the adoption of subordinate and
superodinate attitudes between specific
individuals in face to face relations...There are
no racial groups against which discrimination
ocurs. M(he) issue of racial discrimination is
scarcely a vital one. ower-class whites and lower-
class colored people are alike segregated and
discriminated against, one perhaps slightly more
than the other. .

Another school of thought, which appeared in Sao Paulo in the

1950s and 1960s, focused on the transition from an agrarian slave

society to industrial capitalism in the Southeast of Brazil.

Researchers in this tradition, chiefly Florestan Fernandes (1969,

1972), confirmed the presence of racial discrimination in contemporary

Brazil. Yet, Fernandes emphasized the legacy of slavery, and the

different starting point of whites and blacks at the amment of

abolition, as explanations for the current social inequality between

the races. As such, Fernandes' view implied that racial inequality was












an aberration destined to decline with economic development (Hasenbalg,

1985).

Taking a somewhat different position, others scholars in the Sao

Paulo school, such as Fernando Henrique Cardoso (1962) and Octavio

lanni (1962), stressed the adaptation of racism to the structural

characteristics of industrial capitalism. According to this idea,

recently expanded by Carlos Haserbalg, racial antagonism, which had its

origins in slavery, did not wane with the rise of industrial capitalism

in Brazil but assumed a new role and meaning. "Race prejudice and

discrimination" in Haserbalg's view, ". .far from being a simple

survival from the past, are functionally related to the material and

symbolic benefits obtained by whites through the disqualification of

nonwhites as ampetitors" (1985:27).

While Brazil's favorable reputation in the areas of race relations

was initially bolstered by oonparison with other systems, particularly

the United States, a new generation of scholars has reevaluated the

place that race and color play in Brazil's stratification system.

Hasenbalg (1985) has shown that nonwhites in Brazil are exposed to a

cycle of cumulative disadvantages in intergenerational social mobility.

Using data from the 1976 national household survey, Hasenbalg

demonstrated that, other things being equal, the probability of

improving social status from one generation to the next was

considerably smaller for nonwhites than for whites. Moreover,

interracial differences in the opportunities for upward mobility

increased with higher social status of origin. Among the small groups
of nonwhites born in families of high social standing, the risk of












social demotion was =xch greater oozpared to whites of the same

background. Hasenbalg (1985:32) concluded that racial inequalities that

began under slavery were perpetuated "through discriminatory practices

and cultural stereotyping by whites of the role 'adequate' for blacks

and mulattoes."

Other forms of discrimination take place in the labor market.

Using the 1976 nation al household survey, Nelson do Valle Silva (1985)

found that nar whites received lower wages compared to whites with the

same amount of education and job experience. In 1976, average inkcme

for whites was twice that of naxuhites. A third of that difference

could be attributed to discriminatory labor practices. Silva concluded

that the monetary disadvantage suffered by nonhites due to

discrimination in the labor market was around 60 dollars a month (566

Cruzeiros). These findings were consistent with the results of his

earlier study (Silva 1978) of black-white income differentials in

Brazil in 1960, as well as more recent analyses by ovell Webster and

Dwyer (1987, 1988), based on a sample of Brazil's 1980 demographic

census.

Recent studies of racial inequality go beyond the concern for

social mobility and the determinants of urban wages. Wood and Lovell

Webster (1988), for example, used child mortality rates as the

dependent variable in the study of racial inequality in metropolitan

Brazil. Estimates of life expectancy at birth, derived from demographic

censuses, showed that children born to white women outlived those born

to nonwhite women by 7.5 years in 1950. Despite absolute improvements

among both groups over the next thirty years, the mortality gap between












whites and nonwhites remained about the same (6.7 years) in 1980.

Calculating a mortality ratio for individual wamen, regression analyses

of sample data from 1980 census showed that racial differences in child

mortality persisted after controlling for individual and demographic

characteristics. Wood and lovell Webster concluded that an interaction

between race and membership in terms of the social security system

suggested that discriination against ncmhites in the provision of

health care services contributed to higher mortality of nonwhite

children.


New Directions

It is hardly surprising that the racial democracy theme was

endorsed by privileged social and economic groups in Brazil. This was

especially evident during the period of military rule, from 1964 to

1985. Frtm the outset of the revolution, the military leaders who took

control of the civilian political apparatus repeatedly justified their

actions in terms of national security and national unity. The two

themes had important consequences for the role that race and race

relations played in academic research and in political debate. The

appeal to the unity and security of the nation implied, among other

things, that any recognition of racial inequality or interracial

tension was considered divisive, even subversive. The effect was to

forcibly repress any mention of racial issues in public debate, and to

censor the publication of "treasonous" research findings on racial

inequality.












In keeping with this political climate, dgmgraphers at the census

bureau came under pressure from military authorities to delete the

racial identification question from the 1970 generation. A related

example of government repression was the purge of the faculty at the

University of Sao Paulo. Prominent social scientists involved in the

study of racial inequality were forced by the military government to

"retire." Florestan Fernandes, Fernando Henrique Cardoso and Octavio

lanni were cassados for ten years under the dictatorship. As a result

of the political climate and the resulting lack of data, it has been

practically impossible to monitor the evolution of the relative

position of blacks in Brazilian society over the past few decades

(Hasenbalg and Silva 1987).

Since the military abrogated power in 1985, the issue of racial

inequality has emerged as an important rallying point under the new

democratic regime. oughly 600 Afro-Brazilian organizations have sprung

up in the more liberal political climate of the last several years. For

the first time in Brazilian history, black activists, with a variety of

foci, both political and cultural, have begun to challenge the popular

ideology of racial equality.

One of the most powerful interest groups is the recently

established Unified Black Movemnt (Movimento Negro Unificado, MNU).

The MNU urges black workers to organize independently within unions,

neighborhood groups and samba clubs to protest discrimination and to

campaign against police violence (MNU 1988). In 1988, which was the

Centennial of the Abolition of Slavery, the MNU organized hundreds of

anti-racist demonstrations across the country. The depth of Afro-












Brazilian disquiet was reflected on May 13, 1988, the day that marked

the signing of the royal degree by Princess Isabel abolishing slavery.

Rather than celebrate "one hundred years of lies," many black

organizations called for recognition of November 20 as Afro-Brazilian

consciousness day. The date canm rates the anniversary of the death

of Zumbi, a slave leader captured and killed in 1695 at the fall of

Palmares, a free territory in the Northeast, where thousands of escaped

slaves (quilc~hbo) resisted slavery for nearly one hundred years.

Across Brazil, the 1988 centennial unleashed a round of lively

debates on racial inequality. Exhibits, articles, and new books

retelling the story of slavery in Brazil were released in an

unprecedented frenzy of activity. For the first time, the Brazilian

press broke the old taboo against examining race relations and for the

first time in Brazilian history, it was reported publicly that blacks

eat less, get less education, earn less, and die earlier than whites

(Simnns 1988). Not only was national attention focused on the

centennial, but it was covered by the international press as well.

However promising this abertura or opening appears, recent events

indicate that public debate on racial inequality still elicits

government repression. For example, on the eve of the centenary of

abolition, a demonstration by black groups in Rio de Janeiro was

stopped by a massive police and army presence. The following day, a

nationally televised debate on racial issues was taken off the air at

the last moment without explanation (Rocha 1988). These are only two of

numerous examples which suggest that even though there has been












considerable public denouncment of racial inequality, the ucope of the

debate rains bounded by the persistent myth of racial pquaity.

As recent events suggest, the need to analyze racial equality in

cortenporary Brazil is especially critical in light of the changing

political climate. 7his topic is important for other reasons as well.

Fron the standpoint of basic information, there exists limited data on

the magnitude of race differences in the labor market, and of how these

differences vary by region, industry, and occupation. A straightforward

ooaparison of such findings is therefore an essential step in an

assesment of racial groups in the country.


Research Dsie n

This dissertation focuses on the process of labor market

discrimination among white, black and mulatto men employed in

metropolitan Brazil. The conceptualization of the labor market most

useful to the analysis of inequality is one that allows for

differentiation of labor markets along regional, industrial, and

occupational criteria.2 Chapter 2 presents a review of the sociological

and economic literature on earnings differentials and racial

discrimination. Particular attention is paid to the Human Capital

theory of wage determination, and to the literature on segmented

economies and labor markets.

Chapter 3 uses sample data from the 1980 census to identify racial

differences in wages, job experience, education, geographic

distribution, motion, nation, hours worked, access to social security, and

migrant and marital status. In addition to generating a socioeoncanic











profile of whites, blacks, and mulattoes in Brazil, these variables

provide the basis for analyses of wage discrimination in Chapter 4. The

hypotheses regarding wage discrimination developed in Chapter 4 are

premised on the assiaption that the degree of wage discrimination

varies in different labor market settings. The latter can be defined in

three different ways.

The first criterion far disagregatin is by geographic region of

the country, as shown in Chapter 5. Given the regional disparities in

ecnrxmic development, we can expect to find that wages are lower for

all workers in the Northeast compared to those in the Center-South.

Once these regional disparities are accounted for, the central question

is whether wage discrimination is positively or negatively correlated

with the level of regional development and other variables, such as the

percent of white employers in the nine metropolitan regions.

Wage discrimination is also likely to vary by industrial sector.

The approach, used in Chapter 6, disaggregates the Brazilian economy

into the transformative sector (further subdivided into modern and

traditional industries), and the service sector (further subdivided

into producer, social, distributive, and personal services). Each of

these distinctions are associated with different ways that labor is

hired, used, and reunerated. The broad distinction between the

transformative and the service sector, for example, captures

fundamental differences in size of firm and in the labor process

itself: Workers in the modern transformative sector produce a tangible

good within large-scale firms in which the labor process is highly

capital intensive; workers in the service sector produce intangible











services of various kinds, depending on whether they are engaged in

producer, social, distributive, or personal services. The point is

that the process of wage determination and degree of wage

discrimination, between whites, blacks and mulattoes, are likely to

vary specific to each economic subsetting.

The third approach focuses attention on occupational categories

rather than industrial sectors. The analysis in Chapter 7 addresses the

issue in two ways. The first step uses a logistic regression technique

to identify the factors associated with whether an individual is

employed in a white collar or blue collar job. The second step measures

the degree of labor market discrimination between whites, mulattoes,

and blacks within three occupational groups: managerial/professicnal,

clerical, and blue collar.


Notes


1. For a review of slavery in Brazil see Tanneibaum (1947); Elkins
(1959); Degler (1971) ; Moura (1988); Reis (1987, 1988); Mattoso (1988);
Gebara (1986); and Schwarcz (1987).


2. In this analysis the term labor markets refers to Althauser and
Kalleberg's definition of labor markets as "arenas in which one or more
of the following are similarly structured: employment, movement
between jobs, development and differentiation of job skills, or wages
(1981:121)." This definition suggests that within different regions,
industries, and occupations, there is similarity in the wage attainment
process due to structural mechanisms that operate with uniformity
across persons. It is important to recognize that handicaps minority
groups face also embody human capital and background characteristics.
The research strategy developed in Chapter 2 is to draw on individual
and structural characteristics in order to understand variation in
wages by ascriptive status.














CHAPTER 2
THEORIES OF WAGE DISCRIMCINATI


The study of inequality lies at the heart of both classical and

cRntenporary sociology. One form of inequality that has received much

attention is the inequity in the average wage rates amog groups of

workers classified by sex, race, or ethnicity. Analysts have dealt with

the question of socioeox nic inequality froa a variety of different

perspectives. Three views on economic inequality, each situated at a

different level of analytical reasoning, wll be reviewed here: (1)

the neoclassical econaic perspective, with an emphasis on ccpetitive

and market equilibrium, (2) the segmnted/split labor market, with an

emphasis on ecornic sectors (industrial labor markets), and (3) the

human capital approach, with an. emphasis on individual characteristics

that workers bring to the labor market.

This review of the literature is motivated by two major concerns.

First, it is important that the subsequent analysis of individual

workers' earnings in Brazil be undertaken within the context of previous

research concerning determinants of wages. The goal is to theoretically

and empirically integrate traditions featuring discussion of individual

earnings with those concerned with labor market structures. Second,

this analysis of racial differentials is motivated by an interest in

discrimination. As indicated in Chapter 1, a statistical dermposition

technique will be used in order to measure the extent to which wage
18











inequality is attributable to discriminatory labor practices. Hence,

this literature review will identify the standard way in which

discrimination is measured in the analysis of individual wages.

Subsequent chapters will draw on the integration of individual and

structurally oriented perspectives to specify a basic model of wage

determination and to interpret measures of income inequality.

Neoclassical Perspectives on Discrimination

Theories of wage determination, in neoclassical economics, are

based on the asswptions that wages are the basic labor market clearing

mechanism; perfect competition exists; majority and minority workers

have equal productive capacity and tastes for work; and factors of

production are hoogenous and interdangeable (Cain 1976). The basic

idea is that on the basis of their skills and qualifications individuals

freely compete against one another for wages. To neoclassical economists,

wage discrimination exists when individuals with the same economic

characteristics receive different wages because of racial, gender, or

other ascribed characteristics (Stiglitz 1973:287).

One of the main tenets in neoclassical economics is that

discrimination is a market imperfection that is ultimately cleared away

by the tendency toward equilibrium inherent in a oonpetitive marketplace

for labor. Employers, seeking to maximize profits, are interested in

worker productivity and potential contribution to profits, not the

worker's skin color. Thus, the pressure of economic xnipetition among

firms can and will overcome the resistance of those few racist employers

who persist in discriminating. Similarly, purchasers of goods and

services will be interested only in the product's price and quality,











not in the race of the workers who produce it. The eventual outcome is

that, in a competitive marketplace, "the inexorable force of competition

in unfettered labor markets will reduce over time all differences in

Images to real differences in human capital attributes between black and

white workers" (Bostn 1988:95). Similarly, as hurow (1975:161)

observed: "if something is a market imperfection, there are always

profits to be made by eliminating it. If markets are basically

om"pettiive, mn sooer or later discovers a way around the

imperfection." Econ=mist, Milton friedman (1982:108) is one of the

ost vocal proponents of this view. "It is a striking historical

fact," Friedman contends, "that the development of capitalism has been

accompanied by a major reduction in the extent to which particular

religious, racial, or social groups have operated under special handicaps

in respect of their ecmnnic activities; have, as the saying goes, been

discriminated against."

Neoclassical theory and its variants have prompted several

criticisms. One of the most damaging is the contention that it cannot

account for the persistence of wage differences (assuming equal worker

productivity). If discrimination exists, and the assuption is that

discrimination and competitive equilibria are incompatible, it only

takes one nondiscriminating employer for the system to break down. It

would either be possible to drive other competitors out of business or,

alternatively, to stay in business white employers would have to

eliminate their preference for white workers. It follows that, if there

is evidence of any long-lasting market imperfection (i.e.











discrimination), it is highly probable that it plays sane function in

the economy.

One of the best-known explanations for such "market imperfections"

is Becker's Econi ics of Discrimination (1957). Becker provided a

quantitative representation of a an employer's "taste for discrimination"

The concept provided one of the first measures for quantifying the cost

of discrimination. This theory stated that if an individual has a taste

for discrimination, they ust be willing to pay sonethin, either

directly or in the form of a reduced inane, to be associated with same

persons instead of others (1957:6). When these tastes are operative,

blacks in order to secure a job must accept lower wages than whites. A

consequence is that while white employees will benefit from higher

wages, employers will incur higher monetary costs. The outcome of a

taste for discrimination will be workplace egregation and the existence

of wage differentials by race (Banton 1983:369).


Economic Sectors and labor Markets

As noted earlier, neoclassical views on discrimination have been

subjected to many criticisms. One of the most important criticisms is the

theory's inability to explain long-run differences in wages. The

shortcomings of neoclassical views led economists in the 1970s to turn

their attention to the structural characteristics of economic sectors

and labor markets and the relative demand for and position of blacks

within them. The North American focus on differentiated economic and

labor sectors is on segmented and split labor markets. The latin American

literature seeks to explain inequality by focusing on formal/informal








22

labor markets. iTese approaches emphasize the differentiation of economic

sectors and their associated labor markets as determinants of wage

inequality.


Segmented Labor Markets

Segmented labor market theory arose from a desire to understand the

persistent poverty of different social groups and the continuation of

racial and gender inequalities (Cain 1976). Ihe main thesis in segmented

labor market theory posits the existence of qualitatively distinct

sectors of labor, restricted intr-sectoral mobility, and a

disproportionate and unfair relegation of certain types of labor (for

example, black or female) to the least favorable sector (Boston

1988:101). Unlike the neoclassical emphasis on equilibrium (which

assumes a hamogeneous labor market), and different from Becker (who

emphasized employer's tastes and preferences), the primary emphasis is

on structural-level determinants of inequality. he disproportionate

allocation of rnahites and women to lw-paying sectors of the eonnmy

and labor market is the key idea in the perspective's account of racial

and gender differences in earnings. There has been much attention

devoted to describing the "balkanization" of the labor market, that is,

the idea that labor markets are fragmented into nonmcnpeting groups.

Perhaps the best known approach is the dual labor market thesis which

suggests that the United State economy is divided into two distinct

sectors: a core and periphery.1

Several explanations are advanced for the existence of labor market

dualism. In its earliest version, Boeke (1953) conceptualized modern and

traditional sectors as separate and parallel within a single society: the











result of an intrusive modern eocnic sector which failed to transform

the rest of society. Other versions have interpreted dualism by focusing

on the industrial structure of the econy. The economy is differentiated

between core industries with advanced technology and capital-intensive

enterprise and peripheral industries which are more labor intensive,

with smaller enterprise size, local rather than imported technology,

and lack of formal education entry require nts (Bluestone et al.,

1973).

Corresponding respectively to the core and peripheral sectors of the

economy are primary and secondary labor markets. 7he primary labor

market is characterized by an array of "good" jobs, which may include

unionization, advaneent opportunities, stable employment, benefits,

and a relative high wage. Jobs in the secondary labor market differ on

each of these dimensions. They have low wages and undesirable, often

unsafe, working conditions. There are limited chances for advancement

and arbitrary enforcement of work rules. These conditions promote

frequent job turnover and job instability. A basic tenet of dual labor

market theory suggests that disadvantaged and underprivileged workers

are more likely to occupy secondary labor market positions than are

white males (Berger and Piore 1980:23).

Piore (1971) described the process by which workers are allocated

to either the primary or secondary market. With technological

development, on-the-job training became more and more important, and

employers invested heavily to train employees. On this basis, workers

were separated into two groups: those possessing the necessary

"trainability" characteristics and canmitment to work, and those that,










for one reason or another, tend to work unreliably and intermittently.

his egmentation led employers to devise a labor queue leading to a

process of "statistical discrimination."2

The concept of "statistical discrimination" refers to the situation

in which employers tend not to employ mmBars of certain groups because

their superficial characteristics seem to be statistically associated

with undesirable behavioral traits. In this case, an individual is

judged aaccrding to the aodal (real or inputed) characteristics of the

group to which they belong rather than by their own characteristics.

Thus, if blacks are considered unreliable workers in terms of their job

stability, then individual blacks will probably suffer from "statistical

discrimination" in the sense that blacks are allocated to a secondary

position in the labor queue, regardless of their objective

characteristics, including personal job ocmintment.

Proponents of the segmented labor market approach further criticize

the assumption, endorsed by neoclassical ecomnistsr that labor market

discrimination is dysfunctional and hence transitory. In a study of

racial inequality in the USA, Reich provided evidence to support a

"divide and rule" hypothesis by calculating an index of income

inequality within labor market segments. His argument was that labor

market segmentation "played a major role in channeling the effects of

past and present race and sex discrimination" (1981:16). The effect of

segmenting workers on the basis of race or gender was viewed as

functional to capital inasmuch as such antagonism enhanced the rate of

accumulation by serving to divide, and thereby weaken, labor's bargaining

power in the workplace.










Ihe dualist perspective contrasts in emphasis with the neoclassical

economic model. In the segmentation perspective, the crucial distinction

is between good and bad jobs, not between skilled and unskilled workers,

as the neoclassical approach suggests. Hence, dual economy theory is

conceptualized at the level of the firm, not at the level of the

individual. The central premise of segmented theory is that economic

characteristics of firms are interrelated with the structure of the

labor market and, as a result, there are conseqnes for earnings. The

most pertinent question for this analysis, which will be taken up in

Chapter 6, is how do worker's earnings vary by economic sector?

Split labor Markets

Introducing racial conflict into a model of racial and ethnic

inequality, Edna Bon~cich (1976) advanced the idea of a "split labor

market." Rather than enhancing capital's control over labor, as the

segmented labor market is said to do, the split labor market is seen as

problematic for capital, yet beneficial to certain groups of workers.

This perspective stressed the existence of a group of dominantt workers"

which is threatened by the competition of "cheap labor." A split labor

market refers to a different in the price of labor between two or more

groups of workers.

Capital turns toward the cheaper labor pool as a more
desirable work force, a choice consistent with the
simple pursuit of higher profits. Higher priced
labor resists being displaced, and the racist
structures they erect to protect themselves are
antagonistic to the interests of capital. (1976:44)

A racially (black/white) split labor market, according to Bnacich

began with slavery and persisted well into the twentieth century in











industrial America. he exclusionary strategies were not racially

motivated or "nationalist" per se, but "the product of historical

accident which produced a correlation between ethnicity and the price

of labor" (Banacidh 1980:14). This correlation stems frm the existence

of a "worldwide division of labor, migration patterns, unequal exchange

and uneven development processes" (Cani & Winant 1986:35). While the

segmented labor market is said to enhance capital's control over labor

by dividing and thus weakening the bargaining power of labor, the split

labor marker is contrary to the interest of capital because "dominant"

workers use exclusionary tactics to keep wages high. Class conflict

takes place between capital and the dominantt group," which seeks to

maintain its wage levels and defend whatever control it may have of

production processes. Racial/ethnic conflict takes place between

"dacinant" and "subordinate" workers as the former seek to prevent the

latter from bidding down the price of their labor.

Infomnal/ al Sectors

The frame of reference for segmented/split labor market theory has

traditionally been the United States. A similar sectoral analysis of

labor absorption applied to latin America is the distinction between

formal/informal modes of labor utilization (Portes 1981; Tokman 1978;

Portes and Benton 1984). Formal labor is defined by contractually

regulated wages controlled under labor laws. Contractual employment and

legal coverage protect workers against arbitrary dismissal and also

give them access to programs of health and disability Insurance,

unemployment compensation, and retirement. As a result, the remuneration

of the formal worker consists of two components: a direct monetary wage








27

and an "indirect wage" formed by the various insurance and other programs

prescribed by law.

Conversely, informal workers do not receive regular money wages, nor

"indirect" wages, and relations with employers are not contractual, and

what agreements exist between worker and employer are beyond the

regulatory scope of the state. Informal workers are remunerated in

various ways that often include a wage that is verbally agreed upon, a

piece rate, and nrmonetary oao nation such as food or hosing. The

low incomes and irregular deployment of informal workers cxpel them to

engage in supplementary activities such as animal raising, food

cultivation, and construction of their own shelter.

Informal labor markets were previously labeled "marginal" in Latin

America because its members were believed to survive outside the modern

capitalist sector and to lack any integration with it. However, studies

severely criticizing the concept of marginality (Perlman 1976) have shown

the integration of the informal worker into the modern economy (Portes

and Benton 1984; Worsley 1984). While particular attention has been

given to the informal employment of female workers, this literature has

not dealt with distinctions on the basis of race.


Human Capital

In the study of earnings inequality, the status attainment tradition

within sociology (Blau and Duncan 1967; 0. Duncan, Featherman, and Duncan

1972; Sewell and Hauser 1972; Haller and Portes 1973; Featherman and

Hauser 1974) and the human capital perspective within economics (Becker

1964; Becker and Chiswick 1966; Thurow 1969; Schultz 1962; Mincer 1970;










Miner and Polachek, 1974) are onmmrnly applied to explain the phenmena

of poverty and wage differentials. Within both traditions, the emphasis

is on individuals rather than labor market factors or economic sectors

as determinants of earnings.

Studies of status attainment emphasize the role of familial origins,

educaticf and othef indicators of-socialization experiences as

determinants of occupational status, ability, and, more recently,

earnings. Similarly, human capital analyses emphasize the importance of

"investment" in education and skills as a means of yielding "returns"

in earnings. Both of these traditions attribute increases in either

social mobility or income to the quantity or quality of individual

attributes. The contribution of human capital and status attainment

models has been to advance analyses of racial inequality beyond the

descriptive level.

Most of the contemporary research on earnings inequality relies on

an analysis of human capital. The justification is that human behavior

and social outcomes are explicable in terns of individual attributes.

Individual attributes, in turn, are viewed as distributed along a

continuum which is reflected in, and used to explain, the different

levels of individual earnings. The approach is rooted in neoclassical

economics, and applies essentially the same terms that are used to

characterize czmnodity markets (Sakamoto 1988). Capital stock is

accumulated as the result of optimizing decisions made by the individual

and their family about the allocation of investments in education and

training over their life cycle (Becker 1964). Wages, in turn, reflect

an individual's marginal productivity, or the value of the goods and










services produced. Emoncmists conclude that wage differentials by race

are due, not to discrimination, but to individual differences in skills

and qualifications.

Analyses in the human capital and status attainment tradition

focus on microlevel eplanatimc of inequality by confining their

attention to properties of the individual. Ihe economic sector and

labor market approach reviewed earlier incrporates structural

characteristics of institution and/or labor markets into the analysis

of inequality. The goal in this study is to draw upon both approaches.

Subsequent analyses of wage inequality retain the individual as the

unit of analysis, yet disaggreate the measure of discrimination by key

structural features of the Brazilian eoncxmy (see Valknen 1969; Farkas

1974; Boyd and Iverson 1979; and Parcel and Mueller 1983 for similar

approaches). The particular labor market ontets investigated in this

study are geographic labor markets, Chapter 5, industrial labor markets

(or economic sectors), Chapter 6, and oupational labor markets,

Chapter 7.


Methodolocal Cveats

Wage differences between equally productive white and black workers

are a standard and intuitively appealing measure of the extent of labor

market inequality (Cain 1976). Assuming that earnings are a function of

human capital stock, the usual test for the effect of race on earnings

is to standardize other theoretically relevant factors. The most

frequently encountered human capital model uses individual-level data

to estimate a regression model relating earnings to schooling, labor











force experience, and other job-related characteristics which influence

an individual's wages. This equation is generically referred to as the

"earnings function." The equation can be estimated in two ways: using

a dummy variable to represent the additive and interactive effects of

race or alternatively separate equations are estimated for each

population.

The value of this approach to estimating wage inequality depends

on the proper specification of the earnings function since the amount

of variance in wages explained by the model depends an the extent to

which the model is correctly specified. Hypothetically, if the model

is precise, then no variance in earnings would be left unexplained.

Given the improbability of ever estimating the perfect model, we make

the most informed choice possible consistent with the literature and

the data available.

The standard earnings function based on education, job experience,

and additional personal, market and demographic variables thought to

influence the wage, has been used in hundreds of studies using data

from virtually every historical period and country for which suitable

data exists. The results of these studies reveal that the model is

robust and applicable across a variety of settings (Willis 1986). Data

permitting, measures of cultural differences in attitudes towards

education, days spent in school, quality of school resources, per-pupil

expenditures, academic achievement, number of math courses taken when

in high school, and similar variables could also be introduced into the

earnings function in order to explain further variance in income

inequality (Sowell 1981:21-23). One can easily see how the list of











potential explanatory variables b unrealistically long thereby

reducing the likelihood of finding a data source of this kind. It is

very doubtful, however, that such data would alter the main conclusions.

Measuri Discrimination

The concept of racial discrimination, although seldom explicitly

defined, is generally ackwledged to exist wheneverr "race is a factor

in predicting the aprtunities _pen to an individual" 7CIurow 1969:1-2).

A measure of labor market discrimination, found in both the eomrnmics

and sociological literature, is obtained by deomposing the mean wage

Difference (estimated in separate earnings functions by race) between

the populations into a "cmposition difference" (e.g. explained

differences due to differing levels of education) and a "rates

difference" (e.g. unexplained differences due to unequal returns to

education). The latter represents wage "discrimination", that is, the

higher wage received by a white worker than by a black worker for the

(assumed) same productivity characteristics.

A fundamental problem with all studies of differential wage

determination by race is the fact that the key concept--discrimination-
S -is inferred rather than measured directly. The inference is based on

6C' the interpretation of a statistical residual. That is, the earnings

function (i.e. education, experience, etc.) is used to account for the

variance in wages associated with different levels of human capital

endowments. The variance that is not amounted for by these variables,

by default, is attributed to discrimination.

This interpretation of the residual must be treated with caution.

This is because other measurede) concepts could aooournt for the










variance unexplained by the earnings function. Fbr example, if blacks

are less motivated or less Intelligent than whites, the residual could

be explained by personality or genetic attributes. Analysts who endorse

such racist assumptions could infer that wage inequality was based on

the inferiority of nawhites. However, there is little supporting

evidence that such heKnmena exist.

Another exlam tion miht be that using years of cooling as an

indicator of "educational attainment" ignores difference in the

"quality" of schooling. In this sense, although two individuals may

have the sae quantity of schooling, the quality of that education

differs. More precisely, it is argued that, compared to whites,

nowhites receive an inferior education. However, as D~ucan (1969:104)

observed: "for one thing, inferior quality at any one level of the

school system is likely to result in impaired chances as proceeding to

the next level. Hence, school years completed has partly built into it

a correlation with quality." his argument suggests that if differences

in quality of education exists, the bias introduced should be relatively

minor (Silva 1978:188).

In the final analysis, the matter cannot be resolved an the basis

of the data at hand. Ihe remaining strategy is to turn to other sources

of relevant information. While the record shows little evidence of either

motivational or genetic differences between racist groups, there is ample

evidence, as reviewed in Chapter 1, of racial prejudice against mulattoes

and blacks in Brazil. Hence, it is plausible to interpret the residual

in the earnings equation as discrimination. The caveat is, discriminatory

behavior is inferred rather than measured directly. We must therefore










treat the conclusion with caution, making the most informed inferences

about residuals which are consistent with the literature and plausible

in light of additional data.

There are four basic ways by which labor market discrimination

against nonwhites can be accomplished (Silva 1978):

1. human capital discrimination-not whites can have their mobility

char els blocked by being prevented frao getting the necessary

qualifications to enter higher paying occupations;

2. Employment discrimination-xnnwhites can suffer more than their

proportional share of unemployment;

3. Occupational discrimination-nonwhites can be prevented fram

entering same better paying occupations, regardless of whether they are

qualified or not;

4. Wage discrimination-ncnwhites can earn less for performing the

same jobs as whites, i.e., unequal pay for equal work.

The first type of discrimination takes place mostly before the

individual enters the labor market, largely still within the schooling

system. This type of discrimination has been addressed by the status-

attainent literature in sociology. 7he second and third types of

discrimination occur within the allocation of laborers to certain

occupations and industries. An analysis of this process is imbedded in

the literature on labor market structures.

The final form of discrimination conceptualized in the human capital

tradition, wage discrimination ,takes place after one's entrance in the

labor market. Discrimination is evidenced by differential returns by

race to equal quantities of human capital endowments. The analysis in










the chapters that follow relies on the basic relationships posited by

the human capital perspective, but takes into aco=unt several additional

factors. Specifically, the objective is to investigate how the degree

of discrimination varies by region and within different sectors of the

ecomy and occupations. In this way the analysis seeks to account both

for the individual determinants of earnings as well as same of the

structural features of the labor force that affect the relationship

between human capital, earnings, and race.







1. Guiding theoretical analyses include Averitt (1968), Gordon (1972),
O'COnner (1973), Edwards (1975), Hodson (1978), Wachter (1974), Doeringer
and Piore (1971), Piore (1975), Bihb and Form (1977), and Wallace and
Kalleberg (1981).

2. ITurow (1975) is another economist who uses the term "statistical
discrimination."














CHAPTER 3
DATA AND VARIABLES


-he .8 percent sample of the 1980 Brazilian census is the data

source for the present study. 'he 1980 census is unique for two

reasons. First, amnng developing countries, the Brazilian Census Bureau

(FIBG) is renowned for the quality ard overage of its statistics.

he decennial censuses meet or exceed international standards (Wood and

Browning 1988) and the last two are available in sample form. The 1980

sample, which contains nearly one million cases, permits maximum

flexibility in the analysis of individual characteristics. The

individual level indicators include race, sex, age, incroe, education,

occupation, industry, number of hours worked, social security coverage,

migration, and marital status. These variables provide the basis for

generating a profile of socioeoc ic and human capital differences

between the white, black and mulatto populations.

The second advantage lies in the fact that it is the only census

available to the public that allows disaggregation by race. The

question about color was included in the 1950 and 1960 censuses.

However, sample data are not available from the 1950 census, and the

results of the 1960 enumeration were never published in their entirety.

The 1970 Census, which was carried out during the military's rule, did

not include the race item at all. Moreover, the 1976 national household











survey dealt with the question of race, but the data have not been

distributed by the bureau.

In order to permit inter-region analyses, the 1980 sample (.8

percent version) was constructed in such a way that no less than 4,000

households were selected at the lowest level of geographic

disaggregation (rural and urban areas; metropolitan areas within

states). As a result of this, the unweighted sample over-represents the

smaller units and under-represents the larger ones. Contry-level

analyses thus require that the sample data be weighted on the basis of

the proportion of the actual population size in each state reported in

published census volumes. Hene, in this study, weight factors are used

in all analyses except in Qiapter 5, when the regions are examined

separately.

The sample used in this study is restricted to black, mulatto, and

white metropolitan male employees aged 18 to 64 who receive a wage.1

The race classification in the census data is a self-identification

question with a four-category scheme: white (rano), black (pret),

brown (pLrdo), and yellow (asarelo .2 Since the Oriental populations

constitute about 0.1 percent of the total population, Asians were

excluded.

The Validity of the Census Question on Race


There has been discussion in Brazil regarding the validity of the

four-fold census classification of race. The issue is pertinent in

light of anthropological research on race relations in Brazil that

documents fine distinctions Brazilians make when asked to identify a











person's race. Harris (1964), for example, used a set of nine portrait

drawings to explore the range of terms that might be applied to a given

individual. Variable in hair shade, hair texture, nasal and lip width

and in skin tone, the pictures elicited forty different racial types.

The wide range of terms people use to identify color variations between

the two extremes of black and white would appear to invalidate the

simple fourfold classification system currently used in the census.

onerned with the validity of the forced choice question,

researchers at the census bureau set out to investigate the extent to

which the census scheme departs from an individual's self-

classification if allowed other options. To do this, the 1976 National

Household Survey (PNAD) was designed with two questions on race. The

first was an open-eded item which permitted resp~ dents to use

whatever term they wished. The second was the standard fourfold

classification. Analyses of the open-ended item showed that, despite

the wide range of terms, the four categories (white, black, brown, and

yellow) accounted for about 57.1 percent of the responses. Three

additional classifications (which range from light to dark brown)

proved to be important: cla (2.5 percent); morena clara (2.8

percent), and morea (34.4 percent). Further analyses found that nearly

all of the people who declared themselves rena in the open-ended

question classified themselves as brown (arog) when confronted with

the pre-coded options.

Since the four-category scheme accounted for apprmdmately 95

percent of all responses (Oliveira, Porcaro and Costa 1981), analysts

in the Census Bureau concluded that the forced-choice method, although











not perfect, was sufficiently reliable to be used in the 1980 census

enumeration. Because the census is based on self-identification, and in

light of the cultural content of the way people classify each other in

Brazil, the census typology must be understood in the social sense,

making no claims that it is an accurate means to classify biological

groups.

Sample Restrictions

The primary objective of this study-the analysis of wage

discrimination-required certain sample restrictions. Wmnen, rural

workers, the unemployed, employers and the self-enployed were not

included in the analysis. Female workers were excluded for three

reason: (1) occupational segregation is uch higher for women than

men, thus restricting the number of occupatinal categories available

for analysis; (2) earnings are subject to underestimation because

inooe for a very large number of women are not determined by employers

but by self-enployment, oommissions, piece-rates, and other forms of

accpensations; and, (3) employment for women tends to be non-continuous

due to family responsibilities that require women to leave the labor

market during pregnancy and at childbirth. Even when mothers stay in

the labor force, family responsibility frequently constrains job

choice. As a consequence, wcuen,'s work history, and the factors that

determine women's occupation and wages, are substantially different

from men's. An additional reason for restricting the sample to males

is that men constitute the empirical base of all previous studies of

wage inequality in Brazil (See Silva 1978, 1985). This opens the










possibility for comparison between these earlier studies and the

results of this research.

Rural workers were also excluded from the sample because of

similar difficulties in assessing earnings. Most rural workers are

employed as agricultural laborers on a seasonal basis and are

frequently rmunerated in terms other than wages. An additional reason

for excluding rural workers, is that Brazil's population, unlike most

other Latin America countries, is pr;dminantly urban. By 1980 the

rural population in Brazil ormprised only 32 percent of the total

population (Haserbalg and Silva 1987).

Other restrictions of the sample relate to men who were unemployed

and those who were enployprs or self-enployed. The inclusion of these

three groups would have introduced a variety of factors unrelated to

the wage/labor issue. Workers reporting no earnings were excluded to

eliminate missing data problems with the deper~ent variable (see Beck,

Horan and Tolbert 1978; and Hauser 1980, for a dismssion concerning

this issue). By restricting the sample to men who receive a wage, the

result is to reduce the wage gap below what it would be for the full

population of black and white male workers, given that blacks males are

more likely than whites to be unemployed.

Variables Used in the Analysis

The sanple used in the following analysis includes 38,711

respondents. Eight explanatory variables are used throughout the study.

Subsequent tables present the sample means by race for the various

indicators. This section describes these variables and summarizes the

rationale for including each of them.











Color. The distribution of the 1980 metropolitan sample

population by race is presented in Table 3.1. The white population

accounts for 63 percent of the total population. Mulattoes and blacks

amount for 30 and 7 percent, respectively.


Table 3.1 Distribution by ace, Male Workers 18-64 Years of Age,
Metropolitan Brazil, 1980


Black Mulatto iaite Total

% 7 30 63 100

N 2574 11581 24556 38711

Source: 1980 Brazilian Census, .8 percent subsanple.


Wages. Earnings are the-key dependent variable in investigations

of labor market discrimination. Wages are measured as "average monthly

earnings frmn the primary occupation." Mrothly wages, rather than

annual income, is used because this measure cirumvents the need to

control for length of time worked during the year (information which is

not available in the 1980 Census). Failure to control for spells of

unemployment would bias the estimate of earnings. Wages are reported in

ruzeiros, the Brazilian currency in 1980. AcoordiIn to the exchange

rate in August 1980, Cz$5552.8 were equal to $100 United States. Table

3.2 presents the distribution of average monthly income by race.

The estimated wages for each racial group indicate the existence

of substantial racial differences. The average monthly earnings for the

white population (Cz$21,867.15) is more than twice that of blacks

(Cz$9,003.78). Interestingly, there is a relatively small difference










(Cz$2049.52) in average earnings between the black and mulatto

populations.


Table 3.2 Mean Monthly Wages, in Cruzeiros by Race, Male Workers 18-
64 Years of Age, Metropolitan Brazil, 1980


Black Mulatto White Tital

Cz$9003.78 CZ$11053.30 Cz$21867.15 Cz$17776.75

Source: 1980 Brazilian Census .8 percent m*ammple.

Schoolin. E nation is a standard operational definition of the

human capital a worker brings to the labor market (Becker 1967).

Bducation imparts valuable skills and training that directly enhance

new's productivity and potential earnings. In addition, education may

increase a worker's ability to be more easily trained while employed.

Ecroanists further ocrtend that more highly educated workers may have

additional mental and social skills that facilitate on-the-dob training

and job mobility. Education is thus considered a key factor in the

determination of income.

Table 3.3 presents the distribution of schooling by race for male

workers. The averages for the black and mulatto populations are

relatively close to each other but are much lower than those for the

white population. The same pattern was observed before in the income

distribution by race. White workers are more than seven times as likely

as blacks to have twelve or more years of schooling. Blacks are more

than twice as likely as whites to have no education.











Table 3.3 Schooling Distribution by Race, Male Workers 18-64 Years of
Age, Metropolitan Brazil, 1980


Yrs Black Mulatto White Total


0 16% 13% 6% 9%
1-4 53 51 38 43
5-8 21 22 22 22
9-11 9 10 19 15
12+ 2 4 15 11

Total 100 0000% 100% 10% 100%
N 2574 11581 24556 38711

Source: 1980 Brazilian Census .8 percent subsanle.


Exerience. Labor market experience is another major component

of an individual's human capital. This variable is often measured in

terms of the number of years that a worker has been in the work force.

As with education, the effect of experience on earnings is due to

enhanced productivity. Also, like education, an individual's investment

in their stock of on-the-job training is said to reflect a conscious

decision to maximize lifetime earnings.

The 1980 census does not contain information about the

respondent's work history. Therefore, a transformation of age is used

as a proxy measure of labor force experience. Labor force experience

is calculated by subtracting from the individual's age the number of

years of school cncpleted and a constant of six, the average age at

which a child begins school: Experience = Age (years of schooling +

6). Although use of this substitute assumes continuous labor force

involvement, it provides a more meaningful measure than other ccmmon

proxies, such as age (Silva 1978). Table 3.4 presents the average labor

force experience distribution for blacks, mulattoes, and whites. These











findings show that job experience is almost identical across racial

groups.


Table 3.4 Average Years of labor Force Experience by Race, Male Workers
18-64 Years of Age, Metropolitan Brazil, 1980



Black Milatto White Total


22.45 21.58 20.59 21.01

Source: 1980 Brazilian Census .8 percent sumeanple.



Region. Pronouned regional inequality has long characterized

Brazil. Development has favored central and southern Brazil, while the

north and northeastern states lag far behind the rest of the country in

terms of inwcme, educational achievement and other standard of living

indicators. Thus, the ability to capture regional difference is

central to the study of wage inequality in Brazil. The nine

metropolitan breakdowns suggested by the Council of Statistics-Belem,

Fortaleza, Recife, Salvador, Belo Horizante, Rio de Janeiro, Sao Paulo,

Curitiba, and Porto Alegre-offer a broad view of the most salient

features of Brazil's spatial diversity and are used in the following

analysis.

Table 3.5 presents the regional distribution by color. The pattern

is clear: a disproportionate number of blacks and mulattoes, as

ocmpared to whites, live in the underdeveloped, agrarian North and

Northeast of Brazil, where educational and ecorKnic opportunities are











Table 3.5 Regional Distribution by Race, Male Workers 18-64 Years of
Age, Metropolitan Brazil, 1980

Region Black Mulatto White Itatal

Belem 4% 71% 25% 100%
Fortaleza 29 69 3 100


NCRIHEAST
Recife 38 57 6 100
Salvador 24 59 17 100


SCUIHEAST
Belo Hor 9 39 53 100
Rio de Jan 11 29 60 100
Sao Paulo 4 23 73 100



Curitiba 2 14 84 100
Porto Aleg 5 8 86 100


Source: 1980 Brazilian Census .8 percent subsanple.


scarcer than in the Southeast. The greater part of the white population

is concentrated in the South, the most econically developed in

Brazil. For example, in Salvador, nnhites comprise 83 percent of the

population. In Sao Paulo, 73 percent of the male workforce is white.

Location. The 1980 Brazilian Census used a detailed

classification scheme to record respondent's primary occupation in the

12 month period preceding the Census. Unfortunately, the Brazilian

Census Bureau's codes confound conventional occupation categories with

industrial sector (Merrick and Graham 1979). This results in an

occupational breakdown which often reflects the branch of activity as

much as skill and status differences found in more conventional

occupational classifications. To account for the range of status and











skill levels, the original occupational classifications have been

recorded into four summary categories (See Appendix 1 for the Census

categories and recedes). The categories are: Managers/Anministrators;

Professional/Technical; Clerical; and Blue Collar.

Table 3.6 presents the distribution of workers by occupation and

race. There are substantial difference in occupational attainment

between racial groups. Wite workers are more likely than m.lattoes

and blacks to be in higher skilled copations. Fr emmnple, while the

proportion of whites in administrative and professional oc4upations

(the first two occupatinal groups) constitutes 27 percent of the white

population, the corresponding figures for the mulatto and black

populations are ten and seven percent, respectively. On the other hand,

the estimated proportion of blacks in blue collar positions is 84

percent, compared to 79 and 58 percent for the mulatto and white

populations, respectively.


Table 3.6 Occupational Distribution of Workers by Race, Male Workers
18-64 Years of Age, Metropolitan Brazil, 1980

Black Mulatto white Total

Man/Adm 2% 4% 11% 9%
Prof/Tech 5 6 16 12
Clerical 9 11 15 14
Blue Coll 84 79 58 66

Total 100% 100% 100% 100%

Source: 1980 Brazilian Census .8 percent subsaxrple.


ours worked. The total hours worked per week is an important

explanatory variable because it has a direct impact on earnings. In











particular, if the observations made by dual labor market theorists are

correct, nowhites may be restricted to jobs in the secondary labor

market. Secondary sector jobs are characterized by periodic or chronic

unemployment. Clearly, if nonwhites are overrepresented in this sector,

differences in income could be the result of differences in hours

worked.

The measure of hours worked used in this analysis is a dichotnous

variable coded for either part-time or full-time weekly employment in

the primary occupation. Table 3.7 presents the distribution of full-

time employment by color. The results show that all but roughly five

percent of each group is likely to be employed full-time.

Table 3.7 Proportion IEployed Full-Time by Race, Male Workers 18-64
Years of Age, Metropolitan Brazil, 1980

Black Mulatto White Total

95% 95% 94% 94%

Source: 1980 Brazilian Census .8 percent subsanple


Social Security. In addition to oupation and hours worked,

social security coverage is an indicator of the wage/labor

relationship. According to Portes (1985), social security coverage

represents the best empirical identifier of the formal/informal

economic sectors. The formal sector is characterized by regular

employment and higher wages which are contractually established and

regulated under existing labor laws. On the other hand, workers in the

informal sector do not receive regular contractual wages nor do they

receive the "indirect" wage of social security. The purpose of










introducing this measure is to control for sectoral earnings

differences by race.

The validity of social security as a proxy measure of employment

in firms that ocuply with labor legislation is questionable in Brazil.

Given is that individuals can elect to pay into the social security

system regardless of their source of employment. For example, a self-

employed street vendor has the same option as a government employee to

belong to the social security system. Table 3.8, indeed, shows that

the majority of metropolitan workers receive social security coverage.

White workers are only three percent more likely than blacks or

mulattoes to be covered by social security. these results, if used as a

measure of the formal/informal labor market, overestimate the

proportion employed in the formal sector.

These observations do not totally invalidate the use of this

measure in the analysis that follows. Membership in the social security

system is positively correlated with earnings. The point is that

caution must be used in interpreting social security, as Portes and

others would, as valid measures of economic sector.


Table 3.8 Proportion Covered by Social Security by Race, Male Workers
18-64 Years of Age, Metropolitan Brazil, 1980


Black Mulatto White Total

92% 92% 95% 94%

Source: 1980 Brazilian Census .8 percent subsample.


Background Variables. Two background variables, migratory and

marital status, are also included in the analysis. Migratory status











contributes to the explanation of earnings differentials in an

industrialized environment. Migrants tend to move to areas with

increased opportunities, bringing with them characteristics which

compare favorably to natives in the new environment. Thus, studies of

Brazil have shwn that the wages of migrants usually are higher than

those of natives (Martine 1977). Migrants are defined as those not born

in the state of residence. As Table 3.9 demstrates, mulattoes tend to

migrate slightly more than whites. Blacks are the least likely to

migrate.


Table 3.9 Proportion of Migrants by Race, Male Workers 18-64 Years
of Age, Metropolitan Brazil, 1980


Black Mulatto White Total

30% 40% 37% 37%

Source: 1980 Brazilian Census .8 percent subsanple.


The second background variable, marital status, is a measure of

the respondent's family responsibilities and is often used as an

indicator of the individual's commitment to work. In this sense,

marriage is presumed to affect a worker's productivity and hence, their

employability and income. This is the rationale for including this

variable in the present study. Marital status is coded in two

categories, married and not married,the latter includes those who are

single, separated, divorced and widowed. Table 3.10 presents the

proportion married by race. Nearly two-thirds of all workers in the

sample are married.












Table 3.10 Proportion Married by Race, Male Workers 18-64 Years of
Age, Metropolitan Brazil, 1980

Black Mulatto White Total

64% 66% 66% 66%

Source: 1980 Brazilian Census .8 percent subsample.

umnarv and Discussion

Ihe socioeconxnic profile of the three racial groups shows

surprising similarities in terms of background characteristic such as

job experience, hours worked, social security coverage and marital

status. Yet the data attest to sharp racial inequalities in wages,

education, regional distribution, and occupation. In 1980 the average

wage of black metropolitan workers was less than half that of white

workers (Table 3.2). Unequal aces to educational oortnities is

another crucial dimension of racial inequality. Table 3.3 shows that

blacks and nulattoes are twice as likely as whites to be illiterate. At

the upper end of the educational hierarchy, whites have a 3.1 greater

chance of completing nine or more years of school than blacks. The

unequal geographic distribution of whites and noaKhites is yet another

determinant of oantenporary racial inequality in Brazil. The greatest

part of the white population resides in the most economically developed

regions, the South and Southeast, while blacks and mulattoes are

disproportionately concentrated in the underdeveloped, agrarian North

and Northeast. Finally, the partition of the three racial groups in the

labor force, with respect to occupation, is another dimension of racial

inequality. Blacks and mulattoes are concentrated in blue collar jobs,










the occupations that absorb most of the unskilled and underpaid

workers. Conversely, whites are concentrated in administrative and

professional occupations, where better-paid jobs and higher skilled
labor are found.

It is reasonable to expect that the inequalities between whites

and norrhites in schooling, geographical location, and employment

structure have a strong effect an individual earnings. The following

chapter addresses a fundamental question: Once we statistically

control for the differences in education, region, and occupation, does

wage inequality persist or disappear?




Notes



1. Cases with missing data in any dependent or explanatory variable
have been eliminated.

2. The ar (brown) category refers to individuals who declare
themselves to be somewhere in between black and white (e.g. mulatto,
caboclo, ,ore and indio). Because it is impossible to distinguish
between those of Indian and African descent, I will use the term
"mulatto" and "nmowhite" rather than "Afro-Brazilian" to refer to
panos. Given that the primary objective of this study is to determine
if workers are paid differentially on the basis of skin color, the
inability to distinguish between ndis and those of African descent is
not a crucial issue.















CHAPTER 4
MEASURING WGE INQUALTY


The goal of this chapter is to examine the earnings gap that

exists between the three racial groups in Brazil. The tabulations given

in the previous chapter (Table 3.2) clearly show that white male

workers have significantly higher earnings than do nonwhites. kMreover,

within the nonwhite group, nlattoes earn more than blacks. But what

aaounts for wage differences?

The explanation posited by both the "class over race" and human

capital arguments, reviewed in Chapters 1 and 2, suggests that the

incone gap results from different levels of endowments (e.g.

education). This explanation is plausible, especially in light of the

findings (Table 3.3) that show nulattoes and blacks have lower levels

of schooling than whites. Given that education is one of the central

determinants of wages, it is possible that once simultaneous controls

for education and additional wage-predictors such as occupation are

introduced, the earnings gap between the races may lessen or even

disappear. Three avenues of research, each of which requires a separate

analysis, will be followed to test whether wage differences persist

between whites and nonwhites after controlling for human capital and

demographic factors.











The Complete Earninms Function


The first step in the analysis is to estimate a model using

ordinary least squares regression techniques that relates a vector of

individual characteristics to earnings. The equation will introduce

controls for race, experience, education, occupation, region, hours

worked, social sewrity coverage, and migrant and marital status. The

owariate analysis of the relatiarnhip between earnings and the wage-

predictors listed above is modeled as:

Y 80 + BIRaoe B2EXP + B3EDP2 + BiSCHi + BiRDGi + BiOCCi + B19HRS +
B20SOC + B21MIG + B22MAR

where Y is the log of monthly earnings1 from a primary occupation; Race

represents the three populations; EXP is number of years in the

workforce; EXP s iS the experience-squared term; SCoi's represent

different schooling categories; REXi's are the different metropolitan

regions; OCCi's represent different occupational categories; HRS

represents total hours worked; SOC is the social security indicator;

MIG is the migrant status indicator; and MAR is the marital status

measure. In the analysis that follows particular attention will be paid

to the dummy variable for race.

The intercept of the OLS regression model represents white male

respondents with no schooling residing in Sao Paulo who are employed in

blue collar occupations, work part time, do not receive social security

coverage, are non-migrants, and are unmarried. Table 4.1 shows the

units in which the variables are measured and Table 4.2 presents sample

means and standard deviations for the various indicators.











Table 4.1 Definitions of Variables Used in Earnings Regressions


Definition


EPENDNT VARIABLE
Natural log of
wages


Average monthly wage in
Cruzeirs


Race
White
Mulatto
Black

Experience
Experience2


Schooling
0
1-4
5-8
9-11
12+


Region
Sao Paul
Belem
Fortaleza
Recife
Salvador
Belo Horizonte
Rio de Janeiro
Curitiba
Porto Alegre

Occupation
Blue Collar
Manager/Admin
Prof/Tec
Clerical


Hours


Social Security

Migrant


Marital


-0 if
-1 if
-I1 if


white
mulatto, 0 otherwise
black, 0 otherwise


Age (years of schooling + 6)
Square of experience


-0 if 0 years
-1 if 1 thru 4 years, 0 otherwise
-1 if 5 thru 8 years, 0 otherwise
=1 if 9 thru 11 years, 0 otherwise
-1 if 12+ years, 0 otherwise


-0 if Sao Paulo
-1 if Belem, 0 otherwise
=1 if Fortaleza, 0 otherwise
-1 if Recife, 0 otherwise
-1 if Salvador, 0 otherwise
-1 if Belo Horizonte, 0 otherwise
-1 if Rio de Janeiro, 0 otherwise
-1 if Curitiba, 0 otherwise
=1 if Porto Alegre, 0 otherwise


-0 if Blue Collar
-1 if Manager/Admin, 0 otherwise
-1 if Prof/Tec, 0 otherwise
=1 if Clerical, 0 otherwise

-0 if part-time, 1 if full-time

=0 if not a member, 1 if a member

=0 if a native, 1 if a migrant

=0 if not married, 1 if married


Variable











Means and Standard Deviations of Variables Used in
Regression Equation with Race Dummy, Male Workers Aged
18-64, Metropolitan Brazil, 1980


Variable Mean SD


Wage


Mulatto
Black

Experience


Schooling
1-4
5-8
9-11
12+

Region
Belem
Fortaleza
Recife
Salvador
Belo Horizonte
Rio de Janeiro
Curitiba
Porto Alegre

ocupation
Manager/Admin
Prof/Tech
Clerical

Hours

Social Security

Migratory Status

Marital Status


17776.752

.299
.066

21.011

588.018


23911.858

.458
.249

12.107

628.772


.426
.220
.153
.110


.021
.032
.053
.041
.072
.258
.040
.071


.085
.122
.135

.943

.938

.373

.662


.495
.414
.360
.313


.144
.177
.225
.199
.258
.438
.196
.256


.279
.327
.342

.232

.242

.484

.473


38711


Source: 1980 Brazilian Census .8 percent subsanple.


Table 4.2











The results of the fit of the earnings model to the data are

presented in Table 4.3. The sign and significance of the coefficients

correspond to theoretical expectations. For example, increases in

experience, education, and hours worked contribute positively to wages,

as does working in any other occupation than blue collar. Working in a

metropolitan region other than Sao Paulo has a negative impact on

amnthly wages.

Consider next the coefficients for race. The mulatto coefficient

(-.145) and the black coefficient (-.233) are both negative indicating

that, controlling for the eight wage-predictors, the earnings for

nonwhite workers are lower than those for whites. Being black results

in the lowest of the three populations. Both the "class over race" and

the human capital arguments reviewed in Chapters 1 and 2 imply that, in

the absence of discrimination, factors such as education and occupation

acconmt for all of the variance in earnings, and the race variable will

not be statistically significant. Yet, as shown in Table 4.3, this is

not the case. After introducing eight explanatory variables into the

equation, a worker's race remains significant in explaining wage

differences.

The absolute value of the race coefficients represents the

variation in wages that is unexplained after other job-related

attributes have been controlled. This value can be interpreted as an

indirect measure of the impact of wage discrimination on monthly

earnings of mulattoes and blacks (Boston 1988). If we accept this

interpretation, the findings show that discrimination is higher among

blacks otnpared to mulattoes. This finding suggests the relationship












Table 4.3


COuplete Earnings-Function with Race Dummy for Male
Workers Aged 18-64, Metropolitan Brazil, 1980


Variable Coefficient SE

Mulatto -.145** .007


Black


Experience
Experience2


Schooling
1-4
5-8
9-11
12+

Region
Belem
Fortaleza
Recife
Salvador
Belo Horiz
Rio de Jan
Curitiba
Porto Alegre

Occupation
Manager/Admin
Prof/Tech
Clerical

Hours
Social Sec

Migrant

Marital

Constant


.061**
-.0009**


.293**
.593**
1.023**
1.564**


-.498**
-.572**
-.492**
-.220**
-.177**
-.265**
-.320**
-.274**


.551**
.421**
.035**

.077**
.270**

.013*

.241**


7.611


.011

.001
.00002


.010
.012
.014
.015


.020
.016
.130
.015
.011
.007
.014
.011


.011
.010
.009

.012
.011

.006

.007

.022


R2 .593
N 38711


* p < .05.
**p < .001.
Source: 1980 Brazilian


Census .8 percent subsanple.











between earnings and the predictors of earnings differs by race in the

Brazilian labor market. The subsequent analysis will examine why race

continues to have an effect on wages net of individual attributes.


The Interaction Model

The second analytical step tests whether the equations for whites

and blacks and mulattoes differ significantly from one another. The

procedure for this test is the following: a model is fitted which

includes multiplicative terms for the variable race (mulatto and black)

with each independent variable. If the interaction terms are

statistically significant, it means that the slope of the relationship

between the mean wage and say, education, differs for each racial

group. For example, if the rate of change in wages as a function of

education varied by race, then lower wages among nonwhites could be

attributed to their lower wage returns to education. Interaction terms

empirically test for such a relationship.

The coefficients for the interaction terms and their standard

errors are presented in Table 4.4. The results show significant

interactions between race and experience, education, region, occupation

and background characteristics. The pattern differs between blacks and

mulattoes. More interactions by region are significant for mulattoes

while interactions by occupation are only significant for blacks. In

addition, the results from a general F test indicated that separate

equations should be estimated by race.3

We can conclude that the equations for blacks and mulattoes are

significantly different from the equation for whites. Specifically what











Table 4.4 Interaction Coefficients for Male Workers Aged 18-64,
Metropolitan Brazil, 1980


Variable Coefficient SE

MulXExp -.001* .0006
MUlXYsl -.051* .022
NMlXYs5 -.094** .025
PElXYs9 -.155** .031
kIlXYsl2 -.133** .039
MulXBelem .040 .045
MtlXFortaleza .071* .036
MelXRcife .045 .028
iulXSalvador -.070* .035
HulXBelo .056* .025
MulXRio .006 .016
hlXCouritiba .035 .041
MulXPorto .004 .039
MUtlXan/Adm -.044 .029
MulXProf/Tec -.027 .026
MulXClerical .006 .021
MulXHours -.060* .027
HulXSocial -.054* .024
MLilXMigrant .047** .014
uilXMarital -.130** .014
BlXExp -.003* .001
BlXYsl -.107* .035
BiXYs5 -.165** .043
B1XYs9 -.228** .055
81XYs12 -.413** .099
BlXBelem .107 .106
BIXFortaleza .153 .098
BlXRecife .046 .059
BlXSalvador -.047 .049
BlXBelo .052 .045
BlXRio -.006 .028
BlXCuritiba .100 .099
BlXPorto .123* .052
BLXMan/Adm -.300** .074
B1XProf/Tec -.082 .058
BlXClerical .006 .040
BlXHours -.130* .052
BlXSocial -.129* .042
BlXMigrant .067* .027
BlXMarital -.086** .026

N 38711

* Significant at .05 or less.
** Significant at .001 or less.
Source: 1980 Brazilian Census .8 percent subsarple.










this means is that increases in experience, education and hours worked,

employment in higher salaried ocupations, belonging to the social

security system, and being a migrant and married yield higher wage

returns to whites than to milattoes and blacks. Simply put, this means

that mulattoes and blacks earn less for performing the same jobs as

whites (i.e. in the same location and with the same qualifications).

The conclusion is consistent with a hypothesis of wage discrimination.

Still, the presence of discriminatory behavior is inferred rather than

measured directly.


Human Capital and Back)raund Characteristic Correlations With Earinrs


Table 4.5 presents the zero-order correlations between the basic

human capital and background variables and earnings for each racial

group. For experience, positive relationships are observed for each

group. Likewise the correlation between schooling and earnings is

positive for the equivalent of high school or better and generally

negative for less than or equal to junior high achievement. Within

racial categories, the zero-order correlation between earnings and the

highest dumy variables used to capture schooling, 9-11 years and 12+

years, is stronger than the relationship between lower levels of

education. Thus, overall, the schooling-earnings relationship is

linear. Moreover, the strongest association between predictors and

earnings, for all racial groups, is the relationship between education

and wages.

The correlations between region and wages are generally negative.

In the case of blacks, ccupared to the omitted region of Sao Paulo, a











Table 4.5


Zero-Order Correlations Between
and Background Variables


Earnings and Human Capital


Variables Black Mulatto White


Expience
Experience2


Sdcooling
1-4
5-8
9-11
12+

Region
Belem
Fortaleza
Recife
Salvador
Belo Horiz
Rio de Jan
Curitiba
Porto Alegre

Occupation
Mananger/Adm
Prof/Tec
Clerical

Hours

Social Sec

Migrant

Marital


Source: 1980 Brazilian Census .8 percent subsample.


.053
.017


-.065
.068
.175
.239


-.060
-.077
-.123
-.016
-.017
-.077
-.003
.026


.099
.245
.058

-.011

.147

.091

.277


.035
-.009


-.096
-.008
.148
.331


-.063
-.146
-.125
.036
.013
.014
-.024
-.019


.250
.264
.048

.022

.211

.122

.214


.050
-.005


-.293
-.107
.114
.535


-.020
-.060
-.054
.041
-.005
-.023
-.062
-.067


.372
.378
-.062

-.013

.166

-.021

.295











positive relationship between the southern-mst region of Porto Alegre

and earnings is observed. In contrast, mulattoes earn more when

employed in Salvador, Belo Horizonte, or Rio de Janeiro. For white

workers, wages are highest in Sao Paulo with the one exception of

Salvador.

Occupations other than blue collar positions, the reference

category, are positively related to wages with the negative exception

of white workers in clerical positions. Within occupation, the

strangest positive relationship for all three status groups is between

professional/technical jobs and earnings. The remaining correlations

between hours worked, social security coverage, and migrant and marital

status and earnings are generally positive, with marital status

relationships exceeding the other three background characteristics

within each racial group.


lbe EaZninMs-Fmction by Color



Having concluded that the wage equations differ by race, it is

necessary to estimate separate equations for the black, mulatto, and

white populations. 7he results of these separate equations are central

to the subsequent analysis of wage discrimination.

The means and standard deviations of the variables included in the

separate analysis by race are presented in Table 4.6. Consistent with

previous findings, an examination of race differences indicate that

whites earn more than nonwhites, and blacks have the lowest wages.











Table 4.6


Means and Standard Deviations (in parentheses) of Variables
Used in the Analysis of Earnings by Race, Metopolitan
Brazil, 1980


Variable Blacks Mulattoes Whites


Wage


Experience2

Schooling
1-4

5-8

9-11

12+

Region
Belem

Fortaleza

Recife

Salvador

Belo Horiz

Rio de Jan

Curitiba

Porto Ale

Occupation
Manager/Admin

Prof/Tech

Clerical


9003.777
(6820.906)
22.449
(12.053)
649.199
(656.245)


.528
(.499)
.213
(.409)
.085
(.279)
.017
(.128)

.012
(.108)
.014
(.116)
.044
(.206)
.106
(.308)
.092
(.289)
.411
(.492)
.012
(.109)
.057
(.233)

.022
(.147)
.046
(.210)
.093
(.291)


11053.300
(11087.286)
21.583
(11.789)
604.797
(625.193)


.505
(.500)
.224
(.417)
.100
(.301)
.037
(.188)

.050
(.219)
.074
(.262)
.101
(.301)
.082
(.275)
.092
(.290)
.252
(.434)
.018
(.134)
.020
(.139)

.043
(.204)
.064
(.245)
.106
(.307)


21867.155
(28149.807)
20.590
(12.237)
573.690
(626.956)


.379
(.485)
.219
(.413)
.185
(.388)
.154
(.361)

.008
(.091)
.015
(.120)
.032
(.175)
.015
(.123)
.060
(.238)
.245
(.430)
.054
(.225)
.096
(.295)

.111
(.315)
.157
(.364)
.154
(.361)











Table 4.6 (con't)


Variable Blacks Malattoes Wiites



Hours .952 .947 .940
(.213) (.223) (.237)
Social Sec .920 .920 .948
(.272) (.271) (.222)
Migrant .299 .399 .368
(.458) (.490) (.482)
Marital .638 .664 .663
(.481) (.472) (.473)

N 2574 11581 24556

Source: 1980 Brazilian Census .8 percent subsample.











Black males earn 41 percent of what white males earn; mulatto males

earn 51 percent of what white males earn; and black workers

earn 81 percent of what mulatto males earn. Ctnocening differences in

other individual-level variables, job experience is nearly identical

across race groups while level of education varies substantially.

Whites attain the highest number of years of schooling, blacks the

least. thus an the average, nrmdhites are 20 percent more likely than

whites to have only an elementary level education or less.

Turning now to the variables representing the metropolitan

regions, the findings show that the proportion of the population which

is white increases as you move south. Occupational status varies by

race, with whites clearly favored over nonwhites in the higher

skill/wage occupations. Race variation in hours worked and marital

status is minimal. There are racial differences in migratory status,

with mulattoes being the most likely to have migrated to their present

residence.

The estimate of wages as a function of human capital and job

related characteristics within each group is presented in Table 4.7.

Overall, the magnitude of the coefficients vary by race, the largest

coefficients being estimated for white workers. The racial difference

in the magnitude of the coefficients suggests that whites are able to

translate their investment in education and job experience into higher

wages than either blacks or mulattoes. This finding lends further

support to the conclusion of unequal returns to equal investment.

Regionally, all three groups receive the highest ages in Sao

Paulo, the reference category. However, the magnitude of negative











Complete Earnings-Functions by Race,
64, Metropolitan Brazil, 1980


Male Workers Aged 18-


Variable Blacks Mulattoes Whites


Experience

Experience2

Schooling
1-4

5-8

9-11

12+

Region
Belem

Fortaleza

Recife

Salvador

Belo Horiz

Rio de Jan

Curitiba

Porto Ale


Occupation
Manager/Admin

Prof/Tech

Clerical


Hours


social
Security


Table 4.7


.041**
(.004)
-.0007**
(-.00006)

.227**
(.027)
.460**
(.035)
.803**
(.047)
1.170**
(.085)

-.412**
(.087)
-.461**
(.082)
-.463**
(.048)
-.193**
(.355)
-.143**
(.037)
-.260**
(.024)
-.231**
(.086)
-.153**
(.044)


.254**
(.063)
.355
(.050)
.028
(.033)

.068
(.044)


.182**
(.034)


.055**
(.002)
-.0009**
(.00003)

.280**
(.015)
.549**
(.019)
.922**
(.025)
1.476**
(.033)

-.473**
(.024)
-.538**
(.021)
-.461**
(.019)
-.219**
(.020)
-.141**
(.019)
-.254**
(.014)
-.294**
(.037)
-.277**
(.036)


.513**
(.025)
.402**
(.022)
.042*
(.017)

.141**
(.022)


.254**
(.018)


.064**
(.001)
-.001**
(-.00002)

.327**
(.015)
.645**
(.017)
1.093**
(.019)
1.617**
(.020)

-.514**
(.038)
-.610**
(.029)
-.507**
(.020)
-.151**
(.029)
-.197**
(.015)
-.262**
(.009)
-.327**
(.016)
-.282**
(.013)


.551**
(.013)
.426**
(.012)
.039**
(.011)

.199**
(.015)


.305**
(.016)






Table 4.7 (oon't)


Variable Blacks Mulattoes Whites


Migratory
Status .069* .048** -.002
(.023) (.012) (.008)
Marital Status .234** .166** .281**
(.023) (.012) (.009)
Constant 7.460 7.642 7.880
(.029) (.038) (.074)
R2 .332 .417 .604

N 2574 11581 24556
* u p < .05.
** p < .001.
Note: Staxnard errors are embers in parentheses.
Source: 1980 Brazilian Census .8 percent subsaple.











returns by region is greater for whites than nonwhites. With regard to

occupational status, the first variation in patterns of statistical

significance is observed. All oc~aticral positions are significant

for mulattoes and whites. For blacks, only managerial/administrative

positions are significantly different than blue collar jobs. Such a

finding may reflect the tendency for black males to be segregated into

low paying positions regardless of credentials. This explanation will

be evaluated in Chapter 7.

The number of hours worked is also not a significant predictor of

wages for blacks though it is for whites and mulattoes. Blacks and

mulattoes who have migrated to their current residence receive higher

wages than their native counterparts. For whites, having migrated

translates into lower wages. There are similarities in the relationship

between social security coverage and marital status across groups.

Regardless of race, the effect of being protected by social security

and being married is to increase wages.

As far as the overall predictive success of the model is

concerned, the explained variance differs substantially across racial

groups. The model accounts for more of the variance in earnings for

whites than nonwhites. The magnitude of this difference is greatest

among blacks and whites where the included variables explain 33 percent

of the variance in wages for blacks, 60 percent for whites.


A Summary Measure of Discrimination

In comparative social research, investigators attempt to

disentangle the factors producing differences between groups in the










level of the dependent variable being studied. Use of decm ition

techniques for the study of wage inequality is a standard practice4.

The idea is to partition the observed wage differences between whites

and nonwhites into three co pnents: "unexplained" differences;

"ccupositian" (e.g. education) differences; and "joint effects" (Jones

and Kelley 1984). The most canm way to decompose the differences in

individual wages is to standardize the results of a least-squares

regression equation like the basic earnings function modeled in Table

4.7. The race-specific means and coefficients of each variable

resulting frcn the separate equations by race

(Tables 4.6 and 4.7) are then inserted into the following deronposition

model:


(yh yl)= [(ah al) + EX(bh B1)] + bl (Xh -X1) +

(a) (b)

E(b bl) (Xh- X1)

(c)

Gap = (a) unexplained differences + (b) position + (c) joint
effects

The notation throughout follows that of lams and Thornton (1975)

where:

h superscript indicates the high-wage group (always white males in
this study)

1 superscript indicates the low-wage group (alternatively, black and

mulatto males in this study) Ystands for the natural logarithm of

the wage rate











Xi is the mean of the ith explanatory variable

a is the regression constant; and

bi is the partial regression coefficient for the ith

explanatory variable

The first term, [(ah a1) + 1 (bh Bl)], is interpreted as that

part of the wage gap attributable to discrimination. This ornponent is

"unexplained" because it represents the residual that is left after all

relevant factors are controlled. Statistically, it is the amount due to

the difference between the intercept of the white's and nowhite's

equations, plus the differences in the coefficients standardized in

terms of the crmposition (means) of the low earning group. The

substantive interpretation of the estimate is how much of the income

gap results from differences in how the low group's endowments are

actually valued in the labor market and how they would be valued if

they enjoyed the same rates of return of the high income group. More

specifically, this term measures the amount of wages that blacks would

gain if they were paid the equivalent wages of equally qualified white

workers.

The second term, E~ (Xh X1), represents the amount of the gap

that is due to cnrpositional differences (such as different levels of

education or hours worked) valued at the discriminatory rates of return

under which the low earning group labors. It estimates the amount by

which blacks and mulattoes average income is depressed (oxnpared to

whites) because of a compositional deficit (e.g. lower levels of

education). Substantively, term (b) reflects how much more unohites










would earn if they had the same composition as whites but nothing else

changed.
The third term, E(bh bl) (Xh X1), is the amount due to the

joint effects of the differences in composition (means) and returns

(coefficients). lhis term is interpreted as the effect of jointly

changing both means and regression coefficients over the effects of

changing them one at a time (Iams and Thornton 1975:344). That is, it

reflects the amount blacks would gain if they had the same

characteristics of whites an received the same rates of return to

those characteristics. This term depends jointly on both differences in

composition and returns beyond their individual effects.

The ability to separate the effect of unequal ccrposition (term b)

from the effect of unequal returns to (term a) has obvious policy

implications. For example, it may be that blacks generally have less

education and work in lower paying jobs than do whites or mulattoes,

and that these disadvantaged d aracteristic contribute to the lower

income of blacks. If this were the case, the highest proportion of the

wage gap would be attributed to differences in composition (term b).

From a policy standpoint, the implication would be that the income

difference could be eliminated by "giving" blacks the same

characteristics as whites.

On the other hand, it may be that blacks receive less income for

doing the sane type and amount of work. That is, blacks may be deprived

because they are unable to convert their personal characteristics into

earnings at the same rate as whites. If the policy envisioned is to











increase the returns of blacks to match those of whites, then the

"discrimination" component (term a) would be examined.

7he goal of the following analysis is to decmnpose the wage gap

between whites and narwhites in order to examine which of the three

de Bnposition ocpnents explains the highest proportion of the wage

variance. In light of the interactions between race and the wage-

predictors dmstrated previously, the hypothesis is that nnwhites
suffer wage discrimination, and that discrimination is higher among

blacks ocqpared to mulattoes.

The findings presented in Table 4.8 sunmarize the results of the

deounr sition analysis. The proportion of the wage gap attributed to

discrimination (unexplained differences) is positive for both blacks

and nulattoes. Ihe findings show that 25 percent (Cz$2750.99) of the

wage difference between mulattoes and whites occurs because mulattoes

earn less for performing the same jobs as equally qualified whites.

Table 4.8 Dec~mposition of Average Incane Differentials by Race
Metropolitan Brazil, 1980 (Base Groupx*hite)


Ocmponent Mulatto Black
(1) (2) (3)
Total
Difference 10813.86 (100%) 1286.38 (100%)

Unexplained
Difference 2750.99 (25%) 4103.68 (32%)

Oamposition
Difference 3735.12 (35%) 2891.54 (22%)

Joint Effects 4327.73 (40%) 5868.16 (46%)

Source: 1980 Brazilian Census .8 percent subsanple.










Likewise, 32 percent (Cz$4103.68) of the wage gap between black and

white can be interpreted as that part of the wage differential due to

discrimination. Fram these results, it appears that compared to

mulattoes, blacks suffer more wage discrimination.

In ocaparing differences in composition (e.g. education and

occupation), 29 percent (Cz$3196.84) of the earnings gap for mulattoes,

and 16 percent (Cz$1968.61) for blacks can be attributed to deficits in

human capital and demographic factors. These results suggests that

racial differences in access to wage-related resources are more

important for explaining earnings differences among nulattoes than

blacks. The analysis of the acaposition cqncpment is of particular

interest given the "class over race" argument in the Brazilian

literature. These results demonstrate that unequal social standing does

explain a substantial portion of the wage gap between whites and

nonwhites. However, the most important conclusion is that the entire

wage gap, indeed not even one half of the gap, is explained by unequal

endowments.

In addition, a sizeable positive joint effects term for both

groups (46 percent for blacks, 40 percent for mulattoes) suggests that

jointly changing both wage-related resources and returns to those

resources would have an additional positive impact on earnings for both

blacks and mulattoes. Thus, we can estimate that the monetary

disadvantages suffered by backs and mulattoes due to wage

discrimination was US$50 for mulattoes and US$74 for blacks. This

amount was the "c-st of being nonwhite" in the 1980 Brazilian

metropolitan labor market.












Sunmary and Discussion

Several observations merge from the analysis of racial wage

differentials. The first is that there are substantial differences in

earnings between the races, even when controls for the variables

considered relevant in the process of wage attaiunent are introduced.

That is, differential wages are paid on the basis of skin color, not

qualifications. Second, the nanwhite category is not hamogenous, as

there are significant differences between the black and mulatto

population. This is particularly true when one examies the patterns of

returns to human capital and other variables. This finding has

important empirical implications. Specifically, to consider blacks and

nulattoes as comprising a homogeneous racial group of "nonwhites" is

unjustified, at least as far as analyses of the labor market is

concerned.

Third, the decomposition of the wage gap showed that a substantial

proportion of the income differences can be attributed to

discriminatory factors operating in the labor market. The data

provided suggest that policies oriented to increase factors such as

education among nonwhites will not, in and of themselves, eliminate the

wage gap. Rather, the prospects for racial equality in wages in Brazil

seem to be contingent on two factors, equal pay for equal work plus

increased access to specific resources such as education and higher

paying occupations.













1. Because these data are highly right-skewed, the log of monthly
earnings is used in the OLS regression models, a practice that is
standard in research using human capital earnings functions (Becker
1975; Becker and Chiswick 1966; Hauser 1980; Pugel 1980; Qhiswick 1983;
Willis 1986). There are two primary benefits of using the log form of
earnings, according to Hodson (1985:376): 1) "The functional form has
the benefit of 'pulling in' large positive outliers that might
otherwise bias the regression results because of hetero ticity in
error variance," and 2) "the regression coefficient's produced by an
analysis of log earnings can be roughly interpreted as percentage
returns or rates of return, to the earning determinant under
consideration "
2. The effect of experience an earnings is such that one earns more
with experience, but each additional year of experience nets less gain
than the previous one. Hence, the effect of experience is a declining
positive effect which when plotted takes on the shape of a parabolic
curve. 2he standard statistical correction far this is to square the
experience term.
3. At dfl 42 and df2 38645 the F-statistic of 7.70 was significant at
p<.005 indicating a rejection of the mll hypothesis that a single
population plane far whites, blacks, and mulattoes combined is the true
plane (Afifi and Clark 1984:150).
4. See Duncan 1969; Althauser and Wigler 1972; Oaxaca 1973: Blinder
1973; lams and Thornton 1975; Mincer and Polachek 1974; orcoran and
Duncan 1979; and Jones and elley 1984 for a review.
5. This decmpostion model was chosen for both substantive and
methodological reasons. First, the model identifies three mrponents
that are most appropirate for the questions addressed in this study.
Second, following the recommendation of Jones and Kelley (1984) and
Winsborough and Dickinson (1971), the "discrimination" component in
this model includes only intercepts and slopes. A parallel model
adopted by Blinder (1973), instead, adds the "joint effects" term to
"discrimination." This procedure, in practice, inflates that proportion
of the wage-gap attributed to "discrimination." Another model used by
Oaxaca (1973), adds the "joint effect" term to the composition
component, thus inflating that proportion of the wage-gap explained by
differential endowments. The Jones and Kelley model, therefore, is the
most conservative measure of both composition and discrimination.















CHAPrER 5
EARNINGS AND REGIONAL INEMUALITY


Since slavery, Brazil's population and economy have been

characterized by sharp regional inequalities. Population redistribution

and a series of economic boon and bust cycles are the domestic

nsequees of Brazil's insertion into the international market.

Respading to increased demand in Western Europe during the sixteenth

and seventeenth centuries, Brazilian suar production rcnoentrated

wealth and population in the Northeast. In the eighteenth century, with

the discovery of gold in Minas Gerais and the oncuomitant decline in

sugar, the eoonmaic center of the country shifted to central and

southern Brazil. The late nineteenth and early twentieth centuries saw

the expansion of coffee exports and significant international migration

that led to the incipient industrialization of Sao Paulo. Building on

it's initial advantages, Sao Paulo has remained the leading economic

center of the country, while the Northeast suffers from severe

droughts, extremely high mortality rates, and prolonged economic

stagnation.

The geographic polarization of the economy and population is said

to be "one of the basic determinants of contemporary racial inequality

in Brazil" (Hasenbalg 1985:27). Of the total population in 1980, 75.5

percent of the nonwhite population lived in the Northeast, while the










white population was concentrated in the South (Hasenbalg 1985).

Equivalent estimates for the male workforce showed similar

inequalities: blacks and mulattoes are more than twice as likely as

whites to reside in the Northeast (Table 3.5).

It is reasonable to expect that unequal eoomic opportunities

associated with regional disparities have a strong effect on the

distribution of wages. Given that the dynamic sectors of the economy

are concentrated in the South and the majority of the nonwhite

population in concentrated in the North, wage differences among the

racial groups can undoubtedly be accounted for in part by this

geographical polarization. This finding is supported in the preceding

Chapter. Earnings equations estimated separately by race (Table 4.6)

included controls for each of the metropolitan regions. As expected,

wages differed significantly by geographic location. Workers receive

the highest wages in the South, the lowest in the North. While we can

conclude that regional wage inequality exists, we do not know the

magnitude of wage discrimination internal to each region. Hence, the

following analysis of wage inequality is disaggregated by metropolitan

region. he central question addressed in this Chapter is whether wage

discrimination is positively or negatively correlated with the level of

regional development.


Beional Comparisons

The review of the historical pattern of unequal racial and

economic distribution by region suggests two testable hypotheses:











(1) Based an the effects of geographical segregation of whites and

nowhites, we can speculate that a more racist envirorent exists in

the South, indeed sme research suggests this to be the case (Cardoso

and lani 1960; Fernandes 1971). One way to operationalize this

hypothesis is to focs on the racial distribution of employers.

Specifically, one can posit a positive relationship between the

proportion of white employers and wage discrimination. hus, if we

assume that there is a higher percentage of white employers in the

South, we can expect discrimination to be highest in the southern

regions. (2) The limited resources and economic opportunities available

to nonwhites in the North have resulted in lower levels of human

capital here. Hence, because a larger proportion of the wage

differential will be captured by the czmposition Ipoe nt, we can

expect differences in composition to explain a larger percentage of the

wage gap in the North than South. The implication of this second

hypothesis, like that of the first, is that discrimination will be

higher in the South.


Refornulation of the Earnings Function

Following the same procedures used to analyze Brazil as a whole,

the nine metropolitan regions are examined in three stages. First,

earnings functions are estimated for each region. However, the regional

analysis required sane simplifications in the basic earnings model

specified in Chapter 4. In particular, schooling and occupation will be

treated as dichotmous variables. This is necessary because, for an

inter-regional analysis restricted to workers only, the number of











blacks in the upper schooling and occupational categories is too small

to obtain stable estimates. thus schooling is measured in terms of the

equivalent of elementary education or less (coded 0) versus greater

than elementary schooling (coded 1). This gross distinction is

legitimate given the fact that the mean level of schooling in Brazil is

a little over 4 years (Lam and Imvison 1987). Occupation is

dichotomized as either blue collar (coded 0) or white collar (coded 1).

In stage two, the models test whether the equations for whites and

nonwhites are significantly different from one another by including a

dummy variable for race (white-0) plus interaction terms for the race

variable with all other independent variables. Ihe interaction models

showed that, for all regions, interactions between race and several

independent variables were significant (findings not shown). In

addition, results of general F-tests (Table 5.3) rejected the null

hypothesis that a pooled model should be fitted. In step three,

therefore, models for whites, blacks, and mulattoes are estimated

separately. Finally, the earnings differences between the three groups

are decomposed into unexplained, composition, and joint effects

cWmponents.


The Earnings Function by Region

Table 5.1 presents sample means and standard deviations for the

independent variables by region. As confirmed in the previous chapter,

there are clear regional differences in the average monthly wage.

Earnings are highest in Sao Paulo (Cz$19883.60), the most




























-V


a a


" p-


WS





* -





N s
- I

,,-


IcJ
*


A
N I
Sff





as



4 4


N

o -
1
t^
83


*


Ia


Vol v
- N5
> *


5JS
*


95
N








a4
* *


-V


:


LI


vi
e;


U% I


0 M
4 4















Ie
S*
* *


wo

S ?


in *


gN


-V


NI-,
O N
N -


Ci


Nc -
N 4


m
-
j


sR


ea


in h










industrialized region, and lowest in Fortaleza (Cz$10798.55), a

traditional northeastern area. he one exception to the north/south

polarization is Salvador, the capital of the northeastern state of

Bahia, where the mean wage approimates that of Belo Horizonte and Rio

de Janeiro.

An examination of the racial distribution (columns 2 and 3)

indicates that the highest proportion of mulattoes reside in Belem

(71%) and Fortaleza (69%), the highest proportion of blacks in Salvador

(17%), Belo Horizonte (20%), and Rio de Janeiro (22%). However, whites

cxnprise the majority as one moves south from Belo Horizonte to Porto

Alegre. Labor farce experience varies little across region. Schooling

levels show that nearly one half of the male working population has an

elementary education or less.1

Occupation varies little across the nine regions. Roughly 70

percent of all working men are employed in blue collar jobs. Over 90

percent of the population in each region works full time with slightly

higher proportions employed 40 hours or more per week in the Southern

regions. Similarly, the majority of workers receive social security

benefits, with the lowest percent covered in the North, the highest in

the South. The proportion of migrants by region does vary considerably.

In Rio de Janeiro, 42 percent of the male work force are migrants,

while 52 percent of the workers in Sao Paulo were born outside of the

state they currently reside in. Again, there are few differences by

region in terms of marital status. Over 60 percent of all male

employees are married in each metropolitan region.










Table 5.2 presents the results of the regional earnings model with

the race dummy included. DEamining the coefficients for race (columns 1

and 2) across all regions, tw observations are relevant: the sign of

the coefficients is negative for both blacks and mulattoes; and the

race variables are statistically significant. The results indicate

that, controlling for the eight background variables, wages for

nmwhite workers are significantly lower than for whites. This result

along with the significant interaction terms (unreported) and general F

tests (Table 5.3) verify that the structure of the relationship between

earnings and the predictors of earnings differ among whites, blacks,

and mulattoes across all metropolitan regions.

For all regions, increases in experience and education, working

outside of blue collar jobs, receiving social security coverage, and

being married are positively associated with earnings. ien hours

worked are examined we find the first variation in patterns of

statistical significance. Number of hours worked per week is not

significant for the metropolitan region of Sao Paulo. Such a finding

might reflect the tendency of wages to be higher here, in comparison to

other regions, regardless of hours worked. Sao Paulo was also unique in

terms of the measure of migration. Being a migrant in Sao Paulo is,

unlike all other regions, associated with a decrease in wages.

In regard to the overall predictive success of the model, the

explained variance (column 11) differs only slightly by region. The

model works best for the metropolitan region of Recife (47 percent of

the variance is explained). The least amount of variance (42 percent)

is explained in Porto Alegre.












-~ -
R ag~Nr -


N .


N- N-


Lo j. i


j~a
Y

r rm
H
,a %a
Y


S
I11


* S
ff


SIA
*


o


,e 0
Is %


rs
fNC
N 0


5%-
N
-c N


Si
* IA


t d


6 S










The Earninas Function by Race and Region

Having determined the slopes of the regression equations differ by

race within each of the nine regions, separate equations are estimated

for the black, mulatto, and white populations. The means and standard

deviations of the variables included in the analysis are presented in

Table 5.4. An examination of the race differences (columns 1 and 2) in

inocne shows a clear pattern. Blacks and mulattoes consistently earn

the lowest wages-all things being equal. Wages are the lowest for

blacks in Recife (Cz$6360.22) and for mulattoes in Fortaleza

(Cz$8550.60). The highest wages for both blacks and mulattoes were

found in Sao Paulo (Cz$10216.16 and Cz$11992.58 respectively), whites

earn the most in Salvador (Cz$31199.44). With the exception of

Curitiba, blacks have lower earnings than mulattoes.


Table 5.3 General F Tests for the Null Hypothesis that a Single
Population Model for Blacks, Whites, and Nulattoes is the True Model.


Region F-Stat dfl df2 Sig
(1) (2) (3) (4) (5)
Belem 4.956 16 2877 <.005
Fortaleza 4.152 16 2803 <.005
Recife 9.548 16 3133 <.005
Salvador 4.765 16 3221 <.005
Belo Horiz 8.627 16 3911 <.005
Rio De Jan 9.394 16 3986 <.005
Sao Paulo 8.228 16 4555 <.005
Curitiba 3.584 16 3817 <.005
Porto Alegre 4.316 16 4386 <.005

Source: 1980 Brazilian Census .8 percent subsample.


Schooling levels vary considerably by race and region. The pattern

within each region is consistent with the overall finding of Chapter 4,

white male workers have higher levels of education than nonwhites,

























S 6 Ri S-i
2:4
a V% tv
OLu


A AS A





m NO & Np
* u %U
rs r^ ^


S N

iSS


0. a



S4s ISF
* N -
i^ i^r


o-
wl&


N1
- v)

N


N 4


SN
2 E


N f
N N



v N


N"


- N
I'.


f> *


4 4




II


-j
I




5












ae
I-











06
UJ


O s


't %


,s ,, ~sS


"" H


u r
A


8^


5


c.
~~ WJ
Y


;a5; sC J
Y
















.4 .
N


ON *;W~
-E St St


-~ St S


* N


N N
N -#


'0.4





v~f


oc


LA


h
0i;0
0.4


*S N
ON 0




Nlr 0 c
N. .4


4D
- N





























&U
8i
-n





Li
>-





0.4^





u N


SM
In


ry


%P St as


S S


P t


v

N M



S00
-* o






0-


M


O
'0p.S
ONU


^
a


*0 N
A


0H0 S


S0'
I.e


40
I's .


1 ;
u.S


Y N
-0 *


S
00
r^


8s


A


S
-


HP ~fi
r ~J
v v


v


N
N
e~ 3N
u u


ffff IF~

Eid ~
















N 5 N
0 0


SC
'C'


N~ 4
-p -









V Y
Ii'C4 N' '


* *N4











*"^


'ONs


,- ,w
-p -V


aI


9 N
4 4


Id


-
-V?


s ~ g
- s -p


U"
4


' S
N 'C
'C


S Cv I


s.


4 0 4- O0






"s s a ss
rus *C


N N 1


a9 a


-V


in
AI
ISH S
rfii p.T


sI si
;i a


4 N


I s. ss t
N M 4
NN -V


0 U
CY
4-
S
a-
0


? s


a
i N

N


04


. a
x N




IA


-
W


0 -


IS S
*o 5


iSB


8
iP


s ?










mulattoes higher than blacks. Regional differences show that the

highest proportion of blacks with better than an elementary level

education reside in the metropolitan areas of Rio de Janeiro (35

percent) and Porto Alegre (38 percent), the highest proportion of

mulattoes in Salvador (46 percent) and Balem (50 percent). The largest

proportion of white workers (74 percent) with better than an elementary

education is found in Salvador, the lowest in Sao Paulo (52 percent).

These findings indicate that the highest educated black work force

resides in the South and the highest educated mulatto work force

resides in the North.

Ocaupational differences by race and region are also striking, yet

consistent with previous findings. White males are much more likely

than nonwhites to be employed in white collar occupations in each

region. Internal to the norwhite category, mulattoes in the northern

regions are often twice as likely as blacks to hold white collar jobs.

This is not true in the South, where from Sao Paulo to Porto Alegre

blacks are generally as likely as mulattoes to work in white collar

jobs. The general conclusion is that, compared to other regions, the

South is actually a more favorable location than the North for blacks

who possess the skills and education to obtain white collar jobs.

Again, hours worked and social security status vary little by race

or region. Migratory status does differ substantially by race and

region. Mulattoes are the most likely to have migrated to their current

residence and the industrialized South is their destination. Marital

status differs only slightly by race and region with slightly more

whites than nonwhites being married.










Turning now to the estimates of wages as a function of human

capital and wage predictors in Table 5.5, the findings shw that there

are significant racial and regional differences in the wage related job

characteristics. Job experience has a positive effect on wages for all

races and regions, although it is not always a statistically

significant indicator for blacks. Such a finding may reflect the

tendency for black workers to be segregated into entry-level positions

in which there is minimal chance for advancement based an the

acmmlation of job-relevant skills. It may be that aulattoes and

whites are more likely to ocuipy positions where experience is directly

rewarded in terms of praations aoqpanied by pay increments. This

explanation suggests that there may also be industrial differences in

access to positions for which experience is rewarded. This possibility

will be evaluated in Chapter 6.

Increases in schooling are positively associated with wages for

all three group across the nine regions. Again, like job experience,

schooling is not always statistically significant for blacks and may

reflect the tendency of blacks to occupy positions where advanced

education is not rewarded by wage increments. The magnitude of the

coefficients vary considerably by race and region. The education

coefficients are generally twice as high for whites as blacks,

mulattoes oocipying a middle ground. In Salvador, for example, the

schooling coefficients are .376, .493, and .617 for black, mulattoes,

and whites respectively. The equivalent estimates in Sao Paulo are 070,

.381, and .492. Such findings suggest that returns to human capital

inputs differ by region and race. Wage returns to education for

nonwhites, in particular, are consistently lower than for whites.

















, R


- N 0




s S
*


s'0-



:4-
vf C


Sii



9rN


a a a
00*


A
* N
- N


*0 0
N
-


$K


*0


2e a


-i


4- 0




s I I


N PIB
ES E* I


'0 u


-V








S


~N o
m -


sp

0-'


hs
p. -'


a -


a


t i


Ul -





IL






N


IL
*N
y3





wu
un -f


a


"8


s~ if
























Pd 1_
Ou "


Pd C






0-















N

























P
Kw







OK


NL


i
,* .*


* S
Pd


fly
i *

* 4


g 4
* *


SO


*0
*
5 s


i% Be 5
~I~* r


Ad


let


Pd -
N -


N
r^ ^


v

5 -
P -





t ?


.O ni


a
',









t



Ls









, C






*^
- 5
-


ON
- C
0 I-
v




P C




* S
I C











IL
*


t0 0


*~
Pd~
















0 ~
-? N (


N
0> w


11%
s?
1^*


I



-
z



I. -








LI-



b 6


3t
N
V; N


r4


-- C
C-u


5N
'*0*


1!


-c 6
- S v S


NC *'*


So C e P i P


9~
p. p


N








a


S CS


C.
*O -*


* -
*o


C
I-.


- C


- -


!LS
go
* -


I


N



Bf
wu


,- E



1s a
9


*


9 C!
N 0
o00


C N
99


v v
C a

*


P 9
8 5










Looking at occupation, similar patterns hold. The occupation

coefficients are positive for whites and nonwhites, yet, in some cases

insignificant for blacks. Again, the interpretation is that higher

skill/status occupations do not have the same impact on wages for

blacks as they do for mulattoes and whites. The coefficients for whites

in white collar jobs generally exceed those for nonwhites. Regionally,

the effect of occupation is highest for blacks in Curitiba in the Soth

(.561) and in Recife (.412) in the North, for aulattoes the greatest

returns to occupation are in the northern metropolitan regions. Returns

to whites for occupation are the lowest in Sao Paulo (.492) which

oorrespxnds to the finding that of all regions, whites in Sao Paulo

have the lowest mean education.

Consistent with previous findings, working full time has a

positive effect on incne across regions and races as does access to

the social security system. However, in many regions these two

variables are not significant predictors of wages for the black

poxplation, indicating that blacks are rewarded differentailly for

these characteristics. The same pattern holds for migratory status.

Being married, on the other hand, is always a positive predictor of

wages in all regions and for all three populations.

As far as the overall predictive success of the models by region

and race, the explained variance differs substantially. In all cases,

the model predicts more of the variance in wages for whites than

nonwhites, and the least for blacks internal to the nonwhite category.

Regionally, the model explains the most variance in wages for blacks in

Belo Horizonte (31 percent), Recife (27 percent), and Curitiba (27











percent). For nulattoes, the model explains the greatest amount of

variance in Fortaleza (38 percent), Belem (36 percent), and Salvador

(36 percent). For whites, the greatest amount of variance is captured

in the regions of Recife (51 percent), Salvador (51 percent), and Belo

Horizonte (48 percent). The two general conclusions for this analysis

are: nonwhites receive lower returns than whites to human capital

irpats and demographic characteristics; and, returns vary by region.


IDecouosition of the Waoe-Gap

The findings presented in Table 5.6 summarize the results of the

decaxposition of the racial wage differential. At first glance we see

in column 1 that the wage gap between white and black is generally

greater than that between white and mulatto. Regionally, the smallest

white/nowhite wage differences are in the northern region of Fortaleza

(Cz$7855.67 for mulattoes, Cz$7972.77 for blacks) and the two

southernmost regions of Curitiba and Porto Alegre. The largest wage

differential between whites and nanwhites is in Salvador (Cz$17404.75

for mulattoes, Cz$21531.79 for blacks).

In terms of the amount of the wage gap attributable to

discrimination, the hypotheses predicted higher discrimination in the

South for two reasons. First, because of limited access to education

and economic opportunities, nonwhites in the North would be

disadvantaged, relative to the South, in human capital inputs. As a

result, that proportion of the wage-gap attributed to composition is

expected to be highest in the North. The discrimination coampnent, by

default, would be higher in the South. Second, discrimination was