View
0
Download
0
Category
Preview:
Citation preview
The Distribution of Income in Brazil
SWP356
World Bank Staff Working Paper No. 356
September 1979
lhd by: Guy Pierre PfeffermannCountry Programs Department IILatin America and the Caribbean Regional Office
X Richard WebbDevelopment Economics Department
|___ _ Development Policy Staff
Copyright © 1979The World Bank1818 H Street, N.W. !
Washington, D.C. 20433, U.S.A.
The views and interpretations in this document are those of the authors;and should not be attributed to the World Bank, to its affiliatedorganizations, or to any individual acting in their behalf.
Pub
lic D
iscl
osur
e A
utho
rized
Pub
lic D
iscl
osur
e A
utho
rized
Pub
lic D
iscl
osur
e A
utho
rized
Pub
lic D
iscl
osur
e A
utho
rized
The views and interpretations in this document are those of the authorsand should not be attributed to the World Bank, to its affiliatedorganizations, or to any individual acting in their behalf.
WORLD BANK
Working Paper No. 356
September 1979
THE DISTRIBUTION OF INCOME IN BRAZIL
This paper asks whether growth has been widely shared and,*in particular, whether it has raised the absolute incomes of thepoorest groups in Brazil. The review of evidence covers the period1960-1976/77. New sources, in particular a national household budgetsurvey (ENDEF) and the 1976 household employment survey,have madepossible both a re-examination of previous sources and a clearerseparation in the analysis of income trends before and during thefast growth period between 1968 and 1974.
There is no set of comparable and accurate sources that pro-vide a definite answer to questions regarding income trends since 1960.There is, however, a great deal of fragmentary and imperfect evidencewhich, when examined as a whole, suggests that (i) the extent of incomegrowth throughout the population, including many in the poorest groups,has been underestimated in previous studies, and (ii) the rate ofpoverty reduction was faster during the rapid growth years 1968-1974.
Prepared by: Guy Pierre PfeffermannCountry Programs Department IILatin America and the Caribbean Regional Office
Richard WebbDevelopment Economics DepartmentDevelopment Policy Staff
JNTEMNATIONAL IMOlETARY FUND
Copyright © 1979 JOINT LIBRARY
The World Bank1818 H Street, N.W. [TE 2 '93Washington, D.C. 20433, U.S.A.
1-iTERhATIOPHAL BANi' rORrLEOlTF.-.ClCIOi LN:D-, LC', fOAENT
TABLE OF CONTENTS
Page No.
I. Introduction . . . . . . . . . . . . . . . . . . . . . 1
II. Overall Context . . . . . . . . . . . . . . . . . . . . 2
III. Trends in Inequality: Census and Survey Data . . . .. 7
1. The Available Data: A Summary . . . . . . . . . . 8
2. ENDEF .... . . . . . ..... . . . . . . . . 13
3. Underreporting .... . . . ..... . . . . . . 15
4. The 1960 Census .... . . . .. . . . . . . . . 24
5. Other Years ............. -.------------ 29
6. Censuses and Surveys: Conclusions . . .. . . . . 32
IV. Trends in Inequality: Partial Data . . .. . . .. . . 38
1. Education and Income Differentials . . .. . . . . 38
2. Urban Modern Sector Differentials: Law of 2/3 49
3. Urban Modern Sector Differentials: Wage Surveys 59
4. Regional and Sectoral Differentials . . . .. . . . 68
5. Partial Data: Conclusions . . . . . .. . . . . . 77
V. Trends in Employment and Incomes . . . . . . . . . . . 79
1. The Structure of Employment . . . . . . . . . . . 80
2. The Structure of Incomes . . . . . . . . . . . . 87
3. Income Trends .... . . . . . . . . . . . . . . . 92
VT. The Extent of Poverty in 1974/75 . . . . . 99
VII. Appendices . . . . . . . . . . . . . 103
Internal Terms of Trade . . . . . . . . . . . . . . . 103
Wage Statistics . 107
LIST OF TABLES
Page No.
1. Estimated Family Income Shares . . . . . . . . . . . 10
2. Comparison of Estimates of Annual Personal Incomeand Consumption . . . . . . . . . . . . . . . . . . . 16
3. Nonmonetary as a Proportion of Monetary Family Income 21
4. Educational and Occupational Income Differentials1960-1976 . . . . . . . . . . . . . . . . . . . . . . 41
5. Ske-wness in the Upper Tail of Income Distributions:Ratios Between Mean Income in Open-Ended Class andLower Limit Income-, for Alternative Percentiles . . . 43
6. Population According to Grade Level . . . . . . . . . 48
7. Coverage of Law of 2/3 and 1959 Economic Census . . . 53
8. Trends in Modern Sector Wage Differentials:Selected Data . . . . . . . . . . . . . . . . . . . . 60
9. Trends in Regional and Sectoral Income Differentials 69
10. Life Expectancy by Regions (1950, 1970) . . . . . . . 74
11. The Structure of Employment 1960-1976 . . . . . . . . 81
12. Ranking of Occupational and Regional Groups byIncome Level (Household Heads and Household Income)1974/75 . . . . . . . . . . . . . . . . . . . . . . . 89
13. Farm Household Incomes and Regional AgriculturalProductivity . . . . . . . . . . . . . . . . . . . 91
14. Income and Wage Trends: Indices . . . . . . . . . . 94
15. Comparison of Deflators . . . . . . . . . . . . . . 96
16. Proportion of Households Owning Durables . . . . . . 98
17. The Extent of Poverty: Families Under Two MinimumWages: 1974-1975 . . . . . ... . . . . . . . . . . . 100
Figure 1. Brazil - Percent of Average Life Expectancy(1950, 1970) . . . . . . . . . . . . . . . . . 75
THE DISTRIBUTION OF INCOME
IN BRAZIL
Guy Pfeffermannand
Richard Webb
I. INTRODUCTION
The procedure we follow is to first examine data on trends in
the distribution of income. Since per capita income '.n Brazil doubled
between 1960 and 1976, substantial income growth for all groups will have
resulted unless it was offset by changes in distributive shares. The
analysis of changes in shares thus provides an ir:direct answer to the
question of income trends. The evidence on trends -r. distribution consists
of aggregate data from censuses and surveys, and of partial data on income
differentials. The review of that information is then supplemented by a look
at direct evidence of change in emplcyment incomes occurring both through
the shift to better jobs and through increases in the incomes received in
specific types of employment.
-2-
II. THE OVERALL CONTEXT
Before entering into a discussion of income statistics, it will be
helpful to draw attention to certain features of the economy. The growth and
concentration of income can be expected to bear some relationship first to
the growth and concentration of output, and second, to the institutional
arrangements that affect the division of that output. Though both
sets of factors contribute a great deal to explaining the income distri-
bution, it is easier to be quantitative about output than about
social and political behavior. Also, in the case of both output and
institutions, it is necessary to know something about the situation at the
starting point in 1960, and about how they have evolved since then. Again,
it is easier to provide numbers on trends in sectoral and regional output
and on the location of the population vis a vis that output, than
it is, for example, to quantify changes in share-cropping arrangements,
in the enforcement of minimum wage legislation, or in the extent to which
local monopoly powers of landlords, lenders and wholesalers are being
broken down by better roads and communications.
Since the issue of income distribution in Brazil is most often
posed as a case of "rapid growth with growing concentration" it is useful to
establish some facts on growth. These are first that growth rates have
been very variable, ranging from 3 consecutive years (1963-65) when per
capita growth was negative to 3 consecutive years (1971-73) when it sur-
passed 8.5% per year. At the same time this variability in real growth
was accompanied by high and also variable rates of inflation ranging from
91% in 1964 to 16% in 1972. Second, annual per capita growth in GDP was
relatively slow for over half the intercensal period; between 1962 and
1967 it averaged only 0.7%; for the intercensal period as a whole it
averaged about 3Z. The intercensal decade (1960-1970) therefore, is not
a convenient period for posing questions about distribution during rapid
growth. Starting in 1968, per capita output grew by close to 8% per year
until 1974 and moderated to about 3% since then. Much of the debate about
distribution in Brazil took place in the early seventies and was based on
data for the sixties, particularly the censuses. The present is therefore
an opportune moment for reviewing the growth-distribution relationship
under the very different growth contexts of pre-1968 and post-1968.
Questions should also be raised about the impact of so much short-term price
and output variability on relative incomes and the responses of employers and
workers. Thus, the short-term behavior of incomes is likely to be.highly
variable, adding to the difficulty of discerning long-term trends.
A third feature of the growth record is the comparatively good
performance of agriculture which grew at 4.6% per year between. 1960-1977,
almost twice the rate for the rest of Latin America. Non-agricultural output grew
faster, but the substantial growth in agriculture satisfied a major necessary
condition for improvement in rural incomes. Furthermore, with the help of massive
rural-urban migration, which reduced the increase in farm families to about 0.6%1/
per year, and of favorable terms of trade, these output gains were trans-
lated into a growth in real agricultural value added per household of almost
4.0% per year. This poses the question of how productivity gains were dis-
tributed within agriculture.
It is difficult to evaluate income distribution in Brazil without
reference to the considerable degree of economic and social diversity. Some
1/ See Appendix on Internal Terms of Trade.
-4-
of this can be traced to the different origins of immigrants, but the most
powerful factors are probably ecological differences, and the speed and
uneven spread of technological change. A frequent but misleading expression
of this diversity is the description of Brazil as a dual economy, or a
1/"Belindia", stressing the contrast between modern and traditional technologies
and regions. This expression misleads in that structural differences are
important within the modern and traditional categories.
Thus the hourly wage for a bricklayer's assistant in Sao Paulo is twice that
in much of the North-East (Cr.$7.00 vs. 3.28); the rate of illiteracy amongst
male farm laborers in Sao Paulo 1973 was 25%, vs. 58% in the North-East.
There are powerful regional income differences within the North-East: in
1974/75 an index of average family income was 100 in rural areas, 200 in non-
metropolitan urban areas, 300 in Fortaleza and 590 in Salvador. Furthermore,
there are major income differences within rural society associated with ecology,
land tenure, and with a variety of contractual traditions in land, labor and
product markets.
The failure to notice this diversity underlies many oversimplifica-
tions about income distribution in Brazil. The population is frequently reduced
to a few categories that fit theories or paradigms, such as industrial labor,
the Northeast landless, and senior executives. But factory workers and the
Northeast landless together account for only 12% of the Brazilian labor
force, while the total income of senior managers is also a relatively
small proportion of total personal income. The great bulk of
income and employment is left out in discussions that center on those
categories. Within the Northeast, for example, there are as many owners and
self-employed in nonfarm businesses as there are landless farm families; the former
are growing much faster in number, and their average income is over three times
1/ From a juxtaposition of Belgium and India.
- 5 -
that of the landless. There are almost as many domestic servants as factory
workers. In 1974, family employment on small and medium farms in the South
region alone-(States of Parana, Santa Catarina, and Rio Grande do Sul)
was as large as that in all factories with 50 or more workers in Brazil, and
total income received by each group was similar. In 1976 there were twice as
many workers in Government and welfare services as in factories. Market
and institutional forces affecting employment and incomes in these other,
generally neglected components of the economy are thus, by far, the biggest
part of the story.
Finally, one cannot examine income distribution in Brazil without
noting the highly dynamic and mobile character of the population and its
economy. In such an environment, the traditional attention given to cross-
section or static measures of inequality needs to be complemented with measures
that reveal the income experience of individuals or groups over time. This
mobility has several aspects, the most familiar being the high rate of net
rural-urban migration. But less familiar is the extent of movement within
rural and urban areas, and the large urban-rural return migration. One
measure of all this locational change is the 1970 Census statistic that 25%
of all families had had less than two years' residence in their present home.
Another is the extent of labor turnover: in 1970, 60% of all manufacturing
employees in Sao Paulo State (the sample covers mostly medium and large firms)
had been employed less than two years in their current job. The extent of
turnover is related to the youth of the labor force: 1 in 3 workers were
less than 20 years old. A further aspect is that Brazil is still a frontier country:
cultivated acreage expanded by 35% between 1960 and 1975. The urban counter-
part is that around 1969-1970 about 15,000 manufacturing establishments of
-6-
1/5 or more workers were being created annually. So much movement leads one
to expect a great deal of individual income change with both upward and
downward mobility. Extensive movement could also have high pril_te costs,
particularly to groups such as casual farm and construction workers who move
frequently, and it may hamper the development of local and grassroots
organization and thus the capacity for communal self-help.
1/ Censo Industrial 1970, p. 270. The net increase in establishments of 5+workers between 1960 and 1970 was only 28,000, implying considerable turn-over of establishments.
-7-
III. TRENDS IN INEQUALITY: CENSUS AND SURVEY DATA
It is useful to begin the analysis by breaking down the issue
of overall inequality into two questions: First, what is the share of
the poorest 40%; and second, how is the remainder being distributed
between middle and top income groups. This distinction is helpful
because the questions have unequal priority, and because the causes
underlying changes in shares, and the corresponding policy implications,1/
are likely to differ in each case. From the point of view of welfare,
the most important aspect of the aggregate distribution is the share2/
accruing to the very poor. The extent of absolute poverty is the product
of that share and of the general level of development. The 3 to 4%
growth in per capita GDP since 1960 implies a substantial reduction in
the number of absolutely poor families and/or, in the size of the
21 This distinction is lost when a single measure of inequality, suchas the Gini coefficient, is used.
.] The choice of the bottom 40% as "the poor" is, of course, arbitrary.However, it is probably close to the proportion of families fallingbelow the poverty line in 1960, using the two August 1974 minimumwage poverty line that is the basis for the poverty estimates inTable 6. Also, the statistical error in estimating income shares fora smaller poverty group, e.g., bottom 20%, would have been even greater.
- 8 -
poverty gap (the amount required to bring the poor up to some minimum
needs standard) unless there has been an offsetting fall in the income
share of the poor. A further methodological point: we assume that the
more meaningful unit for evaluating poverty and welfare is the household,
or consumption unit, rather than the individual income recipient.
Ideally, the analysis would be based on household income per adult
equivalent, but this information has not been published.
1. The Available Data: A Summaryi
The discussion of these questions has so far relied principally
on the 1960 and 1970 Demographic Census information on money income.
Some contributions have also referred to the 1972 and 1973 national household
Surveys (PNAD), and to partial sources, such as wage statistics for urban2/
employees and household surveys in a few cities. The recent publication
of the national household surveys for 1974-75 and 1976 provides an
opportunity for evaluating earlier sources, and for extending the data
to include the whole of the fast growth period between 1968-1974. The
1974-75 Estudo Nacional da Despesa Familiar (ENDEF) is the first household
expenditure survey with national coverage. Since there are strong
grounds for suspecting that direct queries on income, like those contained
in Census and employment survey questionnaires, tend to produce underestimates,
1/ The chief exception being Gary Fields' argument that "average realincomes among families defined as poor by Brazilian standardsincreased by as much as 60% while the comparable figure for non-poorfamilies is around 25%." "Who benefits from Economic Development? -A Reexamination of Brazilian Growth in the 1960s," American EconomicReview, 67: 4 (Sept. 1977), pp. 570-582. Below (pp. 33-34) we argue thatthis conclusion is statistically unsound.
2/ The most thorough attempt to draw on other sources is John Wells,"Distribution of Earnings, Growth and the Structure of Demand inBrazil during the 1960s," World Development 2:1 (Jan. 1974), pp. 9-24.
- 9 -
(both through the unintentional omission of occasional and secondary
incomes, as well as through deliberate underreporting) the household
expenditure study is of particular interest as a check on previous
sources. The available data on household income shares, along with our
own adjusted estimates of the possible range for those shares are shown
in Table 1. "Our estimate" is the midpoint of the range in each year
and should be read as no more than an order of magnitude.
- 10 -
Table 1: ESTIMATED FAMILY INCOME SHARES(Percentages of Total Family Income)
PercentileSource Poorest Top
40 10
1. 1960 Census 9.4 44.5
2. Meesook-Fishlow adjustment 11.5 41.0
3. Our estimate 9.8 50.0
(Range) (8.0-11.5) (41.0-59.0)
4. 1970 Census 8.1 46.2
5. Our estimate 8.4 51.5
(Range) (6.8-9.9) (43.0-60.0)
6. 1972 PNAD: money income only 7.4 50.5
7. Our estimate 8.9 53.6
(Range) (8.5-993) (51.4-55.8)
8. 1974/75 ENDEF: Expenditures & gross saving 9.4 46.0
9. Our estimate 8.5 51.9(Range) (7.6-9.4) (46.0-57.7)
10-. 1976 PNAD 7.5 N.A.
11. Our estimate 7.8 N.A.(Range) (6.8-8.8)
Line
1. Albert Fishlow and Astra Meesook, Technical Appendix, Brazilian Size
Distribution of Income, 1960, May 1972, Table B.5.1, p.54. Based on
one in 1,400 sample of Census, excluding North and Central-Westregions (about 7% of population). Principally money income (see fn. 1,
p. 25), including transfers. Original data were already groupedby size classes so shares of bottom 40 and top 10 are sensitive to
assumed mean incomes for the bottom and the open-ended top size classes.
Fishlow-Meesook use a Pareto fit to estimate the mean for the top
open-ended size class (N.Cr.$50.0+) for individual recipients. Their
estimate is N.Cr.$108.7, (slightly higher than Langoni's N.Cr.$87.9).After combining individuals into families, however, the resultant
open-ended class mean falls to N.Cr.$88.3 [Implicit in Table B.5.1,
p.54, op.cit.] This estimate appears to be low if compared with the
relationship between mean family income in. the open-ended class and
the class limit, in ENDEF and PNAD 1972, both of which publish theactual means in the open-ended class. In both surveys, the ratio is
- 11 -
about 2.1 vs. Fishlow-Meesook's 1.75 (87.6 '. 50.0). These
relationships are shown in Table 5 below (p. 43). If the 1960shares are reestimated using the ratio of 2.1, the Meesook-Fishlowshare for the top 10 rises by 3.6 percentage points, from 41.0 to
44.6, and that of the bottom 40 falls from 11.5.to 11.0. The esti-mates in Line 3 would change in the same proportion.
2. Idem. Principal adjustment is imputation of value of direct consump-
tion of product by agricultural families. The imputation is a decreasing
proportion of money income and, on average, adds 39% to total farm
incomes. Other adjustments are imputations for owner-occupied homes,
and food and lodging received by domestic servants.
3. Obtained by allocation of difference between Census income total (after
Meesook-Fishlow imputation of nonmonetary income) and national. accounts
total personal income (shown in Table 2). Range results frpm two extremeassumptions regarding allocation by percentile: high figure assumes equalproportional underreporting in each percentile; low figure allocatesentire difference to top decile. "Our estimate" is the midpoint ofrange. See text pp. 35-36.
4. Based on size distribution in Carlos Geraldo, Langoni, Distribu ao da
Renda E Desenvolvimento Economico do Brasil, (Rio de Janeiro, EditoraExpressao e Cultura, 1973), Table 1.2, p.26. Langoni's figures excludezero-income families (3.7% of total) but we re-included them for greater
comparability with 1960. Exclusion of zero-income families raisesshare of bottom 40 to 9.5%. Presumably only monetary income is recorded
since Census questionnaire and coding procedures have no instructionsfor imputation of nonmonetary income. Also, income at the top isunderestimated by a coding limitation: All incomes above Cr. 9997
were coded as equal to Cr. 9998 (Costa 1977, p.52).
5. Includes three adjustments: (a) addition of nonmonetary income usingENDEF ratio of nonmonetary to monetary by percentile; (b) use of a Paretoestimated mean income for open-ended income class, i.e., Cr.$4251 instead
of Langoni's Cr.$3245; and (c) allocation of difference between Censusincome total (after adjustments (a) and (b)) and national accounts
disposable income as in Line 3.
6. From Table 3.8 of PNAD-2, 1972 40 Trimestre, Todas Regioes, FIGBE,p.54. This table shows total income by income class and thus eliminates
the open-ended class problem. Includes monetary remuneration only(Rendimento Monetario). A more comprehensive income concept (RendaGlobal), including transfers and income in kind, is used in Table 6.1,
but distributed by individuals, not families. Renda Global is 34% larger
than Rendimento Monetario. PNAD 1972 excludes Regions VI and VII(Federal District and North). Only 0.1% of all families are reportedwith zero income, suggesting an element of incomparability with the
1960 and 1970 censuses,
7. Adjusted for nonmonetary income and for the shortfall of survey incomesrelative to national accounts (Table 2) as in Line 3 above.
- 12 -
8. Based on the Estudo Nacional da Despesa Familiar (ENDEF), Table 7,for Regions I-V (excludes Distrito Federal & North). Covers currentexpenditures (including taxes) plus capital expenditures (principallypurchases of real estate) plus net increases in financial assets,including retained earnings, plus debt payments. The financialitems, however, are not net of sales of assets or of borrowing, andtherefore exaggerate net saving.
9. Adjusted to approximate current income by the arbitrary assumptionthat net saving is 75% of gross saving in each income class, i.e.,that sales of assets plus borrowing amount to 25% of gross savings
in each income class as recorded by ENDEF. Actual dissaving(borrowing plus sales assets) is probably higher but (i) the valueof increases in bank accounts is already a net figure; and (ii) notall dissaving needs to be included - only that which is reflected inENDEF's expenditure figures. Since ENDEF expenditure on durables andreal estate is understated by recording only the value of payments during
the reporting year,and not the full cost,there is no corresponding needto deduct the value of consumer and mortgage debt incurred for those purchases.
Also adjusted for shortfall relative to national accounts (Table 2) asin Line 3 above. All references in the text to ENDEF "income" refer tothis estimate of income based on reported expenditures and savings plusour downward adjustment for unreported dissaving plus our upward adjust-ment for unreported expenditures and/or savings.
10. Table 3 (p.74) of the 1976 Pesquisa Nacional por Amostra de Domicilios(PNAD). Excludes rural areas of Region VII (North) or about 1% ofNational income according to national accounts Table XVIII. Share oftop 10% could not be estimated because the top open-ended class isfar too large: it contains 24% of all families. Includes nonmonetaryand transfer incomes. Only 1.0% families recorded with zero income.Income in the open-ended class derived as a residual from total incomecalculated from Table 14 of PNAD 1976. That table covers all individualrecipients and provides a more detailed set of income categories inwhich the open-ended class comprises only 1.4% of all recipients. Meanincome for that class is based on a Pareto function fitted to the lasttwo income categories.
11. Same adjustment as Line 7.
- 13 -
To the extent that any weight is placed on the estimates in Table 1
they support the view that inequality has probably increased. As will be
argued below, the margin for error in that data is very large and there is
no basis for a definite trend estimate. Nevertheless, it should be noted
that these estimates imply less deterioraticn than is usually alleged, and
allow for substantial growth in the absolute incomes of the poor.
A comparison of unadjusted figures (Lines 1 and 6, or 1 and 10)
shows some increase in inequality: between 1960 and 1976 the poor lose 21%
of their 1960 share and the rich gain 13%. Adjustment, however, is necessary,
because there appears to be a great deal of unreported income in the original
sources. The two main adjustments that are required have offsetting effects:
an imputation for no'amonetary income raises the share of the poor, while an
imputation for unreported money income probably increases concentration.
Though these imputations are largely arbitrary, explicit adjustments are
preferable to the implicit assumptions that are involved in using unadjusted
figures: Any comparison requires strong assumptions regarding the extent
and distribution of unreported income. The purpose of Table 1 is to set
out what we consider to be the most plausible statistical message provided
by the available data. The degree of uncertainty in these data will be
explored below.
2. ENDEF
The recently published 1974/75 national household expenditure1/
and nutrition survey (ENDEF) provides the most reliable basis for statements
1/ By the Fundacao6 Instituto de Estatistica e Geografia (FIGBE). Thesurvey covered 55,000 households between August 1974 and August 1975.The sampling framework was based on the 1970 Census updated in May1974. Interviewers visited each household over a week.
- 14 -
about the structure of incomes. Expenditure surveys usually achieve
better coverage of total income than employment surveys, and this was the
l/case with ENDEF (Table 2). In part, coverage is higher because ENDEF con-
tains the first carefully estimated measures of nonmonetary income at
a national level (Table 3). However, ENDEF also yields higher estimates
for monetary income, presumably because the additive procedure of listing
individual expenditures, along with the observation of a household over a
week served as checks on under-reporting. Another advantage of expenditure
surveys is that expenditure is more stable than income. The resulting
measures are therefore closer to an annual "normal" income, and there are
no cases of families with zero income.
One limitation of ENDEF is that the questionnaire data on incomes
have not yet been published. The estimates in Table 1 are based on published
figures for family expenditures and savings. The sum of these items in
principle equals income, but the published ENDEF data on financial trans-
actions are incomplete: sales of assets and of some forms of borrowing
have not been netted out. Income is therefore overstated to the extent that
recorded consumption or investment has been financed by either borrowing
or sales of assets. In deriving the shares shown in Table 1, we made the
arbitrary assumption that true net savings in each income class were 75%2/
of reported "gross savings." Since saving as a proportion of current
1/ Cf. Altimir [1976], pp. 76 and 81.
2/ The term "Gross Savings" is inexact but is used to indicate thatENDEF figures fail to deduct all debits. See footnotes to Lines 7
and 8, Table 1.
- 15 -
expenditure is much higher in the top income brackets - 45% in the top
decile vs. 13% in the bottom 40% of families - the effect of this assump-
tion is to slightly reduce the degree of inequality, as indicated by
Lines 8 and 9 in Table 1. Furthermore, the discrepancy between ENDEF and
national accounts income shown in Table 2 suggests that income remains
underestimated, despite the overall high quality of ENDEF as a statistical
exercise, and despite the fact that ENDEF's coverage of personal income is high
in comparison with other Brazilian sources, and with most surveys in other1/
countries (Table 2).
3. Underreporting
Some indication of the extent of underreporting in each source
is provided by the comparison with national accounts income and consumption
totals in Table 2. The national accounts figures have been adjusted to
approximate the Census/survey concepts. Crude assumptions were required
because there are no current estimates of corporate retained earnings
(missing in survey/census income), or of transfers within the private sector
(netted out in national accounts National Income). The adjustments, however,
are adequate to indicate orders of magnitude: the discrepancies between
Census/survey and national accounts figures are large and they differ over2/
time. They are surely the product of both unintentional omission of
occasional and secondary sources of income, and of deliberate underreporting.
This interpretation is supported by the fact that, by virtue of more
thorough questioning, both PNAD 1972 and ENDEF discover higher incomes.
It seems likely that even in the latter sources, much income remains
unreported: the obvious candidates are illegal incomes, reinvested profits
1/ See footnote 1, p. 17.
2/ An index of real income per household constructed from unadjustedcensus and survey data rises as follows: 1970=100, 1972=170, 1974/75=232.By contrast, the index of real GDP per capita is 1970=100, 1972"120,1974/75=143.
- 16 -
Table 2: COMPARISON OF ESTIMATES OF ANNUAL PERSONAL
INCOME AND CONSUMPTION
(in billions of current Cr.$)
Survey// National %Census a Accountsb Discrepancy
1. 1960 Census 1.46 2.50 42
2. (plus nonmonetary) 1.73 2.50 31
3. 1970 Census 93.2 166 44
4. 1972 PNAD (money factor income) 164 290 43
5. (plus nonmonetary and transferincome) 221 290 24
6. 1974-75 ENDEF (expenditures plus
saving) 483 600 20
7. (Consumption) 395 480 18
8. 1976 PNAD 916 1279 28
/a Total personal income figures of survey/censuses were adjusted (i) to
include Regions VI & VII (Federal District and North) when coverage
was incomplete, using their share in NNP according to Table XVIII of
Contas Nacionais; (ii) to represent the calendar year, using Rio cost
of living index and real growth trend during the year: PNAD were
in fourth quarter, censuses in August (1960) and September (1970),
ENDEF was August 1974 - August 1975 but all ENDEF values are expressed
in August 1974 prices; and (iii) to deduct intra-family sector
transfers, estimated to equal 5% of total family income in all surveys,
and 3% in the Censuses (since less transfer income is reported in
Censuses). Figures are approximations based on 1972 PNAD figure of
12% for all transfer income, less social security transfers from
Sinopse Estatistico do Brasil 1977, pp. 511-518.
/b National accounts Personal Income estimated here as equal to NationalIncome plus Transfers and Subsidies, less Retained Earnings assumed
to equal 10% of National Income. No estimates of corporate retained
earnings are published, but the national accounts provides a ceiling
in the figure for private sector saving less depreciation, if one assumes
positive household sector saving. The latter assumption is strongly supported
by ENDEF, which slows household savings to be 16% of total income (after dbwn-
ward adjustment for unreported dissaving explained in Table 1, Line 9.)
Private sector saving less depreciation as a proportion of national
income was 6.2% in 1959, 8.3% in 1970, 9.7% in 1972, 11.5% in 1974 and
6.5% in 1976. These figures imply retained earnings well below 1/0%, but they
are subject to much error. Figures for other LDCs range between 5 and 10%.
Thus, the 10% used here is probably an upper bound for the seventies,
and too high for 1960 given the rapid growth of the corporate sector.
- 17 -
Source: Contas Nacionais, Revisao e Atualizacao, July 1977. Thispublication contains revised estimates for earlier years, includingthe benchmark years 1949 and 1959. The revision implies a higher GNPin 1960 than earlier estimates. Since Table VI does not contain anestimate for 1960, the 1960 figure used here is a projection of 1959using the implicit GDP deflator and real growth rate.
Line 1. From Fishlow and Meesook, Technical Appendix, p. 54. Censusreference month for income was August, but if August had not been a typicalmonth, a monthly average for the entire prior years was solicited. Itseems likely that a majority reported current (August) receipts.
Line 2. Same source as Line 1. Includes estimated nonmonetary income (imputed rent,pay in kind of domestics, and own-consumption of agricultural output). The Fishlow-Meesook original reconciliation with the national accounts obtained a maximumdiscrepancy of 10%. Their comparison is with an earlier estimate of PersonalIncome published in the Revista Brasileira de Economia. The figure forDisposable Income used here is based on the recently revised Contas Nacionaisop.cit which revised 1959 NNP upwards by 28%. About half sx thz increaae isdue to the revision of value added in Manufacturing and Commerce. The 1960Economic Censuses, which provide the benchmarks for value added in thosesectors, appear to have classified several components of value added as formsof intermediate purchases under the item "Other Expenditures." This becameapparent from a comparison of input-output ratios in the 1950 and 1960 censusesand 1966-1970 annual establishment surveys by DEICOM/FIGBE. Thus, the ratio ofvalue added to total output in Manufacturing ranges between 39 and 40% in allthose years except 1960, when it was raported as only 30%.
AW discussed in the text (pp. 22-26) the 1960 discrepancies shownhere are probably minimum estimates.
Line 3. From Langoni, Carlos Geraldo, Distribuiga6 da Renda e DesenvolvimentoEconomico do Brasil (Rio: Editora Expressao e Cultura, 1973), p. 64.Langoni's mean income, however, was adjusted upwards by re-estimating theupper open-class mean income with a Pareto fit to correct for the ceilingincome of Cr. 9998 imposed on his data.
Line 4. Table 3.7, PNAD-2, 1972, FIGBE. Defined as monetary remuneration,including cash rents and other property income. (See footnote 6, Table 1,above.)
Line 5. Table 6.1, PNAD-2, 1972. Defined as "global income" includes non-monetary remuneration and transfer income from government (pensions, dis-ability payments) from corporate sector (insurance payments, pensions) andfrom other families (remittances to students, gifts, etc.). Nonmonetaryremuneration is 11.3% and transfers 13.4% of "global income." Rural non-monetary income, however, appears to be underestimated (see text p. 29).Unfortunately this table is a distribution by individual recipients; thereis no table which distributes "global income" by household income classes.Thus the full and relative high coverage achieved by PNAD-1972 is notavailable for comparison with household distributions in other years.
- 18 -
Line 6. Estudo Nacional da Despesa Familiar 1974-75, FIGBE, Table 7.Corresponds to our estimate of family income based on reported expenditureand savings plus adjustment for unreported dissaving. (See footnoteslines 8 & 9, Table 1 above.)
Line 7. Idem. Consumption equals current Expenditures plus purchases ofvehicles classified under Increase in Assets.
Line 8. PNAD, FIGBE, Table 14. Pareto fit on upper two income categoriesused to estimate mean income of top open-ended class, which comprises1.4% of recipients. Includes imputed value income in kind, propertyincome, and transfers.
of family enterprises, corporate dividends and other property incomes,
and, in the case of ENDEF socially embarrassing expenditures.
Though the national accounts are also liable to large error, they deserve
greater credence as measures of the level of total income than replies to the brief
question used in Censuses, or even to the fuller questions in surveys. Moreover,
there is a variety of other evidence, reviewed below, which bears out the claim
that Census and survey incomes are too low. In this respect Brazil is similar to
other countries; it is a common experience that expenditure surveys yield lower1/
incomes than the national accounts. Furthermore, the most plausible hypothesis
1/ Evidence for Latin America is provided by Oscar Altimir, "IncomeDistribution Estimates from Household Surveys and Population Censusesin Latin America: An Assessment of Reliability," IBRD/ECLA, Nov. 1976.After reviewing 49 sources in 14 countries Altimir states that "very.few of the surveys and censuses analyzed show income results which arereasonably consistent with macroeconomic estimates," p. 95. Furtherevidence is provided in the following percentages ofsurvey income to personal income or consumption according to thenational accounts:
73% Mexico, Personal Income, 1963 (Navarrette 1970, p.68).66% Korea, Personal Income 1965 (Bhalla, 1979)73% Korea, Personal Income 1976 (Bhalla, 1979)68-71% Philippines, Personal Income, 1970-71 (Berry, 1978, p.315)80% India, Rural Consumption 1970-71 (Ahluwalia, 1978, p. 21)73% Colombia, Personal Income (Berry, 1978, p. 22)60% Turkey, Non-agricultural Disposable Income (Dervis and Robinson,
1977, pp. 29 and 40).
All the above are expenditure surveys: employment and other non-expenditure surveys usually yield even lower income figures.
- 19 -
regarding Brazil's national accounts is that any net error is one of
underestimation, in part because several components of value added are1/
themselves based on earnings levels reported by the 1970 census. The
discrepancies noted in Table 2, therefore, are likely to be minimum
estimates of income not reported in censuses and surveys.
The most obvious component of underreporting is nonmonetary
income, chiefly own-consumption by farm households and imputed rents on
owner-occupied homes. Census and PNAD income data are commonly described
as being money income only. In fact, their income questions do not exclude
nonmonetary income, and further, the 1972, 1973 and 1976 PNAD's have explicit
questions for income in kind. The 1972 PNAD is particularly thorough in
its questionnaire on all forms of income. However, there is good reason to
believe that most nonmonetary income has been omitted by all sources except
PNAD 1972, and that even PNAD 1972 underreports the bulk of rural income in kind._/
This is an a priori to be expectation from reading the survey methodologies,
1/ Census earnings levels are applied to the self-employed outside agri-
culture, and to construction wage-earners. There are other components
of the national accounts methodology which raise questions of possible
underestimation, e.g., value of own-construction, and rural non-monetary income. Also, total income of agricultural households according
to ENDEF is about 40% higher than national accounts value added in agri-
culture (Cr.$96 billion vs. 68 billion in 1974 prices). Presumablyagricultural households are receiving substantial amounts of income from
non-agricultural sources, e.g., construction labor, trading, social
security. Also their imputed rent should be deducted (about 7% of income).On the other hand ENDEF probably omits much property income generated
in agriculture, but not attributed to "farm households," because theproprietors are urban residents with other occupations. This comparison
thus definitely suggests underestimation of agricultural income in thenational accounts. At the same time, the national accounts probably
overestimate some items, but our view is that the net error is not likelyto be on the low side.
2/ In PNAD 1972, rural farm workers (mensalistas and diaristas) are not
asked the value of production for own-consumption. Landowners and
sharecroppers are asked to list quantity and price of up to six productswithdrawn for own use. Out of 107 foods listed by ENDEF in 67 cases
own-consumption accounted for half or more of total rural Northeast
consumption. Accurate recall on such a question would in any case be difficult
to achieve. PNAD 1976 has only a summary question on total income in kind.
- 20 -
but the basic evidence is the much higher values of nonmonetary income
reported by ENDEF and by a set of rural family expenditure surveys carried1/
out in 1962/63 bv the Funda;ico Getalin vargas.
ENDEF provides the first carefully estimated measures of non-
monetary income at a national level and confirms the earlier results of the
FGV: both sources report ratios of rural nonmonetary to monetary income of
about 50% (Table 3). The FGV ratio (46.5%) was used by Fishlow and Meesook
for their adjustment of the 1960 agricultural distribution. Some doubt
was placed on the FGV ratio by two later studies that yielded much lower
figures. Langoni based his discussion of subsistence consumption on a relatively2/
limited survey (597 farm households) which produced a ratio of only 15%. Fishlow,
in a more recent paper, cites the 1972 PNAD which also arrives at much lower figure
for rural households (17%), though its ratio for urban households is close3/
to that of ENDEF (14 and 18% respectively). We have assumed that, on
1/ The 1962/63 surveys were carried out by the Fundapao Getulio Vargas,in rural areas of the States of Pernambuco, Ceara, Minas Gerais,Espirito Santo, Sao Paulo, Santa Catarina and Rio Grande do Sul.
1816 families were interviewed. The 46.5% is for individual earnersand combines imputed rent (about 6%) and imputed consumption of farmproducts (39%). Fishlow-Meesook's Tables B.0.1 and B.3.3 (pp. 25 and45 in Technical Appendix, 1972) show a slightly higher ratio - 48% - forfamilies.
2/ Cited in Langoni, Distribuicao da Renda, pp. 28 and 40. The survey wasby Carlos Langoni and Guilherme Dias, "Avaliaa-o Economica do Servicode Extensao Rural," Convenio ABCAR-IPE, 1972. The ratio is stronglyprogressive, rising from 1.2% for the top rural decile to 44.6% for thepoorest 40%, and 79.6% for the bottom decile.
31 Albert Fishlow was misled by these PNAD 1972 figures into justifyinghis comparison of unadjusted 1960 and 1970 Census and 1972 survey data.From that comparison he concludes that there was a substantialdeterioration in the share of the poor. "These data [PNAD 1972 data onnonmonetary income] therefore decisively reject the promise that aproperly calculated measure of real income would cast the Braziliansize distribution in a more favorable light" in "Brazilian IncomeDistribution: Does Trickle-Down Really Work?" unpublished paper pre-sented to Workdhop on Analysis of Distributional Issues in DevelopmentPlanning, Bellagio, 1977.
Note that 62% of agricultural households are farmers, and that ruralwage-earners have several sources of nonmonetary income, viz. pay inkind (both food and housing) small plots, wood and food gathering, andhome manufacture.
- 21 -
methodological grounds, the two household budget surveys (FGV and ENDEF)
are more likely to measure both monetary and nonmonetary income adequately,
and therefore that the nonmonetary component is in the order of one-third
of total rural income.
Table 3: NONMONETARY AS A PROPORTION OF MONETARY FAMILY INCOME
All Families Urban Rural
ENDEF 23 18 50
Rio 19 18 36
Sao Paulo 19 18 34
South 27 18 50
Minas Gerais 26 18 52
Northeast 30 20 57
FGV 19 47
PNAD 1972 14 14 17
Rural laborers 9
Farmers 17
Notes: 1. ENDEF 1974/75, Despesas Das Familias. Regions I-V, Table 4.Total current and capital expenditure plus saving is used asan estimate of income. Since savings are somewhat overstated(see above p. 14) the above ratios are underestimated.
FCV. Food Consumption in Brazil, Family Budget Surveys in the
Earlv 1960s, Data cited bv Fishlow-<4eesook in echnicalAppendix, pp, 25 and 45.
3. PNAD-2, 1972, Table 4.7 . Refers to farm and nonfarm rather
than urban-rural. The heavy weight of nonfarm plus decimalrounding explain equality between national average and nonfarmratios. Excludes rural areas of Region VII (North).
- 22 -
There is also evidence that the nonmonetary proportion is much
higher for the poorest families. Within the Langoni-Dias sample the ratio1/
rises from 1% in the top rural decile to 80% in the bottom decile. The
more comprehensive FGV survey yields similar results: when income in kind
is added, mean income for the poorest quintile rises 100%, vs. about 40%2/
for the top quintile. The omission of nonmonetary income is therefore
particularly damaging to the measurement of poverty. This is corroborated
by differences in the estimates of families reported to earn one minimum
wage or less in different surveys. The proportion of such families
drops drastically and implausibly, from 47% in
1/ Langoni, Distribuicao, p. 28.
2/ Fishlow-Meesook, Technical Appendix, Tables B.0.1 and B.3.3.
- 23 -
1/1970, to 38% in 1972, to 12% in 1974/75. In 1976 it rises to 20%.
Though rapid income growth during this period may indeed have reduced
poverty, most of the reported change must be due to varying coverage of
nonmonetary income. The intermediate 1976 figure can be explained by
its partial coverage of nonmonetary income. The 1976 income question
requested separate answers to monthly earnings in cash and in kind, but
as with the even more detailed 1972 question, it is likely that coverage
of nonmonetary income was incomplete, as might be expected from any similarly
direct and brief question on non-cash income, and as is suggested by the
fact that the PNAD proportion of families with one minimum wage or less is
almost double that of ENDEF.
There is evidence that monetary income is also underreported in
varying degrees in the different censuses and surveys. The 1960 census
figures, after the substantial Fishlow-Meesook adjustment for income in kind
are still 31% short of the national accounts estimate of personal income.
The ENDEF figure for Family Consumption Expenditure is about 18% short of
the national accounts Personal Consumption. The discrepancy between the
1976 PNAD and the national accounts (28%) is too large to be explained by
missing nonmonetary income, since some nonmonetary income is already included.
Another indication of missing income is provided by the higher coverage that resulted
1/ For 1970, Censo DemograZLi 0 ,able 10, p.226. Three-quarters of theCr. 151-200 class was assumed to earn up to one minimum wage (Cr. 187).The 1972 figure is from Table 3.6 of PNAD-2 which excludes transfer andnonmonetary income. (There is no table for families which includesnonmonetary income, equivalent to Table 6.1 for individual recipientsthough as argued in the text above even Table 6.1 probably underestimatesrural nonmonetary income). The 1976 figure is from Table 3, p.74.
The real minimum wage changed only slightly between 1970 and 1976.
- 24 -
1/
from the much more detailed income questionnaire used in 1972. As shown in
Table 2, overall coverage was close to that of ENDEF - the main difference
being the lesser coverage of rural nonmonetary income by PNAD 1972 - and
significantly higher than that of the Censuses. The composition of 1972
incomes is revealing: Transfers account for 13.4% of total income, income
in kind for 11.3% and monetary remuneration (including property income)2/
for 75.3%. It seems likely that the Censuses, and possibly, later PNAD's
were less complete in reporting transfer income.
4. The 1960 Census
The .1960 census deserves special attention .because it is the
sole basis for long-run comparisons of distribution. Furthermore, it is
the only source which shows a significantly different degree of inequality:
the income shares shown in Table 1 imply a better distribution in 1960
than in later years. It is clear from Table 2, however, that such com-
parisons will be subject to a high degree of uncertainty: about 42% of
personal income was not reported in the Census. Moreover, there is a
range of uncertainty associated with the amount that was "reported." Because
incomes were recorded by class intervals, the top and bottom shares are
sensitive to the assumptions regarding mean incomes in the extreme income
classes. But the more critical questions have to do with the allocation
of unreported income.
1/ The questionnaire listed six different categories of income or recipients(wage-earners, self-employed, rural laborers, rural proprietors and
sharecroppers, property income, other monetary and nonmonetary income)and then further listed various types or components of income withineach category, e.g., wage-earners were asked to report separately each
of the following: Cash wage, bonuses, overtime, paid holiday, severancepay, and others. By contrast, the 1960 and 1970 Censuses simply asked
for average monthly earnings.
2/ Estimated from Tables 2.7 and 6.1, PNAD-2 1972. Transfers include
pensions, family remittances, lottery prizes, gifts, insurance payments, etc,
- 25 -
The omission of nonmonetary income, which chiefly affects the
poorest deciles, was recognized by Fishlow and Meesook and was the subject
of a detailed adjustment procedure, but a comparison with ENDEF raises the
question whether Fishlow-Meesook adjusted enough: they arrive at a ratio
of nonmonetary to monetary income of 18%, which is lower than the 23%1/
found by ENDEF. For the poorest 40%, their ratio is also lower than
ENDEF's - 43% vs. 53%. The real value of nonmonetary income per poor
household in 1974/75 is exactly twice the amount imputed by Fishlow-Meesook2/
in 1960. Perhaps a third of the estimated difference is due to the3/
relative increase in farm prices. The remainder must represent either
growth in the quantity of nonmonetary consumption per household, or
underestimation in 1960. Growth in the quantity of own-consumption
may seem implausible, given the trend towards monetization, but (i) the
income elasticity of expenditure on housing - and therefore
imputed rents - is positive, and probably exceeds 1.0, and (ii) regional
cross-section evidence suggests that rural nonmonetary consumption rises
with income: in 1974-75, nonmonetary expenditure per household
was Cr.$6312 in the South, Cr.$4890 in Sao Paulo, and only Cr.$2742 in4/
the Northeast. Since there is evidence of some real income growth
1/ Fishlow and Meesook op.cit. One aspect of their method raises thepossibility of some compensating overadjustment in the lowest incomecategories. Fishlow-Meesook assume that the Census reports moneyincome only. (The Census question was "How much do you earn, on the
average, per month?") But in the Census, no farm families are reportedwith zero income, though 5% of nonfarm families are reported with zero
income! In the case of farm households, however, zero income wouldimply no cash income during a whole year because variable-income families
were asked to report their mean income for the last 12 months. It seemsimplausible that there were no pure subsistence families in 1960, this
statistic thus suggests that the Census is not pure money income.
2/ In 1974-75, nonmonietary income per family per annum, within the poorest40% families, equalled Cr.$2130. The corresponding Fishlow-Meesookvalue for 1960 (in 1974-75 prices) was Cr.$1056.
3/ The farm/nonfarm terms of trade rose about 40% over the period 1960-1974,
but an index that allows for own-consumption by farmers would rise less.
4/ ENDEF, Table 4.
- 26 -
amongst the poor, some growth in own-consumption is likely.
Urbanization could have been expected to reduce nonmonetary consumption,
but the data show no urbanization of the poor: the urban-rural composition
1/of the bottom 40 remained constant. On the other hand, since the
elasticity of nonmonetary to total expenditure is less than 1.0, and some
degree of monetization is known to have occurred, the evidence still
points to some degree of uunderestimation of 1960 nonmonetary income of
the poor. Unfortunately, there is no basis for a more precise statement,
nor for an estimate of underreported cash income of the poor.
It is also difficult to gauge the error at the top of the 1960
distribution. One difficulty was noted above (notes to Table 1, Line 1):
data from ENDEF arnd P.'AD 1972 suggest that the Fishlow-Meesook mean income
for the open-ended class may be too low, in their household distribution.
In both these more recent surveys the ratio of the mean income
in the open-ended class to the limit of the class is higher (between2/
2.0 and 2.1) than in Fishlow-Meesook's calculation (1.8). This apparently
trivial difference is enough to raise the uncorrected share of the top
10 in 1960 (Table 1, Line 1) from 44.5 to 47.8%, i.e., it eliminates most
of the apparent increase in income concentration at the top since 1960.
In sum, there is an important margin of error in simply establishing what
the Census reports as the share of the rich. This is in addition to the
1/ In fact, the rural share appears to have grown. According to Fishlow-
Meesook, 60% of the bottom 40 in 1960 were agricultural families. In
1974-75, 66% were rural. This could be explained by a combination of
urbanization with faster poverty reduction within urban areas than in
rural. But it may be, to some extent, spurious - caused by a greater
underestimation of urban than rural incomes in 1960 (after the adjust-
ments for income in kind), thus overrepresenting urban poverty in 1960.
2/ See below, Table 5, p. 43. In their uncorrected household income distri-
bution, the implicit mean income for the Cr.$50.0+ class is Cr.$88.3
which is lower than Pareto fitted mean of Cr.$108.7 applied to the
distribution for individual recipients, and lower than the values (Cr.$100.0-
105.0) that would result from using the 1972 and ENDEF ratios of mean income
to limit income.
- 27 -
margin of error associated with the allocation of unreported cash income,
which amounts to some 31% of personal income (Table 2, Line 2).
There is indirect evidence of considerable underreporting by the
rich in 1960. A set of family budget studies carried out'by the Fundaya6-
Getulio Vargas between 1961 and 1963, suggests that incomes at the top of
the distribution were much higher than reported by the Census. About 10%
of FGV's rural sample had family incomes over Cr.50 per month (1960 prices)
vs. only 0.5% in the Census, and 15% of the urban sample, vs. 3.0% in the1/
Census. Also, the top income category in the Census began at Cr.$50.02/
which in 1978 dollars represents about US$800 per month. Only 1 in 6
professionals (doctors, dentists, judges, lawyers and engineers) reported
monthly incomes over that level. Half of all professionals declared
monthly incomes under US$500. Similarly, only 1 in 4 university graduates3/
reported earning US$800 or more. Again, the census reports only 32,000
r 4/agricultural households with monthly incomes over that level. Yet according
1/ A total of 9125 families were interviewed, 7,309 in the urban sector and1,816 in the rural. The survey was not designed as a nationwide sample,but had widespread regional coverage and was stratified to representstate capitals, towns, and rural areas. The sample frame was drawn fromthe 1960 census. Getulio Vargas Foundation, Food Consumption in Brazil,Family Budget Surveys in the Early 1960s, p. 3. FGV data are for urbanand rural, census for agricultural and nonagricultural.
2/ This conversion uses the more realistic 1978 exchange rate of 18.07,after reflating Cr.$50 by the adjusted Rio consumer price index'Using the overvalued 1960 exchange rate (0.187) and US cost of living indexyields a lower dollar figure of $590. In 1960 dollars Cr.$50 was theequivalent of only US$270.
3/ Graduates defined as having completed 5 or more years of superioreducation. Only 0.6% of the labor force were university graduates.
4/ Fishlow and Meesook (op.cit, p. 25) report 0.5% households in agricultureearning N.Cr.$50 or more per month. Censo Demografico 1960, p. -14gives 6.4 million households with head of household employed in agriculture.
- 28 -
to the CIDA study of land tenure, there were 97,400 latifundios in Brazil
1/in 1950 and over 100,000 in 1960. The census income declarations by all "rich"
farm households, defined as those earning $670 or more per month, amount
2/to only between 7 and 9% of 1960 agricultural value added. Those
figures imply that the usual picture of income concentration drawn from land
tenure data is grossly exaggerated - or, that rich farmers underreported3/
substantially in 1960.,
The margin of error in the 1960 income data is thus substantial
and limits any attempt to evaluate either the degree of inequality in
1960, or distributional trends over the sixties.
1/ Defined as farms employing 12 or more workers: they range in size from
100 hectares up. CIDA, Land Tenure Conditions and Socio-Economic Develop-ment of the Agricultural Sector, Brazil, Pan American Union, 1966,
pp. 82-90. Using size alone as a criterion, there were 190,000 farmsof 200 or more hectares in 1960, and 74,000 with 500 or more hectares.
2/ 1959 agricultural value added = 367 million Cr.$ (Table XVII Contas
Nacionais). 1960 value added was approximately Cr. 477 million. Total
income received by the richest 32,000 farm households was Cr.$34 million,using Langoni's estimate of mean income in the open-ended income class(Cr,$87.6 per month) or Cr. 43 million, using a Pareto fitted mean
income (Cr.$112 per month).
3/ The Census figures would also imply extraordinary mobility between 1960and 1974/75 since ENDEF reports about 8 times as many rural households(250,000 vs. 32,000) in that income categorv in 1974-75.
5. Other Years
Most of the comments made about 1960 apply also to the 1970
1/Census income data. The income question was similar; the discrepancy
with the national accounts is of about the same magnitude (44%): under-
reporting clearly includes omission of most income in kind as well as
underreporting of money incomes. In this case there is a better basis for
establishing at least a minimum estimate of underreporting by the rich:
1970 income tax data show twice as many persons in the census top income
bracket as are reported in the Census. It is a minimum for the obvious
reason that income tax declarations are likely to underreport. Langoni
has provided minimum estimates of Census underreporting at the top of the
urban distribution: in 1970 non-agricultural tax-payers reported 84% more
income than Census income recipients with incomes equal to or higher than that
3/of the poorest taxpayers (Cr.148 per month). These findings also support
the argument that top percentile incomes were substantially underreported
in 1960.
The 1972 and 1976 surveys differ in their treatment of income,
between themselves and with respect to the Censuses. The 1972 survey
was particularly thorough, as was noted above (p.19), and yields a
l/ The questionnaire provides for a single entry under "average monthlyincome ; the Instruction Manual for interviewers specifies, as in 1960,that fixed income recipients should state last month's income whilevariable income recipients should average las;: year's income; it alsospecifies that receipts from regular remittances, rents or dividendsshould be included, and that loans, prizes, ir.heritances should be excluded.
2/ The Census (Table 26) reports 328,500 persons in the top bracket (Cr.$2001or over per month) vs. 687,000 according to tax datareported by Virgilio Horacio Samuel Gibbon, "DistribuiFio- de Renda eMobilidade Social: A Experiencia Brasileira," doctoral dissertation,
Fundasa-o Getulio Vargas.
3/ Langoni, Distribuicao da Renda, p. 276.
4/ The 1968 and 1973 surveys did not report the income of agriculturalemployers or own-account farmers, who together made up 31% of the laborforce.
- 30 -
coverage almost as high as ENDEF, though unfortunately the full income
results havenot been published in terms of household incomes. The published
data on household income covers monetary remuneration only, omitting
nonmonetary and transfer income which together amount to 25% of total1/
income recorded by the survey, and 43% of national accounts total personal
income (Table 2). The 1976 survey does include a household table based2/
on total income as measured by the survey, but its income data have the
disadvantage of being somewhat intermediate in coverage between the more
thorough ENDEF and 1972 PNAD,and the more casual censuses. Imputation
of missing income would be complicated by the uncertain degree of coverage
of transfer and nonmonetary incomes. Certainly the margin of error would
be far too large to infer changes during the seventies, while for comparison
with 1960, it would involve more uncertainty than a use nf ENDEF.
Comparisons based on individual recipients rather than households
face additional pitfalls. Bias may be introduced through both apparent
and real changes in the patterns of household structure and labor force
participation. A case of apparent changes arises because the more thorough
surveys of the 1970s not only improved the coverage of income, they also
appear to have found more income recipients than the censuses. This shows
up in a higher female participation ratio, and in a larger number of
1/ The 1972 household income shares presented by Albert Fishlow in"Brazilian Income Distribution: Does Trickle-Down Really Work?" unpublished,
April 1977, p.6, are monetary remuneration only, based on Table 3.1 ofPNAD 1972, and thus omit about 43% of national accounts personal income.
This is also true of the 1972 data on individuals in his Table 1 (p.3).Total survey income measured by PNAD 1972 has so far been published only
by income class of individual recipient, in Table 6.1, p.95 of PNAD-2, but
one hopes, a household tabulation will be made available eventually.
2/ PNAD 1976, Table 3, p. 74. Unfortunately, the table shows only six
income categories and the open-ended top category includes 24% of all
families.
- 31 -
pensioners and other inactive income recipients. Between 1970 and 1972
the female participation ratio (after deducting unpaid family workers)1/
jumps from 16.5 to 22.9%. The proportion of inactive income re-ipients
also rises sharply, from 3.4% in 1960, to 6.0% in 1970, 8.9% in 1972 and
11.4% in 1976. Some of this increase may be the effect of growing pension
coverage and disability payments; some is surely the effect of more intense
questioning in the surveys. The increased coverage of female and inactive
income recipients causes a major change in the structure -f 4.ncome
recipients: the proportion of males aged 20+ in the total falls from
64.4% in 1960 to 50.0% in 1976. Since women and the inactive (pensioners,
persons dependent on remittatices, etc.) are far more concentrated at the
bottom of the distribution, the effect is an apparent increase in poverty,
and probably, greater measured inequality. These biases will arise
particularly in any comparisons involving survey and census incomes.
Another pitfall results from a real demographic change: the
reduction in the proportion of unpaid family workers. This bias has
prejudiced the most widely cited conclusions drawn from the 1960 and 1970
censuses, in particular, those regarding the share of the poor. If the value
of income in kind is not imputed to these workers, the
reduction in their numbers creates an artificial impression of reduced
poverty, through a fall in the number of persons supposedly receiving
"zero income." This bias underlies Fields' and Morley-Williamson's (1977)
1/ There are numerous references in the literature to "rising femaleparticipation." Comparisons of the 1960 and 1970 censuses, and ofthe 1968 and 1973 surveys do show some rise, though at modest rates.Reported female participation rates, especially in agriculture, arenotoriously sensitive to questionnaire wording and procedure and soare subject to large margins of uncertainty.. The jump between 1970and 1972 is clearly the product of questionnaire change. The rate isconstant between 1972 and 1976.
- 32 -
conclusion that the poor fared much better between 1960-1970 than was1/
commonly believed. Langoni's calculation, which instead excludes unpaid
family workers from the comparison, creates an opposite error. The change
in job status of a large number of low productivity women and children -
from family helpers to paid wage-earners or self-employed - causes a spurious
growth in the number of persons employed in the lowest income categories
(they are exciuded in 1960 but included in 1970), and thus biases Langoni's
calculation towards underestimating income growth at the bottom. This particular
problem can be avoided by either imputing an income to family workers, or2/
comparing hoUsehold rather than individual incomes.
6. Censuses and Surveys: Conclusions
We have argued that income distribution estimates based on
census and sUrvey income data carry a large margin of uncertainty. The
main reason is undercoverage. A variety of evidence has been cited pointing
to the understatement of incomes in the censuses and surveys. But margin
for error arises not only from undercoverage; it is caused also by the
need to impUte mean incomes to the open-ended top class, by uncertainties
1/ Fishlow demonstrates that "Eliminating zero income workers from thedistribution fully reverses his [Fields'] resultg. Fields' claim tothe contrary (p.573, p7) is wrong, the result of arithmetic mis-calculation, When zero income wotkers are eliminated both in 1960 and1970, the share of income received by "the poor".... declines from 8.9to 8.0%, [between 1960 and 1970]. That translates into an averagegrowth of absolute income for those under the poverty line of 19%,compared to 32% overall." Albert Fishlow, "The Fallacies of Decompo-sition: A Comment dn Fields' Reexamination of Brazilian Income Distributionin the 1960s,it unpublished, p.3.
2/ In fact, this problem is recognized and discussed by Langoni. He showsthat if one includes zero-income workers, the growth in the Ginicoefficient is reduced from 14% to 9%. (Langoni, op.cit, p.67).He doesn't point out that the share of the poor, as distinctfrom the Gini, is especially sensitive to the treatment ofunpaid family workers. The only statistics he presents on income sharesfor 1960-1970 (in the much-cited Table 3.5) exclude the zero-incomelabor force.
- 33 -
regarding what is and is not covered in the sources, and by changing
methods and definitions. The main difficulty arises in trend comparisons;
ENDEF now provides a much improved cross-section measure of income levels
and of the extent and composition of poverty.
The degree of undercoverage severely qualifies the standard assertions
regarding trends in inequality in Brazil. The two most widely cited contributions
are those by Fishlow-Meesook (1972) and Langoni (1973). Both conclude,
from a comparison of uncorrected census data, and with no explicit assump-
tion regarding changes in the distribution of unreported income, that
inequality increased. Yet, if the estimates of undercoverage in Table 2
are approximately correct, the percentile shares and Gini coefficients
calculated by Fishlow-Meesook and Langoni are measuring dispersion of a1/
subset covering only some 55 to 60% of personal income. Also, as noted
above, these comparisons are prejudiced by being based on individual
recipient rather than household data. All other statements regarding
distribution during the sixties have been based on the abovementioned studies,2/
or on independent calculations using the same uncorrected census data.
1/ As mentioned above, Fishlow-Meesook do correct the 1960 distributionand use the corrected figures for an analysis of poverty and a decom-position analysis of inequality in that year, but their 1960-1970 com-parison is based on uncorrected data only.
2/ A study by Ramonaval Augusto Costa, DistribuicWto da Renda Pessoal oBrasil, 1970 (Rio de Janeiro: FIGBE, 1977) ail6s his own, independentlycalculated Ginis for 1970, but using, again, the uncorrected censusdata. Costa's presentation of the data, by the way, incorrectlycompares Fishlow-Meesook's adjusted 1960 Gini (0.52) with their unadjusted1970 Gini (0.63); the unadjusted 1960 Gini is 0.59. Wells (1974) beginshis paper with references to Fishlow and Langoni and the statementsthat "there was a considerable increase in the degree of inequality,"(p.9) and "Now that it is generally accepted that a deterioration inthe distribution of personal income, as measured in the Census, hasoccurred ......" (p.9). Bacha and Taylor (1978) in their review of thedebate on income distribution also take the worsening as adequately
(Continued on next page)
- 34 -
And, to date, we have seen no attempt to compare the 1972 or 1976 surveys
with 1960 that recognizes and explicitly deals with the problems of under-1/
coverage and change in methods.
What this implies has been stated succintly by Fishlow and Meesook:
.only to the extent that the approximation to a comparable national
accounts measure is close can the calculated distributions pretend to portray
the real distributions accurately. A large divergence allows scope for a
far different profile since the residual need not be distributed identically2/
to that of the income segment included."
What should be added to this statement is first that census and survey income
data are weakest at the extremes of the distribution, thus magnifying the
uncertainty regarding the most important distributional statistic - the
share of the poor. And second, that the degree of uncertainty matters more
for estimates of how much income shares have changed rather than simply,
whether they have gone up or down.
2/ (Continued from p. 33)
proven; their discussion of the basic statistical evidence is limited
to a brief "dismissal"t of the Morley and Williamson (1975) and Fields
(1977) calculations which, because they include zero-income recipients,
show faster relative growth of income at the bottom (see below p. 32).
Independent calculations using uncorrected Census data were also carried
out by Rodolfo Hoffman and J.C. Duarte in "A Distribui,cao da Renda o
Brasil," Revista de Administracab de Empresas, 12(2): 46-66 (1972).
1/ Fishlow's comparison using 1972 household monetary remuneration data
only was referred to above p. 3Q, fn. 1. A study in progress by Regis Bonell
and Paulo Vieira da Cunha of INPES/IPEA, comparing 1960, 1970 and 1976
was cited in Jornal do Brasil, Oct. 22, 1978, p. 38. Their comparison
also uses unadjusted 1960 and 1970 data and compares these data with
1976 survey incomes that has partial coverage of income in kind and
fuller coverage of transfer and secondary incomes, as well as of income
recipients.
2/ Fishlow and Meesook, Technical Appendix, p. 55. This statement did not
inhibit their analysis of 1960 because their comparison with an earlier
estimate of Personal Income by the official national accounts office,
showed a iiuch smaller discrepancy than that obtained here (Table 2).
- 35 -
In Table 1 (p. 10) we provide some orders of magnitude on the
margin of error that may result from undercoverage. The national accounts
provides a basis for estimating the amount of missing income, though the
size of the residual is subject to a large margin of error. The allocation
of that residual is crucial to the calculation of distributional trends, but
there is no information on which to base an allocation of that missing
income by income classes. At one extreme, for instance, the Fishlow-Meesook
procedure assumes that only nonmonetary income was omitted or, what is
equivalent, that the underreporting of money income is proportional to
income. Their adjustments, in consequence, raise the share of the poorest
40 from 9.4 to 11.5% (Table 1, Lines 1 and 2). At another extreme, one could
follow Navarrett' s procedure for Mexico; she assumed that all underreporting
of money income was by the rich, and therefore allocated the entire difference1/
between survey and national accounts incomes to the upper income classes.2/
The same procedure was used by Adelman and Morris. This assumption can be
defended by noting that only the rich have an income tax incentive to
underreport, and that entrepreneurial incomes are more likely to be
underreported than wage incomes.
If the entire discrepancy in money income is allocated to the
top decile, the 1960 share of the poor (including nonmonetary income) is
reduced from the unadjusted census figure of 9.4% to 8.0%, while that of
the top decile is raised from 41% to 59%. Allocating the discrepancy in
money incomes to the rich increases measured inequality in any given year,
but the effect is greater in 1960 than in 1972, 1974/75 or 1976, because
1/ Ifigenia M. de Navarrette, "La Distribucion del Ingreso en Mexico
Tendencias y Perspectivas" in El Perfil de Mexico en 1980, (Mexico
DF: Siglo XXI Editores, 1970), p. 61.
2/ T. Adelman and C.T. Morris, An Anatomy of Income Distribution Patterns
in Developing Nations - A Summary of Findings, IBRD Economic Staff
Ulorking Paper No. 116, 1971.
- 36 -
the size of the discrepancy is greater in 1960. The Fishlow-Meesook
assumption, which is equivalent to making no adjustment for the under-
reporting of money income, thus yields the least inequality in all years
but also the maximum increase in inequality between 1960 and 1974/75:
the share of the bottom 40 falls by 18% (from 11.5% to 9.4%: Lines 2 and
P., Tabl 1). Tbe Na.xRarrette-Adel-marn--Mcrris assumption yields the smallest
increase in inequality: the share of the bottom 40 falls by only 5%1/
(from 8.0% in 1960 to 7.6% in 1974/75).
The true distributions in 1960 and other years, surely lie
somewhere between the limits produced by these extreme assumptions.
Altimir [1977] disaggregated the discrepancy between census/survey data
and the national accounts and found higher underreporting of urban
entrepreneurial incomes than of wages. His own adjustmenit procedure
therefore makes a proportionately greater allocation of the discrepancy2/
to the rich. Any such specific estimates, however, are highly arbitrary. In
Table 1 we have chosen instead to indicate the range of possible estimates
1J Comparison with 1972 and 1976 is more difficult because both yearsinclude an undetermined amount of nonmonetary income. If PNAD 1972monetary remuneration only is used (Line 6, Table 1), and if it isassumed that the 1972 ratio of nonmonetary to total income in thebottom 40 equals that in 1974/75, then the two extreme assumptions yield
bottom 40 shares of 9.2% and 6.4% which, if compared with the correspondingshares for 1960 (11.5 and 8.0), both imply a reduction of 20%. Usingthe same procedures for 1970 also yields equal though smaller (15%)reductions in the share of the bottom 40. However, comparison with1972 total survey income, or with 1976, both of which have better coverageand smaller discrepancies, yields results similar to the comparisonwith ENDEF, i.e., allocating the discrepancy to the rich results in asmaller measured decline in the share of the poor than results from theFishlow-Meesook assumption. The range of variation in the share of thebottom 40 produced by all these different comparisons (comparing 1960with different years and using each of the extreme assumptions) is +6% to-23%.
2/ Altimir [1977], pp. 82-88.
- 37 -
that are made possible by different assumptions regarding the allocation1/
of unreported monetary income.
Those ranges make it clear that there is no basis for definite
statements regarding the amount or even direction of change in distributive
shares. Most alternative comparisons indicate an increase in inequality,
particularly between 1960 and 1970. At the same time, most of those
comparisons indicate that the size of the change in distributive shares
was relatively small in relation to the growth in total per capita income:
between 1960 and the mid-seventies the share of the bottom 40 may have
fallen by about 15 to 20%; during the same period per capita income rose
about 90%. Despite the large margin of error attached to the estimates
of shares, and the probability of some increase in inequality, the data
imply substantial improvement in the real incomes of the poor.
1/ The margin of error implied in that range is in addition to that resultingfrom Fishlow-Meesook's estimates of nonmonet.ry income and our own assump-tion that the ratio of nonmonetary to monetary income as measured byENDEF is applicable to 1970, 1972 and 1976.
- 38 -
IV. TRENDS IN INEQUALITY: PARTIAL DATA
Numerous studies have examined trends in inequality within
one or another component of the Brazilian economy, both as corroboratory
evidence of trends in the aggregate distribution, and as a way to explain
overall trends. As evidence of overall trends, there is a temptation
to infer too much by making casual ceteris paribus assumptions. The
migration of a small farmer to a city may increase urban dispersion,
but reduce overall inequality. Partial analyses each make a limited
contribution towards a better aggregate picture, but one cannot extrapolate
from trends in any one component without strong assumptions about
associated changes in the rest of the economy. With this in mind, we
examine below income differentials by levels of education,
within the urban modern sector, and across regions and
sectors.
1. Education and Income Differentials
C3rowing income differentials by educational groups are fre-
quently cited to corroborate the view that inequality has increased, as1/
well as to provide a causal explanation for that increase. In fact,
Langoni's argument that growing skill differentials explained most of
the increase in inequality became the main issue in the debate on income
1/ Most notably by Langoni in Distribuicao da Renda, 1973.
- 39 -
distribution. The underlying statistical "fact", viz. that income
differentials by educational groups had widened sharply during the sixties,
was not questioned in that debate. Bacha and Taylor [1974] do point out
that the data are a "real puzzle," since schooling grew rapidly in the
sixties, and they cite the failure of attempts to specify plausible2/
models that would reproduce the data. Yet they also cite and implicitly
accept the Census statistics on educational differentials, as set out by3/
Langoni.
The puzzle may be largely artificial: our own estimates for
1960-1970 suggest that widening during the sixties has been exaggerated;
also, data for 1972 and 1976 suggest that incomes of the more educated may
have lagged behind most others since 1970. Furthermore, the most dramatic
figure cited by Langoni - an increase of 52% in college level incomes
between 1960-1970 - is an exceedingly arbitrary estimate. Arbitrariness
is unavoidable because of the large proportion (18.2%) of college level earners
1/ Wells [1974] provides a clear exposition of what he terms Langoni's"neo-classical" or wage differential explanation, and of an alternative"wage squeeze" or policy-based explanation, which he prefers. Thedebate on this issue was reviewed recently by Bacha and Taylor [19781,who also support the wage squeeze argument.
2/ Bacha and Taylor, JDS 1978, p. 282, "Widening skill differentials inthe face of increasing numbers of the skilled during the 1960s make upthe real puzzle in [Langoni's] Table 4." They cite four modellingattempts, by Morley and Williamson [1975], Fishlow [1973], Cardoso[1978] and Lysy and Taylor [1978].
31/ Langoni [1973]: Table 4.2, p.86. In fact, Bacha [1974(a); 1974(b)] alongwith Fishlow [1977] have argued that widening differentials can beexplained by a link between managerial salaries and profits. They sup-port the argument with data showing growing wage differentials,within the manufacturine sector and other laree establishments. Thesedata are discussed in the following Section IV.3, pp. 59-67.
- 40 -
that fell into the open-ended income class in 1960; the average income of all
college graduates in that year is influenced heavily by whatever assumption
is made regarding the mean income in that class. Langoni's assumption, as
we argue below, is a point toward the low end of the range of possible
estimates of college level income in 1960 and, as a result, yields a
particularly high estimate of income growth between 1960 and 1970. Our
alternative results, including a range of arbitrariness for college level
earners, are shown in Table 4.
Our reestimates of college level incomes involve the following
changes:
(i) Langoni's figure for 1970 (Cr. 1706) is raised to correct for
the Cr.9998 problem: the 1970 census register of incomes was truncated
at Cr.9998 - all incomes above Cr. 9997 (slightly less than 1%) were
coded as Cr.9998. The Table 4 figure of Cr.1749 uses
a Pareto fit to the top 5% of the distribution, which yields a mean for the
top 1% equal to Ct. 14,000.
(ii) For all income classes except the top, we assume an average income
equal to the midpoint; this assumption is most important in the modal income
class (Cr. 20 to 50) since it contains almost half (47%) the distribution.
(iii) The rantge for 1960 college level income (1309 to 1438) is obtained
by using two alternative assumptions for the degree of skewness in the
upper tail, as measured by the ratio of mean income in the open-ended
class, to the tower limit of the class (Cr. 50 in 1960 prices, or Cr. 1769
in 1970 pricea). The low estimate (1309) uses a ratio of 1.8 which
corresponds to Langoni's college distribution for 1970 after correcting
for the Cit. 9998 problem as explained above. The original and corrected
- 41 -
Table 4: EDUCATIONAL AND OCCUPATIONAL INCOME DIm ERENTIALS 1960-1976
Average Monthly Income /Educational Percent of (in 1970 Crs.$)/d Percentage Change -Level La Labor Force's. 1960 1970 1972 1960-1970 1960-
1970 1972 Laborncb *b b /b 7-1960 1970 1972 Icl/- Excl -- Incl/- Excl/- Excl/- Langoni Revised 1972
1. No School 7.5 36.0 (25.3) 97, 111 104 112 150 0 7 352. Primary 49.1 49.0 (56.3) 1i4 211 240 341 14 19 623. Junior High 3.6 7.7 9.8 ' 382 482 592 10 26 554. Senior High 3.4 4.9 5.6 619 688 881 28 11 435. College /f 1.3 2.3 3.1 1373 1749 1935 52 27 416. (Range) (1309-1438) (1749) (1935) (22-34)(35-48)7. Total (millions) 23.2 29.5 33.8
Occupational Percent of Nonagricultural Average Monthly Income Percentage ChangeCategory Labor Force (in 1970 Crs.$)id 1960- 1970- 1960- 1960-
1960 1970 1972 1976 1960 1970 1972 1976 1970 1976 1972 1976
8. Administ. & Technical 23.7 26.3 29.3 30.2 546 712 874 1317 30 85 60 1419. (Technical) (6.6) (8.3) (8.2) (9.5) 619 770 971 1417 24 84 57 129
10. Other Nonagricult. 76.3 73.7 70.7 69.8 219 250 341 461 14 84 56 111
Percentage ChangeIncome Differentials in Income Differential
11. Lines 8. 10 2.49 2.85 2.56 2.86 +14 0 +3 415
12. 9 . 10 2.83 3.08 2.85 3.07 + 9 0 +1 + 8
G_neral Note: As noted in Table 2, the Censuses and the 1972 survey each appear to underreport personalincoie by between 40 to 45%. Also, the questions and procedures used to measure income differ in theCensuses and in each suTvey. The 1972 income data underlying this table exclude transfers and nonmenetaryIncome, whlich increases their comparability with the census data, but also increases undercoverage. The1976 data used here also exclude transfers but include the imputed value of pay in kind in the principaloccupation. The greatest margins for error are probably in the No School and College categories, giventhe importance of nunnonetary income to rural workers, and of entrepreneurial incomes to the moreeducated.
Sources:
1. Censo 1960. Tables 20 and 22 for educational levels. Table 24 for occulpational incomes. 2. Censo 1970.Tables 22 and 25 for educational levels. Table 26 for occupational incomes. 3. PNAD-2 1972. Table 2.6for educational levels. Table 2.4 for occupational incomes. Income is monetary remuneration only. The1972 percent of nonagricultural labor force taken from Table 2.3.9 of PNAD 1972. 4. PNAD 1976.Table 29. 5. Langoni [1973, Table 4.2. p. 86].
Notes:
/a Definition of Educational Levels according to grade l,wel achieved:Primary (1-5). Junior High (6-9),Senior High (10-12), College (13-17) except in 1972 wheni, aftcr a reform of the educational system,College became 12-17. Definition of'Junior and Senior High seems to differ from Langoni's definitioni.Nis 1960 percentages show twice as many at Junior High level (5.2%) than at Senior High level (2.7%).This may explain the reversal of our iiicome trends for Junior and Senior High.
/b "Incl." and "Excl." mean inclusion or exclusion of unpaid family workers. Allocation of unpaid byeducational category was calculated from differen.e between Tables 20 and 22 in Censo 1960, anidTables 22 and 25 in Censo 1970. Imputed nean income for unpaid eqial to that of lowest income class.Adjustment applied only to No School and Primary due to small numbEr of unpaid family workers in highereducational categories and greater margin for error in imputing incomc. 1972 data on unpaid familyworkers by education not available.
/c Revised 1960-1970 comparison includes unpaid family workers, but 1960-1972 excludes them due to dataunavailability in 1972.
/d For 1960 - 1970 we used Langoni's index: 1970 = 3537; 1960 = 100 (thc CDP implicit deflator) tomaintain comparability. For 1970-76 we deflated by Cuanabara Cost of Living Index, with an upwardadjustment in 1973,
/e Includes unpaid familv workers. Also includes a small number of illactive income recipients in 1960and 1970 (0.6% in 1960, 2.6% in 1970). Bracketed figures for 1972 No School and Primary are notcomparable with 1960 and 1970 since, in 1972, the bottom group was defined as Illiterate rather thanby lack of school attendance. In 1972, persons with No Schooilng but who were literate were classifiedas Primary.
/f Range of estimates of 1960 mean College level income explained in text.
- 42 -
ratios are shown in Table 5. The high estimate (1438) uses a ratio of
2.2, which implies greater skew in 1960, but which, as may be seen in
Table 5, is entirely within the orders of magnitude of the various
distributions available for Brazil. Indeed that ratio, at the 18th per-
centile from the top (corresponding to the size of the open-ended class
in 1960) equals 2.5 or more in half the distributions.
(iv) Our point estimate for 1960 college income (1373) is
the midpoint of the range, and thus corresponds to a ratio of 2.0, thereby
giving some credit to the lower ratios shown for the 1970 (revised) and
1972 college level distributions. Langoni's 1960 college income (1123 or
Cr.80 in 1960 prices) is based on a ratio of only 1.6 (Table 5). On the
other hand, there is no compelling reason to assume a ratio as low as that
of the 1970 revised college distribution (1.8): the 1972 college distribution
has a higher ratio (2.1) than our point estimate; also, considerable change
in the shape of the college distribution could have occurred over a period1/
of ten years during which the college earner population more than doubled.
Our estimates in Table 4 also involve a downward adjustment of 1960
incomes in the lowest educational categories.. The revision consists of
reintroducing unpaid family workers, with an imputed income equal to the
wage they could earn in agriculture. Because most unpaid family workers
are young or female workers in agriculture, their (imputed) earnings puts
them at the bottom of the income distributions for illiterates and primary
school leavers. Excluding them, therefore, tends to bias upward the 1960
1/ The use of a Pareto fit to estimate the 1960 mean income in the open
ended class had to be ruled out owing to the large percentage of the
distributions (65%) covered by the upper two income classes. The
Pareto fit is reliable only for much smaller upper tails.
- 43 -
Table 5: SKEWNESS IN THE UPPER TAIL OF INCOME DISTRIBUTIONS: RATIOS BETWEEN
MEAN INCOME IN OPEN-ENDED CLASS AND LOWER LIMIT INCOME, FOR ALTERNATIVE PERCENTILES
Ratio of Mean to Lower Limit Incomeof Top Percentile
Top PercentIndividual Income Recipients 2 5 10 18
1. 1960 Census- 1.8 2.0 2.0 2.12. 1970 Census 2.1 2.3 2.5 2.73. 1972 Survey 2.1 2.3 2.5 2.84. 1976 Survey + Tax
Declarations 1.9 2.1 2.5 2.85. 1972 Survey: College Level 1.9 1.9 2.0 2.16. 1972 Survey: Technicians 1.8 1.8 1.9 2.27. 1970 (and 1960) College
Level: Langoni 1.2 1.3 1.4 1.68. 1970 College Level: Revised 1.6 1.6 1.7 1.8
Households
9. 1960 Fishlow-Meesosk 1.8 1.8 2.0 2.210. 1970 Census 2.1 2.1 2.3 2.711. 1972 Survey 2.1 2.3 2.5 2.712. 1974-75 ENDEF 2.0 2.2 2.3 2.5
Note: Ratios were calculated by graphing each distribution and interpolatingto the above percentiles. The second percentile (top 2 percent) islargely determined by the assumption made regarding the open-endedclass. For Lines 3, 11, and 12, however, published data includedtotal earnings so assumptions were not required.
Line1. Langoni Table 23 page 60, Cash income only.2. Censo Demografico 1970, Table 25. Cash income only.3. PNAD-2 1972, Table 6.1. "Global Income": includes transfers and some
income in kind (see note to Table 1, Line 6).4. PNAD 1976, Table 14. Top second and fifth percentiles taken
from income tax data published by V. Gibbon, Mobilidade Social, p. 31.5. PNAD-2 1972. 1972. Table 2.66. PNAD-2 1972. 1972. Table 2.27. Langoni, Table A1.5 page 257.8. Line 7 corrected at top of distribution by Pareto fit (see text).9. Fishlow-Meesook Technical Appendix, page 54. Adjusted distribution.
10. Censo Demografico 1970, Table 9, page 225.11. PNAD-2 1972 , Table 3.8. Cash income only.12. Calculated from regional volumes of ENDEF, Table 7.
- 44 -
mean income in each of those two educational categories and thus, to1/
underestimate income growth between 1960 and 1970. Stated differently,
their exclusion leaves out of the trend comparison the income growth that
is associated with moving out of family enterprises into paid employment.
The effect of this correction is minor, but it further offsets the initial
picture of widening differentials.
The comparison with 1972 presents a major problem because the
definition of the bottom group changed from "no formal schooling" to
"illiteracy." The sharp drop between 1970 and 1972 (36.0 to 25.3%) in the
No School category must mean that about 1 in 4 employed persons with no2/
formal schooling have acquired literacy in some other way. Those persons
were classified in Primary. The 1972 No School and Primary groups are
therefore less educated than the corresponding categories in earlier years:
the No School category in 1972 contains no literateg, whereas it does in
1960 and 1970; the 1972 Primary includes persons with no formal schooling
but who have taken an adult literacy course or acquired literacy on their
own. As a result, the 1972 mean incomes in the No School and Primary
categories are both biased downward, and the 1960-72 income growth rates
for these categories (35 and 62%) may therefore be higher than those shown
in Table 4.
1/ More precisely: exclusion occurs in both 1960 and 1970, but the biasis greater in 1960 because the proportion of unpaid family workers is
larger.
2/ The 1972 PNAD instruction manual defines as illiterate persons whocannot read, or write more than their own name. The definition forprimary level includes persons with at least one complete year ofschool or an adult literacy course or who can write a "simple note."In 1970, 11% of all persons 5 or more years old with no school were
classified as literate. The degree of literacy amongst the unschooledcan be expected to be higher in the employed population than in thetotal.
- 45 -
The 1972 results suggest a change in trends since 1970, but they
also add to the suspicion that the extent of widening of differentials
over the sixties l-s been exaggerated. The higher growth rates over 1960-1972
reflect better coverage of cash pay in 1972, as well as the rapid growth in
real incomes that occurred between 1970 and 1972. There is no basis for
allocating the improved coverage by schooling category, but the 1972 data
weaken the case for widening differentials. The striking result of the
1960-72 comparison is a substantial and roughly parallel growth of incomes
in the primary and secondary school categories relative to college level incomes.
To the extent that there is any pattern to the trends in differentials it is
one of slight narrowing. The lag by illiterates, however, is an important exception
(though the 1972 figure for illiterates is biased downwards, as pointed out above).
The behavior of occupational differentials supports the views
that widening has been exaggerated, and that trends improved after 1970.
Between 1960 and 1976 the mean income of technical and white collar
workers does rise relative to that of all other nonagricultural occupation,
but (i) the total increase, 15%, is modest; (ii) the increase is even
lower - 8% - for the most highly educated occupations, i.e., technical
workers; and (iii) all of the increase occurs between 1960-1970: there
is no widening after 1970. Furthermore, the small change in occupational
differentials actually suggests a falling premium for college education,
because a growing proportion of college graduates are probably being
forced into lower level jobs within the overall category of white collar
and technical occupations. This occupational downgrading is implied by
the much faster rate of increase in the number of college graduates than
- 46 -
in the number of white collar and technical jobs (12.6% vs. 7.1% per year
1/since 1960). There is also casual evidence of flooding in the market
for many categories of professionals. This trend was noted some time
ago by Pastore and Perosa [1971] and Pastore and Bianchi [1972]. The number of
medical doctors, for instance, doubled between 1971 and 1978 and, in
1978, doctors went on strike demanding a floor wage for hospital interns
of 5 minimum wages and alleging that the hourly wage of new doctors was
less than that of construction foremen working on the Rio subway.-
The relationship between education and inequality is determined
not only by income differentials, but also by the distribution of education
across the labor force. Indeed, independently of its role as a productive
asset, education is a value in itself, and should be studied as a component
in the distribution of welfare. Table 6, therefore, shows the entire popu-
lation aged 5 and over according to grade level achieved, while Table 4
shows the labor force according to schooling level.
1/ According to Censo DemQgrafico 1970, Table 14, 90% of all university
graduates, and 82% of all senior high school graduates were absorbed
by the Administrative and Technical occupational categories. Thoughimperfect as proxies for higher educational classes, these categories
have a sufficient overlap with college and secondary schooling to
serve as tests of the widening differential argument. On the other
hand, these categories contain many persons with lower levels of
schooling: 45% of Administrators and 24% of Technicians had no more
than Primary school. Graduates are defined as persons who completedcollege (grades 16-17) and high school (grades 12-15) respectively.
In Table 4, "college" includes persons who attended one or more years
of college, i.e., grades 13-17, and Senior High comprises grades
10-12. Sixty percent of the Table 4 college group had completed grades
16 or 17.
2/ Jornal do Brasil, July 16, 1978, p. 24.
- 47 -
It is evident that the distribution of education is highly
unequal but that there is a trend towards equalization. Tables 4 and 6
show the impressive expansion of secondary and higher education. The
proportion of earners without schooling fell from 42.5% in 1960 to about1/
28% in 1976. The fall was less dramatic in the case of the total
population, however, and the absolute number of persons without schooling
actually rose around 10%.
To translate these figures into a single measure of equity in
the distribution of education one must first make a subjective judgment regarding
the consumption value of each level of schooling, and then choose some measure
of the productive value of schooling at each level. As a first approximation,
we have applied the simplest weighting system, which is self-weighting_
each grade is weighted according to its cardinal value (e.g., 0 for
those who never went to school, 8 for those who reached eighth grade).
These values are roughly parallel to the income differentials shown in
Table 4, which should be considered a maximum range since they do not
allow for consumption value and are not net of private costs of attendance.
The Gini coefficients so calculated for the total population are 0.69 in
1960, 0.65 in 1970 and 0.58 in 1976.
1/ Between 1970 and 1976 the proportion of earners with no schooling isextrapolated on the basis of the trend in that ratio for (i) the totalpopulation, which falls from 43.6 to 34.9% (Table 5); and (ii) heads ofhouseholds, which falls from 40.5% to 31.3% (Censo 1970, Table 5,p. 214, and PNAD 1976, Table 7, p. 78.)
2/ Experiments with different weighting systems changed the level butnot the trend in the Gini's.
- 48 -
Table 6: POPULATION ACCORDING TO GRADE LEVEL(Persons 5 or more years old)
Cumulative
Number (Millions) Percent Percent
1960 1970 1976 1960 1970 1976 1960 1976
No formal schooling 30.1 34.5 32.1 51.0 43.6 34.9 100 100
Primary: Grades 1-3 16.6 22.1 24.2 28.1 27.9 26.3 49.0 65.0
4-5 8.6 13.3 19.6 14.6 16.8 21.3 20.9 38.7
High School 3.3 8.3 13.6 5.6 10.5 14.8 6.3 17.4
College 0.4 0.9 2.4 0.7 1.1 2.6 0.7 2.6
Total 59.0 79.1 91.9
Sources: Censo 1960, Table 15. Censo 1970, Table 15. PNAD 1976, Table 11.
College level ("Superior") is defined in the 1960 and 1970 Censuses
as grades 13-17 but after the 1971 reform of the educational system,
College level is defined as grades 12-17.
Note: The 1976 figures exclude the rural areas of Region VII (North) or
about 3.7% of the total population.
- 49 -
In conclusion, the data relating education and income provide very
weak corroboratory evidence for the contention that inequality has increased
significantly in Brazil. They do suggest that income growth has lagged
for a poverty core, assuming that it is made up largely of illiterates. On
the other hand, this set of data indicates that income growth in the bottom
40 has not been insignificant: it includes modest improvement within the
category of illiterates, faster income growth for those amongst the poor who
have some primary schooling, and additional income growth through the reduc-
tion in the proportion of illiterates. For those with some schooling, the
trends in educational differentials are not clear; they are blurred by the
uncertainty regarding college incomes: also, a definite, though moderate
deterioration emerges only from the least favorable of our 1960-1970
assumptions. The 1972 data instead imply narrowing differentials between
1960-1972, except for a lag by illiterates (Table 4). The 1972 survey and 1970-76
occupational differentials suggest slower income growth for college graduates
since 1970. Other statistics on wage differentials - largely limited to
the urban modern sector - are discussed in the next section. But the more
general evidence from Census and survey data do not support the picture of
sharply increasing differentials during the sixties that are so often cited
in the literature, and in fact suggest a reverse trend during 1970-1976.
2. Urban Modern Sector Wage Differentials: Law of 2/3
Another set of statistics that are frequently cited to
corroborate as well as to explain growing inequality, are wage and
salary data for the modern urban sector. Two types of statistics
are involved. The first are chiefly the earnings figures for individual
- 50 -
workers reported by urban establishments under the Law of 2/3 and
available for the years 1965-1974. Also,the 1959 Economic Censuses
provide earnings data aggregated by income classes. These statistics
represent a large category of earners and lend themselves to the calcu-
lation of aggregate measures of dispersion. The second type of statistics
are occupational wage differentials derived from small samples of closely
specified occupations. Besides their use to support statements about
growing overall inequality, references to these statistics are generally
also part of an attempt to attribute that deterioration to either
or both a wage squeeze through wage policy, and the profits-managerial
salaries link referred to above.
Arguments based on these statistics suffer from two limitations
that have been largely overlooked in the literature. The first is limited
coverage: for most of the period the Law of 2/3 (the most comprehensive
of these data sets) covered less than 20% of employment, and less than
16% of personal income in Brazil. The second consists of problems of data
consistency and quality, particularly the bias introduced by growing
coverage. These limitations are examined more fully below, after a
description of the data and the way in which th7ey have been used.
The most comprehensive of these data sets is a consequence
of the Law of 2/3. The Law of 2/3 requires firms to employ a work force
in which two-thirds of the employees are Brazilian nationals. Under
this law, private sector firms in manufacturing, commerce and service
activities in urban areas are required to report to the Ministry of Labor
data on employment and earnings for the month of April of each year.
- 51 -
In practice, coverage of smaller establishments is
partial. Comparisons of the Law of 2/3 earnings distributions have
been carried out by Wells [1974], and Morley [1976]; Morley's com-
parison was cited by Fishlow [1977]. Wells extends and supplements
the comparison by including the 1959 Economic Census results, and
1966 to 1970 earnings data from industrial establishment surveys.
Both Morley and Wells find increasing dispersion over time
in the earnings data. Wells calculates Gini coefficients for 1959
and 1965-1971. The Ginis show a general upward trend, with the biggest1/
increase occurring during 1965-1966. For those two years he also
presents decile shares showing sharp losses in the bottom decile, and2/
the biggest gains in the top two deciles. Morley compares only 1969
and 1973. He presents the frequency distributions for all earners
according to the Law of 2/3 in those years. Our calculations based on
those data show that the top decile share rises from 37.0 to 41.5%
while the bottom 40 share falls from 16.0 to 14.0%.
1/ To obviate changing sectoral coverage, Wells makes separate butoverlapping calculation for two and three year periods, but thechain is broken, between 1968 and 1969, by a change in the timingof the minimum wage adjustment. (The degree of dispersion inearnings follows a cycle and is lowest immediately after a minimumwage adjustment.) A chain index of Wells' Ginis for combinedindustrial, commerce and service sector earnings is 1959=100,1965=101.8, 1966=114.7, 1967=116.2, 1968=119.3. Then,startingagain,the index is 1969=100, 1970=104.4, 1971=104.9. Excluding theunknown change between 1968-1969, the total increase in the Ginibetween 1959 and 1971 is 25.1%, of which about half occurs between1965-1966.
2/ The share of the bottom decile falls 44% in industry, and 30% incommerce and services; the top 2 deciles gain 10% in industry and5% in commerce and services.
- 52 -
In interpreting these results, Wells is especially concerned
with the timing of the increased dispersion. He argues that the
especially strong increase in 1965-1966 conflicts with the human
capital supply and demand interpretation of growing inequality, and
instead supports a wage squeeze explanation: "....... the major increase in
inequality can be ascribed to the period of most severe wage stabili-1/
zation, 1965-6." Morley notes that the "substantial increase in earnings
inequality" was accompanied by considerable job creating at middle
and high skill levels but that "there was virtually no real salary
increases for those who remained at the bottom." Both Wells and Morley
analyse these partial data sets side by side with comments about aggregate
inequality, without any explicit statistical consideration of the degree
of coverage of their data, and without discussing the necessary assump-
tions about associated changes in the rest of the economy that would
allow one to infer that an increase in dispersion within the urban modern
sector implies increased overall inequality. Neither author provides
data on the number of employees or on the amount of income covered.
The coverage of the 2/3 Law data is summarized in Table 7. In
1965, the initial year of Wells' analysis, it amounted to only 12.5% of
the labor force and 10% of personal income in Brazil. In 1969, the initial
year of Morley's comparison, the figures were 18.i and 15.2% respectively.
Even in relation to the 1969 nonagricultural sector, the 2/3 Law covered
1/ Wells (1974), p. 16.
- 53 -
Table 7: COVERAGE OF LAW OF 2/3 AND 1959 ECONOMIC CENSUS
(1) (2) (3) (4) (5) (6) (7)Reported Employment Reported Earnings Annual Earnings per Worker
(Millions) Billions Percent of ('000 and Cruzeiros)Percent Current Total Personal Total Ratio
Number of Total Cruzeiros Income Reported Personal (5) -(6)Labor Force Income
1959 2.41 (2.13) 10.8 (9.6) 0.21 11.6 0.087 0.082 1.07
1965 3.24 12.5 3.56 10.0 1.13 1.37 0.82
1966 3.46 13.0
1967 3.85 14.1
1968 4.51 16.1 13.6 14.1 3.02 3.45 0.88
1969 5.30 18.5 19.5 15.2 3.68
1970 5.73 19.4 26.4 4.61
1971 6.03 19.8 33.4 5.54
1972 6.88 22.1
1973 7.62 23.7 65.7 16.7 8.62 12.3 0.70
1974 8.80 26.7
Notes:Sources: Reported data from 1959 lndustrial Census and 1959 Commerce and Services Census;Atualidad e Estadfstica Brasil 1969; Anuario Estadfstico do Brasil 1972; Morley [1974].Cols. (1) & (2): Figures in brackets are coverage of data actually used by Wells, afterdeductions of subsectors for comparability. Reported employment corresponds to employees whoseearnings were reported in Table 5s Censo Industrial and Table 7. Excludes proprietors and familyworkers.
Col. (2): Total Labor Force according to 1960 and 1970 Censuses and interpolation. Use of PNAD laborforce definitions which result in higher participation rates, would yield even lower coverage, e.g.,
19.6Z for 1972 vs. 22.1Z.
Qol. (3): 1959 Censuses: Total wages and salaries paid in 1959. Other years: April earnings X 12.Col. (4): Total Personal Income defined as in Table 2 (Note 6).
- SI. -
only 1 in 3 employees and 19% of personal income. Because the number of
earners reported grew rapidly over the whole period, coverage grew, but
reached only 16.7% of personal income in the final year of Morle,'s comparison.
Any change in distribution within the 2/3 Law data is therefore a small part
of the income distribution story.
Aside from its limited coverage, the 2/3 data is liable to
significant biases. The most important is the result of growing coverage:
Between 1965 and 1973 the number of workers reported grew by 11.3% per year
and, as a proportion of the total labor force, coverage almost doubles.
It seems likely that most of the growth came from the incorporation of
smaller firms and workshops where wages are typically lower. Wells was
aware of this potential source of bias: "improved coverage should lead
to an increase in recorded dispersion," but he goes on to dismiss it: ".....
1/
there is no strong indication that coverage has improved." This dismissal
is surprising given the rapid growth shown in Table 7. Coverage grew 58%
over the period (1965-71) compared by Wells, and 29% over the shorter
period (1969-73) studied by Morley. The most likely result was a large2/
relative increase of the population at the bottom of the distribution.
1/ Wells [1974], p. 12. He does not cite the numbers that led to thisjudgment,
2/ The expectation that coverage is better amongst larger, higher-wagefirms is supported by Langoni who finds that the Law of 2/3 significantlyunderrepresents employees who are illiterate, women, under 20 years old,and resident in the North and Northeast regions. Distribuia-o da Renda,pp. 272-274.
- 55 -
Much of the increased dispersion, as measured by Wells and Morley, may
be explained by this change in the size and composition of the group
being studied. More obviously compromised is Morley's conclusion with
regard to the 1969-1973 comparison that "Wages at the very bottom actually
fell, those near the bottom rose .at a slow rate, and the number of people
1/earning those wages rose as well." All these trends are likely con-
sequences of growing coverage of lower-wage employees.
Wells points out a second possible bias but dismisses it, again
erroneously, in our opinion, it is more than likely that employers
will overstate wages paid at' the lower end of the scale, so as not to2_/
be seen to be breaking the minimum wage legislation." But, he sees "no
reason why, over time, the degree of bias imparted to the results should
3/be different." A reason does exist, however, in the reduction of
the real minimum wage during the sixties. Finally, the difficulty of inter-
preting the data is increased by the fact that the incomes reported
are actual earnings during April, rather than monthly pay rates. Earnings
differ from monthly rates for-many obvious reasons: overtime, piece
rates, sickness, part-time work and high turnover, especially in the lower
occupational categories. It is no surprise therefore that, despite
inhibitions about reporting wage rates below the minimum wage, about 1 in
3 workers in these data have monthly earnings below the minimum wage.
1/ Morley [19761, p. 33.
21 Wells [19741, p. 12.
3/ Wells [19742, p. 12.
4/ An index of the real legal minimum wage (deflated bv FGV Guanabara cost of
living index) is 1959=100, 1965=79.4, 1966=73.4, 1969=68.5, 1971=67.2.The real minimum fell by 10% between 3/65 and 3/66, the period that
registers the largest increase in inequality.
5/ Wells [1973], p. 12.
- 56 -
The number involved is too large for a ceteris paribus assumption applied
to trends. The increase in coverage, for instance, could have led to
major changes in the composition of the sample, with different behavior
with respect to frequency of part-time work, piece-rates, turnover, etc.
In sum, though it is difficult to quantify the incidence of these various
potential biases, there are at least two sources of bias acting in the
direction of increasing measured inequality. The growth in coverage, in
particular, is large enough to cause the observed increase in dispersion.
Limited coverage reduces the force of the argument, advanced by
several authors, that much of the growth in income concentration can be1/
attributed to rising managerial salaries. By its very nature, this
argument is about an extremely small group of persons - our ceiling2/
estimate is 1% of the labor force. A larger number would conflict
with the character of the argument, which turns on the link between
senior management salaries and profits, and on the power of senior govern-3/
ment officials to set their own salaries. Assuming that all private
1/ Bacha [1974(a)]; 1974(b)], Fishlow [1973], Wells [1974].
2/ Or about 300,000 persons in 1970, of which about one-third are ingovernment and therefore not covered by Law of 2/3 data. This isan order of magnitude estimate based on following: (1) ENDEF reports250,000 heads of families who are employees in "Cargos de NivelSuperior" defined as employees with high-level responsibility requiringuniversity education or equivalent. Adjust upwards to 300,000 to allowfor non-heads of families and for North Region. (2) 1970 Census:323,000 persons in Administrative, Technical and Defense occupa-tions (excluding proprietors) earning 8 or more minimum wages. Thisis a generous definition because senior managers normally earn more.(3) Law of 2/3 data: in 1969 2.6% of earners, or 138,000 personsearned 9.2 or more minimum wages. Criteria (1) to (3) are clearlyoverestimating the concept of "senior managers" as implied by itsproponents, i.e., the true proportion is therefore well under the 1%figure used here. A fourth criterion yields a much lower figure(4) 1970 Industrial Census: 11,000 establishments with 20-49workers, plus 10,000 with 50 or more workers. Assuming 1 seniormanager per establishment in 20-49 class and 3 per establishment in50+ class yields only 41,000 senior managers in Manufacturing.
2/ Wells [1974], p. 16. "It would be more realistic to see these higherpaid employees, especially when they are executives, as paying them-selves relatively higher salaries in spite of, and possibly as aconsequence of, strict control of the wages of manual workers."
- 57 -
sector managers are covered by the Law of 2/3 data, and that they are
at the top of the distribution, we can calculate that their share of
total personal income in Brazil rose from about 4.0% in 1969 to 4.7%1/
in 1973. The amount involved is minor, both as an absolute share and
as an increase. Managerial salaries may have risen much more than is
reflected in these data,through unreported payments in kind and bonuses,
and there is anecdotal evidence to that effect. But this cannot be
documented with the Law of 2/3 data. Those statistics do not support the
contention that the relative enrichment of the managerial class explains
much of the worsening income distribution. In fact, they weaken the
argument by establishing that only a relatively small proportion of total
income is involved. This negative conclusion applies as much to the
argument that relative enrichment was caused by the pulling up of high
salaries as to the argument that it resulted from compression of wages
at the bottom: either mechanism is summed up in the small percentages
noted above. If the wage squeeze did indeed cause a major redistribution
of total personal income, it is not demonstrated by the earnings data
reviewed here. These data imply that redistribution was minor,
1/ Assume 350,000 senior managers, in 1973. This equals 4.6% of allreported earners. Their share is given by Morley [1974], Table 11,as 28.0% of total earnings. Using our Table 6, Col. (4), this equals(0.28 X 0.167) 4.7% of Personal Income. Assume a growth rate of5.0% per year over 1969-1973: (slower than growth of all reportedearners due to growing coverage.) Number of senior managers in1969 therefore = 288,000 or 5.4% of 1969 reported earners. Theirof Personal Income is given by same tables. If this method is appliedto the Law of 2/3 data for 1965 calculate the share of senior managersin that year the result implies a large fall in their share between1965 and 1969.
- 58 -
or that, if significant concentration did occur, it took the form chiefly
of a transfer from labor as a whole to profits, rather than of redistribu-
tion within labor.-/
1/ We have not reviewed the possibility of a major shift in the functionaldistribution. Such a review would require a detailed examination ofthe national accounts income estimates. Wells downplays this pos-sibility [Wells 1974, pp. 10-11]. An analysis within the manufac-turing sector was carried out by [Macedo 1977]. The national accountsshow relatively little change in the share o:- labor incomes in totalurban income; it falls from 55.4% in 1959 to 52.5% in 1975. (ContasNacionais, Revisao e Atualizacao 1977, Table IX). Macedo finds a fallin the manufacturing labor share between 1960 and 1970, from 26.1 to23.1%, but argues that the measured fall may disappear if growingsocial security payments are taken into account. Manufacturing in anycase accounted for only 31% of urban income in 1970. The nationalaccounts functional income data, however, are subject to large error,and in fact, were dropped from the last publication of Contas Nacionaisin 1978, pending the results of a re-examination.
- 59 -
3. Urban Modern Sector Differentials: Wage Surveys
The argument for growing inequality is also frequently
supported by references to a number of wage surveys. The most commonly
cited have been those published by Suplicy [1974; 1978] and Bacha [1974].
Both cite several wage surveys carried out by business associations1/
and consulting firms. These surveys appear to be limited to a rela-
tively modern, large-firm subset of the manufacturing industry. Two
additional studies, based on larger samples, have been published recently.
One, by DIESSE [1979], is based on wage data obtained from Sao Paulo2/
unions. The second, by Barbosa [1978] uses Law of 2/3 data. Most of
these data, summarized in Table 8 show growing differentials, and
some of the figures have been used as dramatic illustrations of rising3/
inequality. Thus Suplicy notes that "In 1969, the mean wage of the
general manager of a medium or large firm in Sao Paulo and Rio de Janeiro
was 65 times greater than that of a helper (servente) in civil construc-
tion in Sao Paulo; in 1973 it was 81 times greater and in 1975, 90 times
1/ The sources cited are (i) GRUPISA: a Rio based consortium of firms;biannual surveys of about 180 job categories in some 50 large firms.(ii) CSN: Personnel Department of National Steel Company, data for15 large firms. (iii) GE: Payroll Department of General Electric,survey managerial salaries in about 65 large competitors in Rio andSao Paulo. (iv) Morris and Morgan: Sao Paulo consulting firm, surveyof managerial salaries. (v) PRIL: Industrial relations firm: surveyof about 100 manual and administrative job categories in Sao Paulo.
2/ DIESSE (Inter-Union Department for Statistics and Socio-Economic Studies),Distribuicao Salarial em Sao Paulo Segundo Guias da Contribui,cao Salarial
(Sao Paulo: December 1977). Data drawn from union fees, data for asample of Sao Paulo unions. Union contribution is obligatory, butreporting is incomplete. Data for 1956, 1961, 1966, 1971 and 1976,with skill breakdowns.
3/ Morley [1976], p.28, reproduces the PRIL and Morris and Morganstatistics published by Suplicy referring to them as "startling data"which "deserve far wider circulation."
- 60 -
Table 8: TRENDS IN MODERN SECTOR WACE DIFFERENTIALS: SELECTED DATA
IncomeRange
(in .'971/72 Index of Real WageMinJl:Lm 1961 1966 1969 1970 1971 1972 1973 1974 1975 1976
_______ Wages)
CSN Sutrvey
Unskilled 1-2 100 92
Skilled 2-7 100 120
Marnagern 14-32 100 153
San Paulo Industar (PRIL andMorris and Morgan)
Low 1-4 100 106 110 107 107 110 116 137
Middle 4-9 100 109 114 120 123 132 135 151
Middle 9-19 100 105 117 128 129 133 142 151
High 19-66 100 118 127 136 142 150 163 178
Very High 66+ 100 123 128 137 148 148 182 190
Rio: Large Fir.vo (CRUPISA)
Skilled 2-6 100 115
Rio and Sao Paulo (GE)
Managers 13-48 100 117
Sno Paulo Induitrly (DIESSE)
Unskilled 1-2 100 70 59 58
Serni-skilled 1-3 100 74 86 76
Skilled 2-5 100 87 94 108
Forerunf- 3-1l 100 73 104 102
Manager:c. 5-i- 100 85 11i 122
Sano l1'euo l..>w ol 213 (Barbosa)
Unskilled -1.4 100 107
SBon ri,.J-'Ied 1.5-2.6 100 100
Skilled and Whitc Collar 3.5-9.0 100 89
Frofessionnls aind Managers 11-26 100 93
Souirces
CSN: Bacha [1976]. Indices are weighted averages for different occupations calculated frori Table 1.
Pril and Morris ;i-d Morgan: Suplicy [1977], p.l4 6
. Index is unweighted average..
GRUPISA and GE: Bacha [1976]. Indices are unweighted averages.DlESSE: 1)iWriLuicaoSalarial em Sao Paulo segunedo (hnias da Contribuiecao Salarial;" Sao Paulo: Dec. 1977. Deflated by
DIESSE price index.Barbora, Flilton, "Diferenciais de Snlarios entre 07.upacoes: Uma Analise das Variacoes Ocorridas no Periodo 1971-1974,"
tltiiversidade de Sao Paul.o, Tese de Mestre enr Econo-nia, Sao Paulo, 1978.
Note
/a Defined to cover 70% or more. The sources cited above use different deflators. DIESSE's deflator grows especial]y
fast; hence the much lower grouth in real wages. See below Table 15.
- 61 -
greater. Including the additional benefits received by a general
1/manager, the differential in 1975 rose to close to 150 times." tHow
meaningful are such comparisons for the question of aggregate inequality?
A first observation about comparisons between servente and
managerial wages is that very few household heads in the modern sector
are serventes or unskilled workers. Of all blue-collar household heads
employed in manufacturing and construction in Belo Horizonte, only 14%
were classified as unskilled, in a 1972 survey; 86% were skilled
workers or foremen. A nutrition survey in Sao Paulo classified 20%
of all (including informal sector) blue-collar fathers as unskilled,
and only 11% of the entire city sample of fathers were unskilled.
PNAD 1976 data for male employees in Sao Paulo manufacturing and con-
struction provide indirect support for this argument: 40% earned 2
21/minimum wages or less but 35% were aged under 25. These groups almost
3/certainly overlap strongly. Since only 7% of male household heads
were aged under 25, the large majority of employees who were heads must
have been earning over 2 minimum wages, implying skilled occupational
status. Further support comes from ENDEF: in 1974/75, the average household
income for manual employees in the State of Sao Paulo equalled 5.7 minimum
wages, implying that the heads of those households were earning between 4
4/and 5 minimum wages.- Heads employed in the modern sector probably earned
5/more.- By contrast, wages for unskilled workers are generally between
1 and 2 minimum wages (Table 8). Part of this difference must result
1/ Suplicy [1977], p. 143. A similar observation is made by Paulo RenatoSouza in, La Segmentacion? del Mercado de Trabajo Urbano y las Disparidadesde Salarios en Economias Subdesarrolladas, (Santiago: PREALC, 1977), p.27.
2/ PNAD, 1976. Regiao II. Tables 21 and 30.
3/ As suggested by the age-earnings profile shown in PNAD 1976 (Table 28).
4/ ENDEF, Regiao II, Sao Paulo, Table 6.
5/ In 1974, the average wage of production workers in manufacturing firmswith 50 or more workers was 50% higher than that in firms with 5 to 19workers, according to PescuisajludAustlDlil 147-4, FIBGE.
- 62 -
from earnings from second jobs, and part from supplementary pay, in
cash and kind, not included in surveys such as those reported in Table 8.
But the large gap between unskilled wage rates and average earnings
according to ENDEF corroborates the survey findings cited above, that
few modern sector household heads are unskilled. The significant wage
from the point of view of poor households, therefore, is that for skilled
manual workers; the unskilled wage rate is of secondary relevance, since
it applies almost entirely to second earners, mostly young entrants into
1/the labor market.
The relevance of the wage trend for managers must also be
questioned, not only because so small a fraction of the population is
involved, but because some of the increase in managerial wages is probably
attributable to a change in the nature of those positions. Managerial
positions are defined by the amount of responsibility they entail and,
in the case of senior executives, that amount is proportional to the size
of the firm. It is not clear whether the surveys reported in Table 8
have adjusted for the growth in firm size and degree of responsibility
associated with top executive positions. This bias probably does not
affect middle and lower level managerial positions whose degree of res-
ponsibility tends to be held constant as firms expand, with a growth
instead in the number of such positions. In the case of Table 8 data,
this possible bias is most likely to apply to the "very high" salaries
reported by Suplicy from a Morris and Morgan survey, covering positions2/
with annual salaries over US$60,000.
1/ The average household in Sao Paulo has 1.6 earners (poorer households
usually have fewer earners). If the principal earner is a skilled
worker earning 3.5 minimum wages, the contribution of secondary workers
earning 1 minimum wage will on average amount to about 15% of family
income.
2/ In 1978 dollars, or, Cr.35,000 per month in 1975.
- 63 -
A more meaningful wage differential therefore is that between
skilled workers and managers excluding the most senior. The data in
Table 8 are inadequate for a reliable test of this differential:
sampling procedures in all cases are unclear and may not be. appropriate
for trend analysis in the case of the public relations firm surveys,
periods covered differ, and only Barbosa's data appear to extend
beyond the largest and/or most modern sub-set of manufacturing firms.
The coverage involved is thus smaller and more "elite" than the
Law of 2/3 data reviewed above, and may therefore not represent trends
even within the whole of the urban modern sector.
Examining Table 8 with these qualifications in mind, the
differential between managerial and skilled manual wages rises as
follows: CSN survey (1966-72) 27.5%, PRIL survey (1969-76) 17.9%,
GRUPISA and GE surveys (1966-72) 1.7%, DIESSE (1961-76) 13.0%, Barbosa
(1971-74) 4.8%. The more systematic DIESSE study also covers the
longest period, and shows all the increase occurring between 1966-1971.
PRIL, Barbosa and DIESSE all show slower rates of widening during the
seventies. The DIESSE results are confused by the fact that foremen's wages
grow less than those of skilled manuals, while the semi-skilled category,
which overlaps somewhat with the skilled category in other surveys, lags
considerably. The Barbosa results, though they span only 4 years, add to
the confusion, by showing unskilled wages growing faster than managers',
and furthermore, a falling real wage for managers during the fast growth1/
years 1971-1974. Both results directly contradict the PRIL and Morris
1/ When deflated by the adjusted Rio cost of living index. Even DIESSEwhose deflator grows more rapidly, shows a rising real managerial wagebetween 1971-1976.
- 64 -
and Morgan datt cited by Suplicy. The overall impression, however, is
of support for the view that the wage gap increased, though the degree
of widening appears to be less dramatic than implied by most authors.
The question of how much is unsettled owing to the poQr quality of the
statistical evidence involved.
The evidence of widening within large scale manufacturing is
plausible, and there is a considerable literature offering competing
explanations. These arguments have been reviewed in a survey article
by Bacha and Taylor [1978]. One aspect of the trend in factory unskilled
wages that needs to be stressed here, because it bears on the question
of overall inequality in the economy, is the relationship between the
unskilled wage in large scale manufacturing and in the rest of the urban
economy. The stagnation or, possible fall in the factory unskilled
wage rate during the sixties is very likely a direct result of the falling
minimum wage. But a necessary condition for this policy impact was a
factory wage in 1961 that was relatively high in relation to the market
wage for unskilled labor. In 1959, 56% of all employees (white and blue
collar) reported by the Censuses of manufacturing, commerce and services
were earning one minimum wage or less. Since white collar, foremen,
and skilled and semi-skilled manual workers comprise at least three quarters
of total reported employment, the unskilled (largely serventes and ajudantes)
must have been earning well under one minimum wage. Moreover, urban
employees not reported by the economic censuses (almost half of total
urban wage employment) probably earned less. By contrast, the 1966 servente
wage in those firms surveyed by CSN (Table 8) equalled 1.4 minimum wages,
while that in the other unskilled category, ajudantes, equalled 1.8 minimum1/
wages. The average unskilled wage in establishments surveyed by DIESSE.
1/ Bacha [1976], Table 1, p. 120. Calculated from Bacha's table byapplying his 1966-1972 (negative) growth rates to 1972 mean wages.
- 65 -
equalled 1.2 minimum wages in both 1961 and 1966.- Both CSN and DIESSE
are therefore representative of a subclass of unskilled workers that
were relatively privileged in the early sixties: their wages were pro-
bably about twice the average urban unskilled wage rate. This is not
surprising since the large-scale, unionized firms surveyed by DIESSE
and CSN were the firms least likely to evade the minimum wage legislation in the
early sixties.
The narrowing differential between the legal minimum
and the going market wage for unskilled labor made it possible for large
firms to hold or lower their entry wage during the sixties, but there is
no evidence that this occurred for the majority of unskilled workers who are
employed outside the protected sector. Also, this process appears to have
reached its limit toward the end of the decade through the combined effect
of accelerated growth, and a narrower modern sector wage premium. During
the seventies, the minimum wage fell below the market wage in most urban
unskilled employment. By 1976, 93% of all nonfarm employees in the state3/
of Sao Paulo reported earnings greater than the legal minimum.
Thus, the policy impact of minimum wage policy on unskilled wages in the
urban modern sector during the sixties cannot be extrapolated to wages
outside this relatively small (though highly visible) category of employ-
ment, nor to situations in which minimum wages can be overruled by the
market, as they were after 1970.
1/ DIESSE [1979].
2/ An alternative hypothesis is that these firms claimed to be paying theminimum when in fact they were paying less. If true, this would implythat part of the reported widening is artificial.
3/ PNAD 1976, Table 27. Figure excludes domestic servants whose payin kind is not included in the earnings data, and a small number (2%)of part time workers earning the minimum or less.
- 66 -
Suplicy's illustrative data on the range of incomes within the
modern sector also bears comment. Any statement about range is very
arbitrary unless it is interpreted in the most absolute sense: the
range" is highly sensitive to the extent of aggregation, i.e., 'rhether
one takes the top 0.1, 1.0 or 10%. The closer to the
absolute end-points, the less interesting is the comparison, and also
the less reliable as a measure of shape of the rest of the distribution.
A somewhat more meaningful measure of range can be obtained from ENDEF
data on occupational incomes. ENDEF distinguishes a category of "high
level" employees which, in the State of Sao Paulo, comprise the top 4%
of the occupatidnal hierarchy. The range between their family income
and the average for all non-agricultural manual workers in the State is1/
6.1 times (Cr.$150,479 vs. 24,606). If the comparison is made, not with
all manual workers, which are a large category (65% of families), but
with the poorest decile of all urban families in the State, the range is
21 times (Cr.150,479 vs. Cr.7,000). Even these are significant dif-
ferentials, though of a very different order of magnitude from Suplicy's2/
illustrations.
Despite the attention that It has received, the evidence of wage
stretching reviewed in this section is limited to large scale manufacturing,
and is therefore of minor relevance from the point of view of aggregate
inequality in Brazil. In 1970, factories with 50 or more employees employed
only 1.7 million workers, or about 5% of the 31.5 million income recipients
in that year. The factory wage bill amounted to Cr. 9.9 billion or about
1/ Corrected for overestimation of savings and therefore income as
explained in text p. 14.
2/ ENDEF. Region II, Table 6. The mean income for the poorest urban
decile is obtained by interpolation.
- 67 -
6% of personal income in Brazil. Furthermore, the forces acting
on wage structure in that sector, particularly wage policy and the effects
of large and growing organizations, cannot be generalized to the rest
of the economy. There is a need, therefore, for a much broader examination
of trends in differentials and their underlying mechanisms.
- 68 -
4. Regional .tnd Sectoral Income Differentials
A major support for the notion of growing inequality in Brazil
has been the -isible poverty and backwardness of the Northeast, and of
rural areas in general. This support has persisted as a popular notion,
despite e4idence fbom both the national accounts and the 1960 and 1970
censuses that regional ner capita income differentials were converging1/
during the fiftie and -ere relatively constant during the sixties.
Langoni finds al-i.tn r change in the relative per capita income of the
Northeast between 1960 and 1970: Northeast per capita income grew 34%
2/
vs. 37% for Brazil as a whole. Similarly, the national accounts report
an almost constant differential between relative income per capita in the
Northeast and in all Brazil between 1960 and 1970: the differential
3/grows 2% from 2.13 to 2.17. Graham and Merrick calculate an "aggregate
4/
measure of regional inequality" that improves slightly during that decade.
In Table 9 we-present additional evidence,including data through 1977,
that challenge the common presumption that per capita income growth was
concentrated in urban areas and in the Southeast.
1/ See Graham ai;d Mer-ric;; r1979h, Chapter VI. Also John Redwood III,
"Evolucao recente das disparidades de renda regional no Brasil,"
Pesquisa e Planejamento Economico 7:3 (Dec. 1977), pp. 485-549; and
Thompson Alineida Andrade "Desigualdades Regionais no Brasil: Uma
Selecao de Estudos Empiricos," Pesquisa e Planejamento Economico 7:1
(April 1977), pp. 205-226; Langoni [1973], pp. 84 and 113.
2/ Langoni [1973], Table 4.1. His statistical decomposition of the
sources of the increase in inequality finds that, holding other
factors constant (age, sex, education and activity) the regional
variable acts to reduce aggregate inequality (pp. 112-113).
3/ Based on national accounts data from Conjuntura Economica and reported
by Graham a-id Merrick [1979], Chapter VI.
4/ Grahamn and Merrick [19791.
- 69 -
Table 9: TRENDS IN REGIONAL AND SECTORAL INCOME DIFFERENTIALS
1959 1960 1966 1968 1970 1972 1974 1976 1977
Regional per Canita Tncomea
1. Urban - Rural: Brazil 2.17 2.74 2.32 (2.93) 2.332. S. Paulo , Northeast 2.11 1.98 2.64 (2.53) 2.003. S. Paulo Northeast (Urban) 2.00 1.93 1.96 (2.22) 1.964. S. Paulo . Northeast (Rural) 1.86 2.15 2.83 (2.55) 2.175. South Northeast (Rural) 2.11 1.98 2.64 (2.53) 2.006. Brazil , Northeast 1.76 1.80 1.93 (2.02) 1.84
Regional-Sectoral Wages
7. Construct.Unskilled - Rural Casual 1.56 1.49 1.43 1.10 1.15 1.168. Idem: Sao Paulo 1.30 1.20 1.12 1.03 0.88 0.819. Urban Minimum ,- Rural Casual 1.72 1.72 1.64 0.96 1.03 1.09
10. Rural Casual: S.P. and South . NE 1.71 1.81 1.81 1.52 1.58 1.5711. Minimum: Southeast 9 NE 1.65 1.65 1.64 1.50 1.47 1.41 1.41 1.4112. Construct. Unskilled: Rio & S.P. - NE 1.59 1.7413. Construct. Skilled: idem 1.59 1.55
Occupational - Sectoral Wages
13. Construct.' Skilled ,' Unskilled 1.97 1.91 1.92 1.68 1.64 1.6014. Manufac. (20+): Non-plant. Plant 2.37 2.69 2 46 2.5915. Manufac.: Large (20+) Small (5-19) 1.40 1.79 1.72 1.41
16. Manufac.: Plant (20+) , Small (5-19) 1.16 1.47 1.35 1.1017. Govt. , Private (Nonag. Employees) 1.64 1.64 1.41Land Rent
18, Cropland: NE . Rio and South 1.00 0.9619. Pasture: NE 1.00 1.11
Value Added per Household
20, Nonagriculture , Agriculture 5.38 7.70 5.11
Line
1-6: Sources: Langoni (1973) for 1960 and 1970; PNAD 1972, Table 6.1; ENDEF; PNAD 1976 (Table 26). See Notesto Tables 1 and 2 for coverage. All data are for individual recipients except ENDEF (bracketed) whichare for households.
5: South equals Region III Sul (Parana, Sta. Catarina and Rio Grande do Sul).
7-13: See Appendix on Wage Statistics.
14-16: Censo Industrial 1960; Censo Industrial 1970; Pesquisa Industrial 1972; Pesquisa Industrial 1974.All by FIGBE. Plant workers are "Pessoal ligado a Producab:" covers plant engineers and supervisors,skilled and unskilled plant workers. Non-plant includes admin.strative and service workers. Thisbreakdown was calculated here for firms with 20 or more worker:; only because of suspected poor qualityof this statistical distinction in smaller firms.
17: 1960 Census, 1970 Census and 1976 PNAD.
18-20: Conjuntura Economica, June 1977, p. 101. States weighted by agricultural labor force.
- 70 -
Between 1960 and 1976 per capita income in the Northeast kept1/
pace with Sao Paulo (Line 2) and Brazil as a whole (Line 6). This overall
performance shows up in both urban and rural areas. The Sao Paulo - North-
east urban differential is remarkably constant over the five different
years (and sources) shown (Line 3). As might be expected, the rural dif-
ferential is more variable: Northeast rural income lags behind rural income
in Sao Paulo (Line 4) but grows faster than the South (Line: 5). IOependett
support for these results is provided by trends in regional wage differentials:
both agricultural and skilled construction wage rates (Lines 10 and 13) rise
more rapidly in the Northeast than in the South and Southeast, between 1968
and 1977, though unskilled construction wages lag (Line 12). Finally, the
trend in agricultural land rents is consistent with the income and wage
data: the price of both rented cropland (Line 18) and pasture (Line 19)
2/rose at almost the same rates in the Northeast and South. -
The overall urban-rural income ratio rises between 1960-1970, but
then falls between 1970-1976 to a level only 7% higher than in 1960 (Line 1).
Some improvement in the reporting of farm income in kind in the 1972 and
1976 surveys may contribute to this result, but coverage of urban cash
income also rose and there is no basis for knowing whether coverage improved
more in rural or urban areas. Wage statistics support the Census and PNAD
1/ The small changes in the ratios - a lag of 5% by the Northeast relative
to Brazil as a whole, a gain of 5% relative to Sao Paulo - are not
statistically significant, in view of the deficiencies of census and
survey income data discussed above, and in relation to the doubling
of income per capita over the period.
2/ In real terms the rental price of land in the Northeast rose 87% for
crops and 18% for pasture.
- 71 -
income data: the gap between urban and rural unskilled wages falls from1/
56% in 1968 to 16% in 1977 (Line 7). Additional evidence is provided
by national accounts and demographic data. These show a net decline,
between 1959 and 1976/77, of the ratio between nonagricultural and
agricultural productivity (Line 20). The ratio rose substantially bet-
ween 1959 and 1970, but the increase was more than offset after 1970 by
a combination of real output growth, improved farm terms of trade, and
continuing rapid rural outmigration.
One measure of sectoral inequality, is provided by the
differential between the average wage in factories (20 or more workers) and
workshops (5-19 workers) (Line 15). This ratio, which can be interpreted
as a differential between protected and unprotected sectors, grows between
1960 and 1970, but then declines to its original level in 1959. This pat-
tern matches evidence from educational and occupational differentials
that some widening did occur between 1960 and 1970, but that this trend slowed
or was reversed after 1970. Another statistic that fits this pattern is the
wage ratio between non-plant and plant workers in factories (Line 14), which
follows a similar cycle between 1959 and 1974.
A related sectoral measure is the income ratio between govern-
ment and private nonagricultural employees (Line 17). Between 1960 and 1976,
government workers lose relative to private. Factories and government
together account for the bulk of what is usually considered the "modern
1/ The absolute size of this premium is not known exactly since reportedfarm wage rates are on a daily basis, and reported construction ratesare hourly: the measured premium will turn 'on the assumption maderegarding average hours and days worked., Also, payments in boththese markets more often than not include pay in kind, piece rates,and other incentive pay. The published rates are reportedly for"pure!' cash contracts but-the frequency of "impure" contracts allowsconsiderable room for unmeasured differences in the average levels ofthese two key wage rates, as well as in their trends.
- 72 -
sector;" they are also the categories of employment most subject to direct
policy influence on wages. In both, wages are higher than in informal or
small-scale establishments, but that premium shows no net increase between
1959 and 1976. 1n fact, the average wage in the modern sector as a whole
appears i:o fall r .Jative to that in the urban informal sector.
The factory-workshop wage ratio (Line 15) raises questions about
wage determination. We noted above the arguments that wage policy tended
to depress manual wages during the sixties while organizational forces
pulled up supervisory and managerial salaries. These arguments imply
slow or no growth in the average factory wage (including manual and managerial
workers) starting up to about 1968, and faster growth after that year. The
factory wage series (Table 14) tends to bear out this prediction.
However, wages in the small-enterprise sector, which are determined in
relatively competitive markets for skilled and unskilled manual workers,
followed an even more pronounced cycle: the workshop wage lagged behind
factories in the sixties, but grew faster after 1970.
- 73 -
This behavior suggests several, possible complementary explana-
tions. One is that the effects of policy have been exaggerated: sitall enterprise
wages, which are not likely to have been affected by wage policy, grew
even more slowly in the sixties than the wages of production workers in
factories (Line 16). Another is that labor market conditions changed
abruptly after 1968/70 and that the internal labor markets of factories
tended to dampen the effects of this change both before and after. A third
possibility is that the profits-wages link, that has been discussed solely
in connection with senior managerial wages, is equally powerful within
small enterprises. This is made plausible by general knowledge of the
importance of piece rates and other incentive pay, by the fact that many
employees are artisans using their own tools and often working on
a commission basis and the fact that workshop owners are generally on
the payroll paying themselves a combined salary and profit.
Another set of data that bears on the regional distribution
of income and welfare consists of life expectancy rates. These rates
are to some extent an indicator of economic improvement. But the achievement
- 74 -
of longer life expectancy is of course a fundamental objective of
development in itself. The following table analyzes the trend over
t--e and the distribution of life expectancy at birth by regions of
Brazil in 1950 and 1970:
Table I0; LIFE EXPECTANCY BY REGIONS(1950, 1970)
1950 Life 1970 LifeRegion Population Expectancy Population Expectancy
(1000) (Years) (1000) (Years)
Amazonia 1,845 42.7 3,604 54.2
North 2,629 43.7 4,673 50.4
Northeast 9,866 34.0 15,035 44.2
Bahia 5,479 39.2 8,394 49.7
Minas Gerais 8,739 46.1 13,087 55.4
Rio-Guanabara 4,675 48.7 8,995 57.0
Sao Paulo 9,134 49.4 17,781 58.2
Parana 2,116 45.9 6,930 56.6
South 5,725 55.3 9,567 61.9
Central-West 1,737 49.8 5,073 57.5
Brazil 51,945 43.6 93,139 53.4
Source: Census, in Thomas W. Merrick: "Interregional Differences inFertility in Brazil, 1950-1970" in Demography, August, 1974;and Jose Alberto Magno de Carvalho and C.H. Wood: "Renda eConcentracao da Mortalidade no Brazil," IPEA, Rio de Janeiro.
C m~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~C
~~~~~~~~~~iI 0~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~u------------------------------... ........................ ..... .... .... .... .... .. .. .. . . .. .. .... .... .... .... ....
. . .. ' -. -~~~~~~~~~~~~ - - - - - -
- 76 -
The poorest areas in 1970 have the same life expectancy as the country's
average in 1950 (about 44 years). In turn, life expectancy in the Nprtheast
in 1950 was only 34 years. Thus there is a 10-year span between the poorest
and the E .erage just as there is a 10-year improvement between the national
average for 1950 and that for 1970.
Ranking the ten regions in the table according to the life
expectancy of their populations and weighing the latter by the size of
each region's population, one arrives at Figure 1, which shows the
distribution of life expectancy in 1950 and 1970. If all persons had
had the same (average) life expectancy, the distribution curve would be a
vertical line rising at the- 100 index mark. Conversely, perfect inequality
would be represented on the Chart by a horizontal line. Thus, movement
toward the vertical between two observations reflects greater equality.
This is in fact what can be seen on the Chart. Perhaps partly because of
migration from poor to more affluent areas and the gradual assimilation of
migrants with the resident population, not only has life expectancy increased
substantially, but its distribution is more equal in 1970 than in 1950.
Moreover the absolute improvement at the top (61.9 minus 55.3 = 6.6 years), is
smaller than that at the bottom (44.2 minus 34.0 = 10.2 years) reflecting
"diminishing returns" to life-extending efforts as life expectancy and income
increase.
- 77 -
5. Partial Data: Conclusions
The partial measures reviewed in this section do not support
the view that inequality grew significantly between 1960 and 1976. The
two sets of data that cover -the entire economy - educational and regional
income differentials - suggest little change in the structure-of incomes
since 1960. Educational and occupational differentials show a small
increase (Table 4), while regional differentials remain relatively constant
(Table 9). Law of 2/3 and some wage surveys show some widening in the
wage structure within the urban private modern sector. The Law of 2/3 data,
however, is subject to large margins of error and to biases that probably
overstate the increase in inequality. Also, that data cover a relatively
small proportion of total income and employment (Table 7). Wage surveys
also indicate widening differentials within large scale manufacturing
(Table 8), but the sector involved is an even smaller fraction of total
employment, and its wage structure was subject to forces that cannot be
generalized to the rest of the economy. Employees in the modern sector as
a whole, instead, tend to lose relative to those in the informal sector (Table 9).
A second conclusion is that distributive trends have been
more favorable since 1970, than during the sixties. This is true of
educational and occupational differentials, of large vs. small scale
manufacturing wages, and of regional income ratios.
Both conclusions support our analysis-of Census and survey
income data. We found no definite evidence of growing inequality in these
data, though a best guess regarding 1960 shares would imply some deteriora-
tion (Table 1). Also, whatever deterioration did occur probably took
place between 1960 and 1970. This finding helps to reconcile our conclusions
*with the general view that inequality grew significantly. 'That view has
- 78 -
been based almost entirely on data for the sixties. On the other hand,
there has been a tendency to magnify the indicators of increase'.
inequality during the sixties, and to wrongly project t!,at trend into
the seventies.
The apparent improvement in distributive trends after 1970 is
especially interesting, because it implies that rapid growth had a double
effect on poverty reduction: during the boom, total inco-ne grew more rapidlv
and more evenly.
- 79 -
V. TRENDS IN EMPLOYMENT AND INCOMES
For the most part, incomes rise in two ways: as people move to
better jobs, and as the mean income in specific jobs rises. The
following evidence on each of these sources of income growth thus
complements the indirect evidence obtained from trends in distribu-
tion. At the same time, it helps to explain those trends by providing a
more disaggregated, and less anonymous picture of the course of incomes
obtained from employment.
It is worth noting briefly, however, that property and transfer
income is probably larger than usually reported by the available sources,
and changes in the amount received by the poor may be an important compo-
nent of trends in their overall income. The main flows are likely to
be rents from agricultural holdings and urban real estate, and welfare
payments.-/ PNAD 1972 reported transfer, property and non-factor income
equal to 14% of total urban income, and 11% of rural income.- The propor-
tion of non-working income recipients grew from 7% to 11% of total reci-
pients between 1972 and 1976, according to the respective household surveys.
Since both surveys asked specific, separate questions on such income sources,
the increase probably reflects the rapid growth of welfare payments during
the seventies rather than changing statistical coverage. By 1974, the
number of persons receiving pensions from the two main welfare institutions
1/ Mixed property and labor income received by farmers, self-employedworkers and owners of small businesses is included in the data reportedbelow on income from "employment." The omission referred to here is ofpure property income and transfer payments.
2/ From comparison of Tables 2.8 and 6.1.
- 80 -
1/was larger than the number of factory workers. In this paper we have not
attempted to measure the separate contribution of growing welfare payments
on poverty reduction, but their impact is reflected in the household
income levels re,orted by more recent surveys, and in the estimates of
poverty given below in Table 17.
The measurement of trends in income from employment poses
three statistical tasks: the measurement of (i) movement across jobs,
(ii) income differentials between jobs, and (iii) trends in mean earnings
in specific employment categories. Statistical gaps and imprecisions, partic-
ularly with respect to trends in wage rates and income in specific jobs, limit
the reliability of conclusions on these questions. Neverthe.less, the
conclusions obtained here tend to bear out the picture of widespread
income growth and upward mobility that is implied by the relatively small
changes in distributive shares and the substantial growth in per capita income.
1. The Structure of Employment
The two main features of the change in employment patterns
between 1960 and 1976 are a substantial move out of agriculture, and
an upgrading of the urban employment structure (Table 11). The absolute
number of farm households increased by only 11% over 16 years, and the rate2/
of increase appears to have been slowing. Thus, 4 out of every 5 farm
households that would have been created through natural population growth
1/ The two agencies are the National Institute for Social Welfare (INPS) andFUNRURAL, the agencv of the Ministrv of Social Welfare in charge of administeringsocial welfare programs in the rural sector. By 1976, FUNRURAL's
receipts amounted to about 6% of total rural income. In 1974 there
were 2.6 million workers in factories with 50 or more employees (Pe-squisaIndustrial 1974, p. 128, FIGBE), and 3.2 million INPS and FUNRURAL pen-sioners (INPS_ Mensario Estatistico, and Boletin Estatistico do Funrural,1977). One-thiird of the pensions were paid in rural areas, (by FUNRURAL).Total welfare expenditure by these two agencies was about equal to totalwages and salaries paid by factories in 1974 (Cr.$38.5 billion).
2/ A comparison of 1970 Census and 1976 PNAD data shows a minimal changebetween 1970 ind 1976 (an absolute growth of only 2%), but the marginof error is larger than between Censuses.
- 81 -
Table 1I: TIlF STRUC1URE OF 13iPLOYMENT 1960-1976
Labor Force Annual Percen-(in Millions) tage_ . Crowth
Hlotiseihold 1leads Total Labor Force /Ic 1960-19761960 1976/a 1960 1976 3960 1976 Heads Total(Millions) (_ )
Total 12.06 19.25 100 100 22.75 39.72 3.0
Agriculture 6.77 7.49 56.1 38.3 0.6
Nonagriculture 5.29 11.79 43.9 61.2 5.3 5.7
Manufacture and Construction 1.50 4.67 12.4 24.2 7.4TransporL and Commerce 1.67 2.91 13.8 15.1 3.5Personal Services 1.04 1.73 8.6 9.0 3.2Governsment and Social Services 0.67 1.65 5.6 8.6 5.8Others 0.41 0.83 3.4 4.3 4.5
Selected SubgroupsActivity
Construction 0.78 2.69 8.0Factories (504 workers) (1.23)/b(2.95)/b 5.5Women in Personal Services 1.48 3.29 5.1
OccupationAdministrative and Technical 2.53 7.61 7.1Manual (nonagriculture) 8.15 17.59 4.9
Status (Nonagricul ture)Self-employed and Family 2.97 5.29 3.7E.j'1loyees 7.91 19.38 5.8
/5Participation Rates d 1960 1970 1972 1976
Population (10+) 46.6 44.9 52.7 50.8Women (10+) 16.6 18.5 29.9 29.6Men (10+) 77.2 71.8 76.1 73.6Men (10-19) 45.2 38.8 48.4 45.9
Sources: PNAD 1976. Censo Demografico 1960. Censo Industrial 1960. _esquisa Industrial 1974. PNAD 1972.
Notes
/n Adjusted upwards to include rural areas of North region esing 1970 CenslUs share of rural North in totalpopulationi (3.6%). Nonagriculture figures are slightly underestimated due to omission of nonagriculturalworkers in rural areas of North.
lb Estimated from 1959 and 1974 figures using the 5.5% annual growith rate recorded between 1959-1974.
/c As argtied in the text, thie comparisoni of total labor force structure between 1960 and 1976 is biased bydefinitional changes which lead to higher participatlion rates in 1976. The change is especially large inagriculture so only heads are compared in that sector.
/d The purpose of showin g these datn is to demonst.rate the inconmparabilty between Censuses (1960 and 1970)and surveys (1972 and 1976), particularly by shjbiIng the. jump between 1970 and 1972 in the rates for womenand young rnen.
- 82 -
emigrated. As a result, the number of nonagricultural households more than
doubled, growing at 5.3% yearly over the whole period and at a slightly
faster rate since 1970. The contribution to income growth of that movement
is measured by the differential between agricultural and non-agricultural
.mean household incomes. The most reliable measure is provided by ENDEF:
for Brazil as a whole it was 2.5 times in 1974/75, and the size of the
differential appears to have remained roughly constant since 1960.1/ An
approximate adjustment for urban-rural cost of living diffeteices would
still leave a differential in the order of 2 times.-/
The constancy of the differential itself suggests that this
enormous absorption by urban areas occurred without a flooding of the
lower income categories. But more direct evidence is provided by the
favorable trends in the composition of non-agricultural employment. As may
be seen in Table 10, administrative and technical occupations gained rela-
tive to manual; secondary activities gained related to tertiary; within
tertiary, government and social services (mostly teachers and nurses) gained
relative to other services, and wage employment gained relative to self-
employment and unpaid family workers. Amongst heads, the slowest growth was
in personal services. One unfavorable trend appears to be a loss in the share
1/ Table 9, Lines 1 and 19. The urban-rural differential was 2.9 times
in 1974/75.
2/ Using the estimated cost of living differential of 25% explained in
Table 17, footnote 1.
- 83 -
of factories relative to workshops or small-scale manufacturing but factory
employment grows almost as fast (5.5%) as total nonagricultural employment1/
(5.7%). Also. because the Dersonal services sector has a Particularly large
share of secondary earners, its growth is much understated by the household head
figures, but even the figure for all women in personal services, most of
whom are in domestic service, shows a slower rate of growth (5.1%) than that
for all non-agricultural workers (5.7%).
Those trends in some cases refer to household heads, and in
others to the total labor force. The more reliable comparison is that for
heads, because the post-1970 data are obtained from surveys which use dif-
ferent questions and definitions to define activity and, as a result, yield higher
levels of labor force participation than do censuses (Table 11). The rate for
women, for instance, jumps from 18.5% in 1970 to 29.9% in 1972. None of the
trends cited above, however, are significantly affected by this potential bias.
Our estimate that the rate of employment growth in agriculture was only
0.6% per year is based on the notion that households are a preferable basis for
defining employment trends in agriculture than individuals. This approach has
some obvious shortcomings: it will miss changes in the extent of wage-work
performed by women and children, or in the tendency for secondary earners
in farm households to work in construction or other nonagricultural activity.
1/ This statistic may be affected by changes in coverage. Factory employ-ment (establishments with 50+.workers) is taken from Industrial Censusesof 1960 and 1970 and Industrial Survey of 1974. The 5.5% growth rateshown in Table 11 is that recorded between 1959 (1960 Census) and 1974.In principle, all these sources cover all employment in firms of only 5or more workers, so coverage of 50+ firms should not vary. As a test ofthe Industrial Census data, the Census for 1972 yields 2.5 millionemployees, vs. 2.7 million in SEPT data, and 3.7 million employees forall size establishments according to PNAD 1972. The Census-PNAD residualimplies that about 30% of manufacturing employment was missed by theCensus, but the omission may all occur in firms of under 50 workers.
- 64 -
This could be important, for instance, where access to land is declining,
or where school enrollment is increasing. On the other hand, the usual
definitions of labor force participation, where multiple and overlapping
forms of activity are the rule, and where much activity is
directed at production for own-consumption, can lead to large biases in
estimates of the volume of employment. In the context of rural households
in Brazil, standard definitions require highly arbitrary statistical
allocations, and the'resulting estimates are especially sensitive to changes
in definitions or procedures. Thus, the growth rate of total agricultural
employment obtained from a comparison of 1960 census and 1976 survey figures -
1.3% per year - is much higher than that of heads of households in agri-
culture, but the difference is largely attributable to the different treatment
of female activity. Thus, the difference arises only between censuses
and surveys: between 1960 and 1970 the census data record a growth rate
of total agricultural employment of only 0.5% per year, whereas, a
comparison of the 1970 census and 1976 survey yields a growth of 2.5% per
year. If women, and male unpaid family workers are excluded, the growth
of total agricultural employment between 1960 and 1976 is only 0.7%, similar
to that for household heads.
One aspect of employment change with negative implications for
income is an apparent reduction in access to land and parallel increases
in dependence on wage work and in the urbanization of the agricultural labor1/
force. Numerous studies claim the existence of such trends. Some data
for the State of Sao Paulo show an increase in the category of volantes
1/ E.g. 0 Trabalho Volante na Agricultura.Paulista, Ministerio do Trabalho,
Secretaria de Emprego e Salario, Convenio SINE/SEMO, 1978. This was one
of a series of state-level studies. The literature generally refers toboias-frias or volantes,terms applied to temporary wage workers.
- 85 -
(temporary day workers), from around 17% in 1964 to 26% of total1/
agricultural employment in 1975, but this phenomenon is particularly
pronounced in Sao Paulo. There are no available estimates for all Brazil.
Perhaps the most potentially negative component of this process
is urbanization, since it minimizes opportunities for own-production, and
entails higher living costs. Urbanization, of course, also carries
benefits, particularly in the form of greater access to services and to
a wider set of employment opportunities for all family members. In fact,
any trend to urbanization is surely, at least in part, voluntary. Some
data, however, suggest that the extent of urbanization of agricultural wage
earners has been exaggerated in much of the discussion. In 1976,/2
19% of all agricultural households were urban. To approximate the increase,
between 1960 and 1976, in the proportion of farm household heads who were
both landless and urban it is necessary to deduct from that figure (i) farmers
who were urban residents, and (ii) agriciultural wage-earners who were urban
residents in 1960. These statistics are not available, but the net increase
is likely to have been well under that ceiling figure of 19% of farm3/
households.
1/ Data fron Instituto de Economia Agricola, Secretaria de Agricultura,State of Sao Paulo.
2/ The ratio was -37% in thp'eState-7bf Sao Paulo. PNAD 1976, Table -7 (Familias).
3/ What may be concealed in that figure is a larger movement to urban areasby the dependents of primary earners who are migrating to nonagriculturaljobs in towns and cities. Case studies suggest that a relatively largeproportion of volantes and boias frias are secondary earners.
- 86 -
Changes in the proportion of landless, as distinct from urban
agricultural households are more difficult to estimate. The main
approximation to landlessness is the category of agricultural wage-earners.
In practice, there is enormous overlap between work on one's own land
and wage-employment. The frequency and variety of share-cropping arrange-
ments increase the difficulty of classification.
Since classification is normally by principal income source, the
number of wage-earners reported in censuses and surveys overestimates the
landless, because many "wage-earners" have some access to land, or are
members of households with land. On the other hand, changes in the amount
of land available to those with access, or in the terms of access in the
case of share-croppers and renters may be as important as changes in the
number of those who are strictly landless. The following data on wage-
earners are therefore open to several possible biases as indicators of
either the level or trend in the degree of access to land, but they are
presented as a first step towards better estimates.
A proxy for the number of landless households is the number of
men aged 24 and over classified as wage-earners. These grew only slightly
between 1960 and 1970, from 23.5% to 26.0% of all farm households, though
it is during this period, following the extension of protective labor
legislation to rural areas in 1963, that much of the increase should have1/
occurred. Figures for years after 1970 are not comparable owing to a
1/ The Estatuto do Trabalhador Rural. However, changing crops and
technologies may be a greater part of the explanation for the observed
increase in casual wage-employment in many areas.
- 87 -
different treatment of sharecroppers (parceiros) who in 1960 and 1970 were
listed separately, but in 1976 were allocated between the categories of
employees and independents according to the nature of their relationship
with the landowner. In 1976, the proportion of heads in agriculture defined
as employees was 37% - a substantial jump over the 1970 figure of 26% - but
the definitional component of the change is not known. In summary, the aggregate
data suggest that inferences from case studies on localized trends have tended
to exaggerate the aggregate importance of urbanization and increased land-
lessness amongst the agricultural labor force. On the other hand, the
available aggregate data are very imperfect measures of those trends, and
the more definite evidence from case studies indicates that. in some areas,
at least, there has been a negative change in the agricultural employment
structure.
2. The Structure of Income
The broad occupational-regional hierarchy is provided in Table 12
using ENDEF data. The occupations are those of heads of families only,
while the incomes shown include the earnings of secondary workers and are
therefore higher than the earnings of any individual in those categories.
The table illustrates the power of both occupational and regional variables
in creating stratification. One point to note is the big differential
between farm and nonfarm manual labor ("Farm Laborer" and "Employee: Manual").
In all regions, the landless laborer doubles his income by moving to urban
manual employment within his own region. Allowance for urban-rural cost
of living differences would still leave increases of well over 50% since
the landless buy much of their food, while at least half of nonfarm manual
employment is in small cities and towns where cost of living differences
- 88 -
with rural areas are not as large as in metropolitan areas. The rural-Northeast
to urban-Rio move roughly triples income, while the family income of a Sao Paulo
manual workers is 4.7 times that of a Northeast farm laborer.
The size of these differences seems surprising in view of the
apparent extent of mobility and integration of unskilled labor markets.
Urban wage rates in entry jobs in construction, factories, small urban
establishments, and domestic services are generally between 1 and 2 minimum
wages, levels which are now close, in absolute terms, to daily wage rates1/
in agriculture. The observed higher mean earnings of urban manual workers
must therefore reflect the existence of a skill hierarchy. within urban manual
employment, as we noted in discussing wage differentials avove, and thus, the
potential for upward mobility within the category of manual employment. Another
factor may be the greater availability in urban areas of employment opportunities
for secondary workers.
The importance of access to land can also be measured in Table 12.
The term "laborer" must be interpreted as a matter of degree, since ENDEF
data show that "farm laborers" produce a substantial portion of their own
consumption. Small farmers in all regions receive, on average, about
twice the income of agricultural laborers in their region. A major exception
is the Northeast where the difference between farm laborers and small
farmers is minor. A more important qualification is that farm laborers
in all regions outside the Northeast earn more than small farmers in the
Northeast: the most striking difference is in Sao Paulo where the earnings
of a farm laborer's household are 70% higher than those of the Northeast
small farmer.
1/ As shown in Wage Statistics in Appendix.
- L,9
1T * 1 l : I :;:( ( ' I(A; I(iNAI. fI ;:iI (-iAl. (:ROI:iS 1.V INCOME LIEVL
tI;EIiIffi l. A ,:Nj hoU!Tt;f:OLD iN::OME) 19I141.5
Ccce.v-;>ntls.nw -ar ion'Asil : __ _AMNro uAltore -___tu
Category Ne.:n Nudber Mean NwiuberR gieon Income Pre I i es Region Income Fa m:: Jies
.. ____ ("'Ofl _ ('000) ('CCC) ('000)
Total S,219 6.263
Manager Rlo 169.1 78
SP ]67.2 82
MG 143.4 19
NE 137.9 37
Employer or Professional. Mr 136.3 62
SP 123.1 204Rio 120.7 87
Manager South 120.4 34
F.nployer or Professional South 108.5 117
MiddIle M1anagement Sr 79.1 235
Shopkeeper SP 71.8 159
Em:ployer or Professional NE 67.9 84
Middle Manacgement Rio 67.7 139
South 57.6 116Farmer SP 54.5 102
Middle Management Mr 54.0 68
NE 52.1 96
Farmer MG 45.2 150
Rio 45.1 23Shopkeeper South 44.4 117
Eraployce: Nonmanual SP 43.3 405
Shopkeeper Rio 42.9 57
Farmer South 42.5 286
Self-employed SP 38.2 261
Workshop SP 36.8 121
Employee: Nonmanjal Sout)i 35.7 179
Workshop South 35.5 96
Employee: Nonmanual MG 34.4 123Rio 34.2 254
Shopkeeper MG 34.0 96
Workshop Rio 31.0 49
MG 30.9 43
ELn;,loyee: .::inhuil SP 25.6 1,337
Self-Cupio.td RIo 24.9 169
South 24.5 Il5
Srail Farmer SP 23.2 174
Employee: Nore.;:nual NE 23.0 200
Emcployee: Manual Rio 21.0 685
Small Farmer South 20.2 930
Employee: Man,ial Sourh 19.2 714
Self-cemployed MG 17.8 180
Shiopkeeper NE 17.3 343
E.sployee: Manual MC 16.3 526Sm.all Fanrer OG 16.0 439
Farrmer NE 15.9 487
Small Farmer Rio 14.8 41
Workshop NE 14.4 147
Farm Laborer SP 12.6 400
EaLplcyee: ;Ianual NE 11.9 848
Self-e-mployed NE 10.8 457
Farm Laborer South 9.9 329Rio 9.0 71
MG 7.5 480
Small Farmer NE 7.3 1.239
Farm Laborer NE 5.5 1,112
Sources: ENrMEF. Oceunation Is of he;ld of hou,sehold. Ilco.ne is annual household I icome
in August 1974 cr-zviros. Exchange rate - Cr.$/US$.
Definitions
1. Manager: senior manngers, gencrally wit), professional qualificationA.
2. Employe;r or professioena.: tar ldes:; self-emoloyed professionals.
3. Middle m!-snagerent: includes supervisory pos!t, Ines requiring high school level.
4. Shopkeeper: : nciUleCs e raVe:S of service esit:d,lirlmeonts.
5. farime: ridu!Le and largoye s: n I:..0 moloc:. waye-l hor.
6. Employee: nonmamI l: low J--. *.fflr 2ce rers, store and service employees.
7. Wor-kberp: a!d is;n. :-.;ir l.; owl otlher veTI-al .activitles.8. Sf-t;..jM.:.-: csi. if inn :,Jst,r.,iI, cuafi irn establishlmert
9. SnaI.al fa;vn r: hL' I lilt' inti':
10I. I: :.e-r: r i n .'.;r,--c-: I:..- :-
11. RE5 I: j;ef: Sil: s;.c tlc; -: : : I ' :... ! .. :x .I e.ist
- 90 -
The data in Table 12 also indicate the size and composition of
middle income groups; 34% of all family, for instance, have family incomes
in the order of 5 to 20 minimum wages (Cr.$22,500 to 90,000). These include
a sizeable contingent of farmers who achieve incomes well above those of
agricultural laborers and who comprise a rural middle class. Much larger,
however, is the size of the urban middle-income groups made up of skilled
workers and a large number of small business owners.
Within agriculture, there is a strong relationship between
regional average productivity and the incomes of both small farmers and
farm laborer households. These relationships are shown in Table 13. The
comparison in that table is crude, owing chiefly to the lack of regional
value added data for 1974/75, but it serves to establish the relationships
between productivity and incomes. First, small farmer incomes are strongly
correlated with average regional productivity: the high productivity of
the South and Southeast is not limited to large farmers. Second, the
farm laborers also benefit substantially from higher regional productivity:
the elasticity of their household incomes to regional productivity is
between one-third and one-half. That elasticity is about twice that of
the rural wage rate. The Sao Paulo farm wage-earner's household income
is higher than in the Northeast only partly because the regional wage rate
is higher. Over half the higher family income of the Sao Paulo farm laborers
must be explained in other ways. One possibility is more days worked per year,
which may in turn be related to more developed labor markets (better in-
formation, quicker transport, and more efficient intermediation). Another
is more and better paid work opportunities outside agriculture -- in towns
and cities during agricultural slack seasons. These greater employment
opportunities apply as well to secondary earners. Finally, since many
- 91 -
Table 13: FARM HOUSEHOLD INCONES AND REGIONALAGRICULTURAL PRODUCTIVITY
Indices ofHousehold Income
Regional Agricultural Seall Farm RuralValue Added per Worker Farmer Laborer Casual Wage
1970/1972 1974/75 1974
Sao Paulo 452 318 229 149
South 306 277 180 146
Rio 257 203 164 107
Minas Gerais 174 219 136 121
Northeast 100 100 100 100
Source: ENDEF for household incomes. National accounts for 1970 value added.J.F. Graziano da Silva, coordinator, Estrutura Agraria e Producao deSubsistencia na Agricultura Brasileira (Sao Paulo: Editora Hucitec.
1978), Table 51 for 1972 regional productivity per worker. For 1972we used average of productivity per permanent and per maximum (peakseason) annual employment 1970/72 figure shown here is average ofthose two years. FGV for rural casual wage.
- 92 -
wage-earners have access to some land, they may enjoy higher productivity
on those plots than their counterparts in the Northeast.
What is most interesting about the cross-section relationship
between small farmer and wage-earner incomes and regional productivity is
that it suggests that the gains from productivity growth and improved terms
of trade have been widely shared within the agricultural sector.
3. Income Trends
The limited direct evidence on trends in the income of specific
occupational categories are the wage series summarized in Table 14 below
and the wage and income series discussed above and shown in Tables 8 and 9.
The broad conclusions that are suggested by those data are first
that most wages have been rising though at very variable rates. The general
picture is clearly not one of wage stagnation. A second conclusion is
that most wages grew faster during the period of accelerated economic
growth which began in 1968, though the pre-1970 data is scant. A third conclu-
sion is the lack of a clear relationship between the growth of skilled and
unskilled wages. Skilled categories gained relative to unskilled within
large-scale manufacturing, and there was some relative improvement for white-
collar workers over nonagricultural manual workers during the sixties, but the
unskilled fared better in construction, while unskilled casual farm labor
shows substantial growth in all regions of Brazil and a particularly rapid
increase in Sao Paulo. The single most striking and important feature of
those series is the marked increase after 1970 of real wage rates for casual
rural farm laborers. These wage rates went up by 61% in real terms between
1970 and 1977, Excluding the most advanced state of Sao Paulo, the average
- 93 -
wage rate for casual rural laborers still goes up by 56X for the rest of
Brazil. For comparison, per capita income went up by 57% during the period
1970-1977. Thus an important group of workers who are among the poorest
in the country experienced a substantial absolute improvement in their1/
daily earnings.
Two qualifications must be made, however, with regard to these
statistics. One is that not enough is known regarding the precise manner
of measurement. Given the wide dispersion that characterizes wages even
for specific occupations, the accuracy of an index is usually sensitive to
the method of sampling, the moment of sampling, the coverage of the population
being sampled, and the extent to which various components of real remuneration
are included. The second qualification consists of the same set of
questions and degree of uncertainty with respect to the available price
indices. One price index in particular, estimated and published by DIESSE
shows a significantly faster rate of increase. A comparison of this
index with the more commonly used Cost of Living Index for Rio, (corrected
for underestimation in 1973) is shown in Table 15. Price indices for
other cities grow at variable rates, but the size of the differential between
the DIESSE index and others, especially since 1974, should be examined
further.
1/ The rural wage series refer to cash only. After the extension of laborlegislation to the rural areas in the mid-1960s the income in kind ofmany rural workers was reduced; this reduction is unlikely to have hadmuch if any bearing on rural income trends in the 1970s.
Table 14: INCOME AND WAGE TRENDS: INDICES
BaseYear 1970 1971 1972 1973 1974 1975 1976 1977=100
1. Sao Paulo Casual Rural Wage 1960 112 121 128 145 180 184 193 210
2. All Brazil: Casual Rural Wage 1968 97 103 109 130 157 165 159 156
3. Construction: Unskilled 1968 92 95 99 100 111 117 117 116
4. Construction: Skilled 1968 89 91 97 103 94 99 97 94
5. Construction: Foremen 1968 88 105 127 134 154 165 174 178
6. Urban Employees (SEPT) 1965 121 121 144
7. Manufacturing Mean Wage (Central Bank) 1971 100 108 119 119 131 140
8. to (FI3GE/DEICOM) 1959 132 136 149 149 152 161
9. (Sao Paulo: 2/3 Law) 1972 100 128
10. " (Sao Paulo: DIESSE) 1961 92 97
11. *' (with FGV 1961 104 121deflator)
12. Car Industry 1966 114 113 124 129 127 130
13. Basic Industries (ABDIB) 1971 100 108 119 119 131 140
Note: All series deflated by Rio COL (adjusted upwards in 1973) unless otherwise indicated.
Line
1-5: See Appendix on Wage Statistics.6: Law of 2/3, S PT, Ministry of Labor.7: Bolet.n do Barnco Central do Brasil.8: Mean remuneration of plant workers in manufacturing (includes plant supervisors and
foremen) ob..ained from industrial surveys.9: A weighted average of mean industry wages obtained from R. Macedo, Emprego e Salario
No Ciclo Economico 1972/75, Programa de Estudos, Secretaria de Emprego e Salario do MTG,FIPE.
10: DIESSE. Deflated by DIESSE cost of living index.12: ANFAVEA. Cited by Suplicy [1978). DeflAted by FIPE/USP cost of living index.13: Conluntura Economica.
- 95 -
The many questions that surround wage and price statistics make
it difficult to arrive at precise conclusions regarding the extent of
income growth in the important occupational categories covered by those
series. It does not seem plausible, however, that the variety of evidence
that has been cited on income trends in different regions and.occupations
.and sectors is providing a uniformly distorted picture, nor that the degree
of distortion would alter the broad conclusions cited above, i.e., that
real wages.and incomes in most employment categories have been growing at
significant rates, particularly since 1968.
The trend in rural wages is made plausible by the relationship
between regional farm incomes and agricultural productivity (Table 13).
And, as suggested by these data, the trend in wages may not capture the full
extent of income growth generated by rising productivity, since incomes may have
been affected also by increased opportunities for work during the year, and for
primary and secondary earners in both rural and urban.areas. The close
association between the trends in rural wage rate and farm productivity may
be seen in the following figures. An index of agricultural value added
in Brazil, deflated by the cost of living index to reflect consumer purchasing
power moves as follows: 1959=100, 1970=97, 1975=172, 1977=223. The rural
casual wage shown in Table 14 follows the same pattern closely: little
change during the sixties; sharp increase between 1970-1975. The falling
off in the rural wage since 1975 suggests that it was the exceptional
combination of an urban boom with sharp growth in agricultural value added
that generated so strong a response in wages, and that the slowing down of
urban growth has reduced pressure on the supply of rural labor.
- 96 -
Table 15: COMPARISON OF DEFLATORS
Indices Index of
FGV DIESSE DIESSE.FGV
(Rio c.o.l.)
1964 100 100 100
1966 234 254 109
1971 673 761 113
1974 1,273 1,533 120
1977 (Dec.) 3,407 4,885 143
Index for 1974Real Wage Rate Indices Deflator
(1968--'100) FGV DIESSE
Sao Paulo: Daily farm labor 150 125
Rest Brazil: " " " 159 133
Sources: 1. DIESSE, Dez Anos de Politica Salarial. Measures
price changes for a typical consumption basket of
workers in Sao Paulo.
2. FGV: (Fundaca"o Getulio Vargas), Conjuntura, Correctedfor 1973 as explained in note to Appendix Table II1.
- 97 -
This review of statistical evidence on wage differentials, changing
employment structure, and wage trends has found evidence of widespread
income growth resulting from both movement to better jobs and rising incomes
within categories of employment. (Table 16 provides some corroboratory
evidence: it shows the substantial growth in the ownership of durables1/
and use of electricity.) The data are stronger on the shift component of
income growth than on wage rates. Shifts in employment have clearly
contributed to income growth at many levels, including the poorest. The
wage trends also support the hypothesis suggested by other data reviewed
above, that income growth amongst poorer groups accelerated after 1970.
A proper statistical disaggregation of the sources of income growth would
perhaps find that mobility and employment change contributed more to
income growth during the intercensal decade than did rising wages, but
that the independent contribution of wage increases to income growth rose.
after 1970.
1/ See John Wells, "The Diffusion of Durables in Brazil and Its Implicationsfor Recent Controversies Concerning Brazilian Development," CambridgeJournal of Economics 1 (3): 259-79. Wells argues that the growth inconsumption of durables was much broader in terms of income groups thanwas believed.
- 98 -
Table 16: PROPORTION OF HOUSEHOLDS OWNING DURABLES
1960 1970 1972 1974/75 1976
Brazil
Refrigerator 12 26 31 36 42
TV 5 24 32 39 47
Electric Light 39 48 N.A. N.A. 63
Radio 35 59 73 69 76
Northeast
Refrigerator 3 9 11 13 17
TV 0.3 8 10 13 18
Electric Light 16 23 N.A. N.A. 34
Radio 12 35 45 51 59
Sources: 1960 and 1970 Censuses; 1972-1976 PNAD; ENDEF.
N.A.:- Not available.
- 99 -
VI. THE EXTENT OF POVERTY IN 1974/75
An approximate measure of the size and regional distribution of
poverty is presented in Table 16 using ENDEF. The main purpose of identi-
fying the poorest group in the population is, of course, to orient the
location and design of government efforts. The figures
also provide a benchmark for measuring progress in the reduction
of Doverty. These purposes can be served by the arbitrarily
defined poverty line use here. A further contribution could be made by
estimating an absolute poverty line based on nutritional requirements,
though more research is needed on the various dimensions of the relation-
ship between income levels and basic needs.
The publication of partial results of the national household
expenditure survey of 1974-75 (ENDEF) provides a more reliable basis
for a measure of poverty than previously available statistics. As noted
above, the omission of both money and non-monetary income weakens the
value of the Census and PNAD as measures of income levels among the very
poor.
The poverty line chosen here - two Rio minimum wages per family
US$260 per capita - identifies 27% of the population as
poor. A crude adjustment for regional cost of living differences was
carried out. The regional distribution of these families corresponds to
common notions regarding Brazil: 61% of the poor are rural, and one half
are in the Northeast. A finding that is less generally known is that
almost three quarters of the urban poor are in smaller cities and towns
rather than in metropolitan areas. More precise measures would modify
- 100 -
Table 17: THF. EXTENT OF POVERTY: FAMILIES UNDER
TWO MINIMUM WAGES:/1 1974-1975
Total Percent Number Regional Distri-
Regions Families Poor in Poor bution of Pcverty('000) Region ('000) (Percentage)
Rio 2,169 11 244 4Metropolitan 1,784 9 163 3Other Urban 209 14 28 1Rural 176 30 53 1
Sao Paulo 4,168 9 385 7Metropolitan 2,078 5 103 2Other Urban 1,413 11 156 3Rural 677 19 126 2
South 3,548 15 520 9Metropolitan 601 8 48 1Other Urban 1,162 13 147 3Rural 1,785 18 325 6
Minas G. & E.S. 2,592 30 788 14Metropolitan 367 14 52 1Other Urban 1,034 25 260 5Rural 1,191 40 476 9
Northeast 5.791 52 3,008 54Metropolitan 844 27 231 4Other Urban 1,673 44 731 14Rural 3,274 62 2,046 37
North & D.F. 2,032 30 619 11
Metropolitan 156 4 6 -Other Urban 918 25 230 4Rural 958 40 383 7
Total 20,300 27 100Metropolitan 5,830 10 603 11Other Urban 6,409 24 1,554 28Rural 8,061 42 3,409 61
/1 Adjusted for regional cost of living differences. Income conceptis total family expenditure including nonmonetary and capital (realestate, vehicles, stocks) expenditures. (From ENDEF, Table 7.)Basis for poverty line was the highest regional minimum wage inAugust 1974, or Cr.$376.80 per month. Adjustment for regional costof living differences is an arbitrary approximation. Differenceswere assumed to equal 30% between rural and Metropolitan and 20%between rural and other urban. The poverty line is Cr.$9,000 peryear (US$1,300) in Metropolitan areas, Cr.$7,500 in other urban areasand Cr.$6,923 in rural areas. Assuming an average family of 5, thisamounts to a poverty line of US$260 per capita in metropolitan areas.
/2 Data for Ncrth are approximations guided by PNAD 1972 data on familiesby income class and by regional unskilled wage differentials.
- 101 -
these findings. The most important improvement that is required is the
estimation of more accurate regional cost of living differences.
A major finding here is that income levels of the poor have been
underestimated in the past. Estimates, based on PNAD 1972 money income only
data, for instance, place 62% of all households below the two minimum wage
line. A recent ECLA/IBRD study, using PNAD 1972 total income figures and
an almost identical poverty line, estimates that 47% of Brazilian families
were poor in 1972, a figure substantially higher than the 27% obtained1/
from ENDEF. Using a lower poverty line of US$130 per capita, which is
closer to figures normally used in other countries, would reduce the poverty
group to less than 15% of the population.
The most thorough previous measurement of poverty levels and
characteristics was carried out by Meesook and Fishlow for 1960, using2/
Census data. Their poverty line was equal to one Northeast urban minimum3/
wage, or US$130 in 1974 prices. After adjusting for non-monetary income,
they estimate that 31% of all families in 1960 fell below that level. If
we apply the same real poverty line to the ENDEF data, the share of the4/
poor is in the order of 15%, or half of the 1960 level. Since there is
a strong presumption that ENDEF figures are better approximations to true
1/ Oscar Altimir, CEPAL/BIRF, La Dimensioni de la Pobreza on America Latina(March 1978). The poverty line was Cr.$890 per capita per year which,in August 1974 prices, amounts to Cr.$9,700 vs. Cr.$9,000 used here.
2/ Albert Fishlow, "Brazilian Size Distribution of Income," AER 1972.
3/ Applying the Guanabara Retail Price index, plus an upward adjustmentfor the rate of inflation in 1973 which government spokesmen now admitto be closer to 22.5% than the official figure of 12.6%.
4/1 Reflated to 1974 prices using the Guanabara CQst of Living Index.
- 102 -
income than Census-based estimates, this inconsistency has only two possible
explanations. One is that even the corrected Fishlow4Meesook income figures
are severe underestimates. Another is that poverty has been reduced. Since
it would require an implausibly large underestimation to explain the gap,
it seems probable that the much lower poverty share in 1974-75 is the result,
at least in part, of a real reduction- in poverty.
The discovery of higher income levels at the bottom of the
distribution is more a measure of the inaccuracy of earlier statistics than
of the adequacy of those income levels. There is much direct evidence on
the high levels of malnutrition, mortality rates and severely deficient
services and living conditions that correspond to income levels in the
vicinity of two minimum wages. In part this reflects high prices in Brazil
by international standards. More important perhaps is the fact that some
aspects of family welfare in Brazil have lagged behind the growth of money
incomes. This is most obvious with respect to the lag in public services
and relatively high levels of malnutrition, but it may also involve aspects
of welfare that are difficult to quantify, such as low levels of comunal
support and resources, and the costs associated with high degrees of
individual mobility.
- 103 -
APPENDIX
Note on Internal Terms of Trade
An important factor in distribution change is the trend in internal
terms of trade between agriculture and the rest of the economy. Since the
poorer segments of the population include a high percentage of rural families,
the relationship between prices received by farmers for their products and
prices they pay for purchasing goods and services from the non-agricultural
sectors is significant for income distribution. The following table shows
three series: (a) changes in the relationship between wholesale prices
for agricultural output excluding export crops, and wholesale prices for
industrial goods; and (b) changes in the relationship between wholesale prices
for agricultural output excluding export crops, and wholesale prices for
textiles, shoes and clothing, three items of particular importance in the
budget of farmers who grow much of the food they consume. A third index (c)
shows changes in the relationship between wholesale prices for all agricultural
products (including export crops), and wholesale prices for textiles, shoes
and clothing. The table suggests that terms of trade have turned in favor
of agricultural products (whether food producers for the domestic market or
export producers) during the 1970s. While until the 1970s it took at least
one unit of agricultural production to buy a pair of shoes or a dress, it
now only takes 50 to 60% as much agricultural output to purchase these
manufactured goods. For industrial goods as a whole the reduction is less
marked, but still in the order of 25% between 1970 and 1977. These trends
suggest that agricultural producers (who include a large proportion of
small poor farmers) have improved their purchasing power per unit produced
during the 1970s.
- 104 -
Indicators of Internal Terms of Trade(1965-67 = 100)
(a) (b) (c)
1944-1950average 83.1 77.9 82.7
1951-1959average 99.4 93.9 112.0
1960 101.2 95.6 99.91961 96.9 85.0 87.71962 107.2 96.7 99.41963 100.0 91.4 91.0.1964 99.8 96.1 104.31965 91.0 92.4 97.21966 103.5 103.3 102.31967 103.5 102.4 100.81968 92.0 89.3 88.71969 94.4 102.8 100.61970 99.6 108.6 111.01971 108.9 117.7 117.31972 112.8 121.4 123.11973 113.4 119.8 125.41974 112.3 135.8 143.61975 111.5 159.8 163.21976 120.5 165.7 182.41977 124.1 178.5 204.8
Source: Conjuntura Economica (Fundayao Getulio Vargas).
This impression is confirmed by looking at the relationship be-
tween farmgate prices for crops since 1970. These increased much faster
between 1970 and 1977 than either the Rio de Janeiro cost of living index or
its clothing and apparel component, as shown in the following table.
Furthermore, as the table shows, the improvement in terms of trade for the
- 105 -
rural sector is not confined to export products but includes a wide range
of agricultural and livestock products. 1/
Increase in farm-gate prices as a multiple of increasesin (a) the Rio cost of living index and (b) its apparel component
(1970-1977) /1
(a) (b)
Farm Products
Cotton 1.6 2.7Peanuts 1.7 2.8Rice 1.1 1.7Bananas 1.7 2.9Potatoes 1.3 2.2Cocoa 4.0 6.7Coffee 3.2 5.3Sugar 1.2 2.0Beans 1.4 2.4Tobacco 1.7 2.7Oranges 1.2 1.9Manioc 3.5 5.9Corn 1.0 1.7Wheat 0.9 1.6Beef 1.1 1.9Hogs 1.3 2.1Chicken 0.9 1.6Milk 1.4 2.3
All Crops 2.3 3.9Livestock Products 1.2 2.0
All Farm Products 2.1 3.5
Source: Conjuntura Economica.
/1 The cost of living in this table has been corrected for the likely1973 under-estimate.
1/ The food and non-food components of the Rio de Janeiro cost of livingindex have moved as follows between 1970 and 1977:
c.o.l. Food Non-food
1970: 100 100 1001977: 529 590 486
Thus, if computed only over the non-food component of the c.o.1. indexthe figures in column (a) in the table would-be increased by about9 percent.
- 106 -
In sum, the trend in the domestic terms of trade in favor of agri-
culture during the 1970's is fairly clear. The relationship between that
trend and rural incomes, particularly incomes of the poor rural population,
is, of course, very complex, and needs much further exploration.
- 107 -
Appendix on Wage Statistics
The statistical series underlying the discussion of wage trends
are shown in Appendix Tables I to V. They contain the following information:
Appendix Table I: daily wage rate of unskilled rural laborers by states
except frontier states, for the period December 1968 - December 1977.
These rates are shown in current cruzeiros as well as in cruzeiros of
December 1977. States not shown on this table are in Appendix Table II,
as well as averages for all states. The average wage rate for all states
in Appendix Table II reflects the 1972-76 weights of the Fundacao Getulio
Vargas, the source of these data.
Appendix Table II: daily wage rates based on an 8-hour day for construction
workers in the principal city of each state. The table also shows the
legal urban minimum wage based on 27 days per month. 1/ Wages for
"lunskilled" workers are those for serventes; they do such jobs as
excavating without power tools or carrying loads on construction sites.
Wages for the "semi-skilled" are for armadores who have enough formal
or on-the-job training to shape and set metal rods into concrete;
they are the equivalent of skilled workers in manufacturing. Brick-
layers, painters, carpenters also fall into this category; their wage
rates are within 12 percent of those for armadores. "Skilled" workers
in the present context are foremen (mestres de obra) who typically have
either completed secondary school plus 2-3 years' vocational training,
or have an equivalent informal training. They are the equivalent of
technicians in manufacturing industry. The table also includes wage
1/ The legal wage is defined in Cr$ per month. It is unclear whether thedaily rate should be based on 30 or 27 days, but since the data are usedprincipally for trends the possible error is immaterial.-
- 108 -
statistics for Brazil, with and without Sao Paulo. These averages are
weighted, in the case of construction and in the case of the legal minimum
wage, by the 1970 population of the major cities listed in the Table.
Appendix Table III: the same as the previous table, but with all wage
rates brought forward into December 1977 cruzeiros. After checking
various regional deflators it was found that they do not differ enough
to warrant regional deflations. Nor is the difference between food
prices and other prices large or consistent enough to warrant separate
deflation. Therefore all current wage rates in this report are de-
flated by the Rio de Janeiro cost of living index, with an upward
adjustment of the rate of inflation in 1973, from the official figure
-of 12.6% to 22.5%, which government spokesmen now admit to be closer
to the true value for that year.
Appendix Table IV: the Fundacao Getulio Vargas collects from the agri-
cultural extension'service-statistics on monthly wages of permanent.
farm workers in each municipio. The table shows the trend in the
relationship between those monthly (money) wages on one hand and
daily wage rates for casual rural laborers on the other for 14 states
and Brazil (minus Sao Paulo and the frontier states), for the period
1966-1977.
Appendix Table V: Agricultural wage trend for the state of Sao Paulo.
The Instituto de Economia Agricola of the state of Sao Paulo collects
wage statistics through an annual survey of agricultural establishments
and thus provides an independent check to the FGV wages reported by
extensionists. Source for 1948-1968 data is IEA, "Modernization of
Agriculture in the State of Sao Paulo," 1973. The figures after 1969
- 109 -
are drawn from earlier attachment Tables, assuming that the relation-
ship between December wages and annual averages obtains for 1975,
1976 and 1977. The deflator is the Sao Paulo cost of living index
adjusted in the same way as in Appendix Table III for inflation
in 1973. The wages concern only money income. Sundays and holidays
are incorporated in the attached series (which through 1969 is based
on monthly original aggregations). Therefore the daily wage between
1948 and 1963-64 is likely to be overstated. The series is shown in
graphic form in Figure 3 of the main text.
Appendix Table I
Daily Rural Unskilled Wages 1968-77(in Cruzeiros)
Espirito Sta.Maranhao Ceara Paraiba RGN Alagoas Sergipe Santo Catarina RGS
Daily rural unskilled wages
1968 2.45 1.81 2.06 2.40 2.22 2.60 2.66 4.15 3.921969 2.71 2.30 2.40 2.58 2.42 3.00 3.28 4.89 4.891970 3.77 2.57 3.16 2.90 3.49 3.44 4.05 6.09 5.951971 4.63 3.62 4.07 3.50 3.81 4.61 4.85 7.42 7.081972 5.83 4.27 4.73 4.52 4.68 5.13 6.G1 9.07 8.941973 7.70 6.16 6.98 6.23 6.98 6.64 9.02 12.54 12.611974 11.64 10.65 13.27 10.79 12.55 13.10 14.51 20.25 17.761975 16.72 13.65 16.01 15.82 16.60 20.03 20.70 28.90 24.461976 23.12 19.56 23.46 21.49 24.42 23.90 30.65 38.17 34.081977 30.00 29.00 33.00 30.00 37.00 34.00 47.00 53.00 47.00
Daily rural unskilled wages deflated by C.O.L. index.
1968 20.42 15.08 17.17 20.00 18.50 21.67 22.17 34.58 32.671969 18.19 15.44 16.11 17.32 16.24 20.13 22.01 32.82 32.821970 20.94 14.28 17.56 16.11 19.39 19.11 22.50 33.83 33.061971 21.84 17-08 19.20 16.51 17.97 21.75 22.88 35.00 33.401972 24.09 17.64 19.55 18.68 19.34 21.20 26.07 37.48 36.941973 25.83 20.66 23.41 20.90 23.41 22.27 30.25 42.06 42.291974 29.07 26.60 33.14 26.95 31.34 32.72 36.24 50.57 44.351975 31.88 26.03 30.53 30.16 31.65 38.19 39.47 55.10 46.641976 30.40 25.72 30.84 28.25 32.11 31.42 40.30 50.19 44.811977 27.57 26.65 30.33 27.57 34.00 31.25 43.19 48.71 43.19
- ill -
Appendix Table II (paRs I of 2)
DIiy Vaga(in Cruzeiroa)
crNsmRicrloNUrban
Bural Unskilled Skilled Fre n Mlnlmm Wane
Natal (RON)1968 1.97 2.64 4.00 *0. 2.931969 2.27 3.28 4.40 IA.00 3.6
14
1970 2.95 4.16 5.44 30.40 4.621971 3.65 5.04 6.80 30.40 5.6o1972 4.89 6.08 8.32 30.40 6.761973 7.16 6.80 10.00 32.00 7.911974 13.13 10.72 16.00 32.00 9.871975 14.58 12.56 24.00 72.00 13.961976 21.51 21.04 40.00 132.00 20.181977 34.00 26.48 46.00 200.00 29.16
Salvador (Bahia)1968 2.63 3.36 6.80 22.00 3.711969 3.03 4.00 9.60 24.00 4.441970 4.32 4.80 12.00 32.24 5.331971 5.01 5.76 13.60 52.00 6.401972 5.84 6.88 16.00 48.00 7.641973 8.33 8.00 17.60 80.00. 8.891974 14.83 10.80 23.20 123.20 10.931975 20.12 19.20 36.00 160.00 15.471976 28.10 24.00 47.20 240.00 22.311977 37.00 36.00 64.00 318.00 32.18
Rio de Janeiro (Rio)1968 3.07 4.32 9.60 18.80 4.80
1969 3.54 5.20 11.20 24.00 5.78
1970 4.43 6.24 12.00 26.80 6.931971 5.16 7.52 14.80 39.84 8.361972 6.54 8.96 19.76 43.20 9.961973 9.16 10.40 21.60 60.00 11.561974 13.37 14.40 26.40 93.60 13.961975 19.98 20.00 36.00 120.00 19.731976 28.97 32.00 54.00 184.00 28. 44
1977 44.00 48.00 80.00 320.00 40.98
San Paulo (S.P.) 1/S 1968 3.69 4.80 9.60 18.00 4.80
1969 4.36 5.20 11.52 18.40 5.781970 5.21 6.24 12.08 20.00 6.931971 6.61 7.60 14.80 26.56 *.361972 8.01 8.96 17.60 53.28 9.961973 11.18 12.00 24.80 55.84 11.561974 18.60 19.20 30.40 96.00 13.961975 25.00 24.00 38.40 132.00 19.731976 38.00 33.60 52.00 198.00 25.441977 59.00 48.00 72.00 256.32 40.98
Curitiba (Parana)1968 3.42 4.16 6.88 8.24 4.36
1969 4.23 5.12 8.56 *2.e0 5.241970 5.32 6.32 10.96 13.12 6.311971 7.03 7.76 13.60 17.60 7.731972 7.69 9.36 16.32 21.36 9.241973 10.73 10.96 19.92 27.68 10.671974 16.74 16.00 26.00 48.00 12.98
1975 26.45 22.40 40.00 73.60 18.311976 32.52 32.00 59.60 112.00 26.401977 46.00 40.00 76.00 186.80 38.04
Balo Pori20nee (binas Cernia)1968 2.72 4.16 7.12 14.40 4.621969 3.27 4.96 8.80 17.60 5.511970 3.86 5.92 10.08 20.24 6.581971 4.95 7.20 11.28 28.00 8.001972 6.12 8.96 12.32 28.80 9.961973 10.12 10.40 20.00 56.00 11.561974 15.17 14.40 24.00 64.00 13.961975 20.71 24.00 34.00 112.00 19.731976 28.23 35.20 56.00 176.00 28.441977 41.00 48.00 76.40 280.00 I0.98
I/ iaRe rates for 1975, 1976 and 1977 are for December. Earlier ratpet are annual rstes convertd to
December rates on tne basis of the :wlatlonship betwen anymal and December rat-s in 107c
(paRe 2 of 2)
Appendix Table II
Daily Wages - (cont'd)(in Cruzeiros)
CONSTRUCTION____________________________ Urban
Rural Unskilled Skilled Foremen Minimum Wage
Brazil (excl. Sao Paulo)1968 2.64 4.05 7.90 16.28 4.571969 3.14 4.87 9.67 20.32 5.501970 3.86 5.86 11.06 24.11 6.601971 4.81 7.10 13.13 35.07 7.971972 5.86 8.59 16.13 36.48 9.571973 8.65 9.97 19.89 56.86 11.101974 13.97 13.92 24.77 81.42 13.451975 19.33 21.10 36.06 116.25 19.001976 26.48 31.35 53.75 179.39 27.401977 36.33 44.42 74.78 287.03 39.48
Brazil (cinc. Sao Paulo)1968 2.79 4.36 8.60 17.00 4.671969 3.31 5.01 10.44 19.43 5.621970 4.05 6.02 11.48 22.41 6.741971 5.06 7.31 13.82 31.55 8.141972 6.16 8.74 16.74 43.44 9.731973 9.00 10.81 21.92 56.44 11.301974 14.62 16.10 27.10 87.45 13.671975 20.12 22.30 37.03 122.78 19.321976 28,09 32.28 53.03 187.10 27.841977 39.50 45.90 73.63 274.35 40.12
- 113 - (a e 1 Of 2)
Appendix T-,;ie III
Daily Wages Deflated by the C.O.L. index(in 1977 Cruzeiros)
CONSTRIICTION-rban
Rua Unskilled Skilled Foremen___ inims
Natal (RGN)1968 16.42 22.00 33.33 U.67 24.421969 15.23 22.01 29.53 1C7.38 24.431970 16.39 23.11 30.22 168.89 2<.671971 17.22 23.77 32.08 143.401972 20.21 25.12 34.38 125.62 27.9,-1973 24.01 22.81 33.54 107.33 26.531974 32.79 26.77 39.96 79.92 24.651975 27.80 23.95 45.76 137.28 26.6i1976 28.28 27.66 52.59 173.54 26.531977 31.25 24.34 42.27 183.80 26.80
Salvador (Bania)1968 21.92 28.00 56.67 183.33 31.081969 20.34 26.85 64.43 161.07 29.801970 24.00 26.67 66.67 179.11 29.611971 23.63 27.17 64.15 245.28 30.191972 24.13 28.43 66.12 198.35 31.571973 27.94 26.83 59.02 268.32 29.821974 37.04 26.97 57.93 307.66 27.291975 38.36 36.60 68.56 305.06 29.501976 36.94 31.55 62.06 315.54 29.331977 34.00 33.08 58.82 292.24 29.57
Rio de Janeiro (Rio)1968 25.58 36.00 80.00 156.67 40.001969 23.76 34.90 75.17 161.07 38.791970 24.61 34.67 66.67 148.89 38.501971 24.34 .35.47 69.81 187.92 39.431972 27.02 37.02 81.65 178.51 41.161973 30.72 34.89 72.44 201.24 38.771974 33.39 35.96 65.93 233.75 34.861975 38.09 38.13 68.64 228.79 37.611976 38.08 42.07 70.99 241.91 37.391977 40.44 44.11 73.52 294.08 37.66
Sao Pa,,In (S.P 11968 30.75 40.00 80.00 150.00 40.001969 29.26 34.90 77.32 123.49 38.791970 28.94 34.67 67.11 111.11 38.501971 31.18 35.85 69.81 125.28 39.431972 33.10 37.02 72.73 220.17 41.161973 37.50 40.25 83.18 187.29 38.771974 46.45 47.94 75.92 239.74 34.861975 47.67 45.76 73.22 251.68 37.611976 49.96 44.18 68.36 260.32 37.391977 54.22 44.11 66.17 235.56 37.66
Curitiba (Parana)1968 28.50 34.67 57.33 68.67 36.331969 28.39 34.36 57.4! 80.54 35.171970 29.56 35.11 60.85 72.89 35.061971 33.16 36.60 64.15 83.02 36.461972 31.78 38.68 67.44 88.26 38.181973 35.99 36.76 66.81 92.84 35.791974 41.81 39.96 64.93 119.87 32.411975 50.43 42.71 76.27 140.33 34.9;1976 42.75 42.07 78.35 147.25 34.711977 42.27 36.76 69.84 171.67 34.96
Belo Horizonte (Minas Gerais)1968 22.67 34.67 59.33 120.00 38.501969 21.95 33.29 59.06 118.12 36.981970 21.44 32.88 56.00 112.44 36.561971 23.35 33.96 53.21 132.08 37.741972 25.29 37.02 50.91 119.G1 41.161973 33.94 34.89 67.08 187.83 38.771974 37.88 35.96 59.94 159.82 '* ̂1975 39.49 45.76 68.64 213.55 37.6i1976 37.12 46.28 73.62 Z31.40 37.391977 37.68 44.11 70.21 257.32 37.66
(page 2 of 2)Appendix Table III
Daily Wages Deflated by the C.O.L. index - (Cont'd)(in 1977 Cruzeiros)
CONSTRUCTIONUrban
Rural Unskilled Skilled Foremen Minimum Wage
Brazil (excl. Sao Paulo)1968 22.00 33.75 65.83 135.67 38.081969 21.07 32.68 64.90 136.38 36.911970 21.44 32.56 61.44 133.94 36.671971 22.69 33.49 61.93 165.42 37.591972 24.21 35.50 66.65 150.74 39.551973 29.01 33.44 66.71 190.71 37.231974 34.89 34.77 61.86 203.33 33.591975 36.85 40.23 68.75 221.64 36.231976 34.81 41.22 70.67 235.85 36.021977 33.39 40.82 68.72 263.78 36.28
Brazil (incl. Sao Paulo)1968 23.25 36.33 71.67 141.67 38.921969 22.21 33.62 70.07 130.40 37.721970 22.50 33.44 63.78 124.50 37.441971 23.87 34.48 65.19 148.82 38.4o1972 25.45 36.12 69.17 179.50 40.211973 30.19 36.25 73.52 189.30 37.901974 36.51 40.21 67.68 218.39 34.141975 38.36 42.52 70.61 234.10 36.831976 36.93 42.44 69.72 245.99 36.601977 36.30 42.18 67.67 252.13 36.87
- 115 -
Appendix Table IV
Ratio of wage per mDnth of permanent farmworkers to wage per day of casual workers
1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977
Maranhao 32 28 30 32 30 31 28 29 25 23 23 24Ceara 33 27 31 28 29 27 27 27 24 24 25 25Natal (RGN) 28 29 31 28 29 30 25 25 20 24 23 24Paraiba 28 24 26 27 25 25 25 23 21 20 20 21Pernambuco 25 26 26 28 27 28 26 23 21 19 21 24Alagoas 28 25 26 31 30 34 32 26 22 23 20 19Sergipe 29 26 23 22 26 26 29 32 24 20 22 20Bahia 30 27 27 29 27 27 27 27 21 20 20 21Minas Gerais 31 31 29 27 28 27 27 26 24 24 25 24Espirito Santo 29 28 27 26 27 30 28 25 23 24 25 23Rio de Janeiro 29 27 27 26 29 30 29 29 26 24 25 25Parana 24 26 25 25 26 25 27 26 25 21 23 23Santa Catarina 24 25 24 25 25 23 25 23 19 19 21 21R.G.S. 26 29 26 26 26 28 26. 23 21 20 21 23
Brazil (excl. Sao Paulo) 30 30 29 29 30 29 29 28 24 24 24 26
- 116 -
Appendix Table V
Casual Rural Laborers' Wages, State of Sao Paulo: Daili Rates
(in 1977 cruzeiros)
1948 17.28 1963 16.42
1949 19.87 1964 22.29
1950 22.81 1965 22.81
1951 25.75 1966 19.87
1952 24.19 1967 19.35
1953 21.08 1968 19.70
1954 20.56 1969 18.49
1955 21.25 1970 18.66
1956 20.39 1971 19.53
1957 19.35 1972 20.05
1958 18.84 1973 22.39
1959 17.11 1974 29.54
1960 16.59 1975 30.65
1961 16.42 1976 34.28
1962 16.24 1977 37.96
RECENT PAPERS IN THIS SERIES
No. TITLE OF PAPER AUTHOR
322 Ability in Pre-Schoolers, Earnings, and R. GraweHome-Environment
323 Priorities in Education: Pre-School; M. SmilanskyEvidence and Conclusions (consultant)
324 Tropical Root Crops and Rural Development T. Goering
325 Costs and Scale of Bus Services A.A. Walters
326 Social and Cultural Dimensions of Tourism R. Noronha(consultant)
327 Investment in Indian Education: Uneconomic? S. Heyneman
328 Nutrition and Food Needs in Developing 0. KnudsenCountries P.L. Scandizzo
329 The Changing International Division of B. BalassaLabor in Manufactured Goods
330 Application of Shadow Pricing to Country L. SquireEconomic Analysis with an Illustration from I.M.D. LittlePakistan
331 A Survey of the Fertilizer Sector in Ir.dia B. Bumb (consultant)
332 Monitoring and Evaluation in the PIDER M. CerneaRural Development Project - Mexico
333 Determinants of Private Industrial A. Pinell-SilesInvestment in India
334 The "Graduation" Issue in Trade Policy I. FrankToward LDCs (consultant)
335 Balancing Trickle Down and Basic Needs M. SelowskyStrategies: Income Distribution Issuesin Large Middle-Income Countries withSpecial Reference to Latin America
336 Labor Force, Employment and Labor Markets L. Squirein the Course of Economic Development
337 The Population of Thailand: Its Growth S. Cochraneand Welfare
338 Capital Market Imperfections and Economic V.V. BhattDevelopment A.R. Roe
-2-
5No. TITLE OF PAPER AUTP.OR
339 Behavior of Foodgrain Production J. Sarmaand Consumption in India, 1960-77
340 Electric Power Pricing Policy NI. Munasinghe
341 State Intervention in the Industrialization A. Chcksiof Developing Countries: Selected Issues
342 Policies for Efficient and Equitable J. LinnGrowth of Citites in Developing Countries
343 The Capital Goods Sector in LDCs: A Case J. Datta Mitrafor State Intervention?
344 International Technology Transfer: Issues F. Stewartand Policy Options
345 Family Planning Programs: An Evaluation R. Cucaof Experience
346 Prospects for Traditional and Non- D. HughartConventional Energy Sources inDeveloping Countries
347- National Urbanization Policies in Developing B. RenaudCountries
348 Private Direct Foreign Investment in K. BillerbeckDeveloping Countries Y. Yasugi (consultants)
349 Adjustment Policies and Problems in M4. WolfDeveloping Countries
350 Energy Options and Policy Issues in T. ByerDeveloping Countries D. Fallen-Bailey
351 Growth and Equity in Semi-Industrialized J. BergsnanCountries
352 Capital Flows and Developing Country Debt J. Katz
353 Trade Policy for Developing Countries D. Reesing
354 Development Problems of Mineral-Exporting G. NankaniCountries
355 The Global Framework R. Cheetham, S. GuptaA. Sch-wartz
Recommended