wage differentials in brazil: theory and...

Wage Differentials in Brazil: Theory and Evidence

JORGE SABA ARBACHE

The investigation of wage determination and wage differentials indeveloping countries has concentrated on the effects of humancapital and different sources of segmentation associated withinstitutional arrangements and structural characteristics onearnings. In this article, we use micro-data for Brazil for the 1980sand 1990s to test several competitive theories, and models withsegmentation explained by efficiency wages. We find thatunmeasured abilities and efficiency wage models seem to play arole in wage determination, while compensating differentials andtransitory difference theories were found to be irrelevant to wagesformation.

I . INTRODUCTION

One of the most intriguing issues in labour economics has been the patternof inter-industry wage differentials. Economists have found a remarkableregularity in wage dispersion within and between countries, even afterwages have been controlled for differences in human capital, occupation,and other variables [Krueger and Summers, 1987, 1988; Gittleman, andWolff, 1993]. These findings suggest that wage differentials are compatiblewith the functioning of capitalist economies, and thus cast doubt on theappropriateness of competitive theories. Besides competitive arguments,segmentation arising from efficiency wage models has occupied a central

Jorge Saba Arbache, Professor of Economics at the University of Brasília. Correspondenceaddress: Departamento de Economia, Universidade de Brasília, Brasília DF, Caixa Postal 04302,CEP 70910-900, Brazil. E-mail: [email protected]. The author would like to thank Alan Carruth,Andy Dickerson, Owen O’Donnell, Peter Sanfey, Pablo Fajnzylber, the seminar participants atthe Labour Market in LDC’s Seminar – DESG/ESRC, Tenth Annual Conference of the EuropeanAssociation of Labour Economists, University of Miskolc, University of Brasília, FederalUniversity of Minas Gerais/Cedeplar, and two anonymous referees for their useful comments andsuggestions on early versions of this paper. The usual disclaimer applies. Research grant No.300146/1000-0 from Conselho Nacional de Desenvolvimento Científico e Tecnológico isgratefully acknowledged.

The Journal of Development Studies, Vol.38, No.2, December 2001, pp.109–130PUBLISHED BY FRANK CASS, LONDON

382jds04.qxd 30/11/2001 12:28 09

role in the explanation of wage differentials in developed countries [Dickensand Katz, 1987; Krueger and Summers, 1988; Katz and Summers, 1989].Models dealing with discrimination [Daneshvary, 1993; Macpherson andHirsch, 1995], rent-sharing and union action [Blanchflower et al., 1996; DiNardo et al. 1996] also have an important role in the determination of wagedifferentials.

In the case of developing countries, however, investigation of wagedifferentials has focused, firstly, on the contribution of human capitalvariables to earnings [Corbo and Stelcner, 1983; Lam and Levison, 1992;Psacharopoulos and Velez, 1992; Yamada, 1996], and secondly, on thestylised facts that labour market is segmented along the lines of modern-traditional sectors [Harris and Todaro, 1970], formal–informal sectors[Tokman, 1983; Castells and Portes, 1989], public–private sectors[Lindauer and Sabot, 1983; Macedo, 1985; Fields and Wan, 1989; Teal,1996], and foreign-national ownership companies [Morrison, 1994; Teal,1996]. In addition, the effects of restrictive labour legislation [Thomas andValleé, 1996, Marshall, 1996; Rama, 1995, 1997; Carneiro, 1997],minimum wages [Morrison, 1994], and trade unions [Teal, 1996; Arbache1999; Arbache and Carneiro, 1999] are also believed to affect wagedispersion, while competitive theories and segmentation explained byefficiency wage models have received less attention as a source of wagevariance.

In Brazil, wage differentials have long been a source of attention dueto the high income inequality that characterises the country [Langoni,1973; Bacha and Taylor, 1978; Cowell et al., 1996]. Studies of inter-industry wage differentials in the 1980s based on micro-data have foundthat about 50 per cent of inter-industry wage dispersion could beexplained by standard wage equations [Pinheiro and Ramos, 1994; Gaticaet al. 1995]. The unexplained wage dispersion was attributed toalternative sources, especially industry affiliation. Competitive theories,such as compensating differentials and unmeasured abilities, wereconsidered as irrelevant, despite the fact that these models were assessedvery indirectly in these papers. Furthermore, these studies did not test, forexample, the importance of job conditions and the impact of turnovercosts on wage dispersion, which is regarded as an important source ofwage differentials [Krueger and Summers, 1988; Katz and Summers,1989]. Turnover may be particularly important for Brazil, not onlybecause of training aspects, but also as a result of dismissal costs, whichare particularly high in this country [Amadeo and Camargo, 1993]. Allthese variables may play a crucial role in wage determination, and theirabsence prevent us from analysing their effects on wage dispersion. As aconsequence, we may overestimate the variance in wages.

110 THE JOURNAL OF DEVELOPMENT STUDIES

382jds04.qxd 30/11/2001 12:28 Page 110

The aim of this article is to test the importance of several competitiveand non-competitive theories of wage determination. The analysis presentstwo contributions. First, we empirically test many wage determinationtheories for Brazil. Second, the assessment is carried out in a period full ofeconomic reforms, including several stabilisation plans, trade liberalisation,privatisation of state companies and deregulation of markets, thus enablingan evaluation of the effects of structural adjustments on the wage formationand wage differentials of an industrialising country.1

To the extent that the understanding of the causes generating wagedifferentials can provide normative and positive implications to issues suchas income distribution, educational and training programmes, employment,trade and industrial policies, empirically testing a wide range of wagedifferentials theories is essential for development of policies and forassessing their effectiveness.

The article is organised as follows. Section II presents some stylisedfacts on inter-industry wage differentials and a theoretical review of theoriesof wage differentials and their explanation of wage dispersion. Section IIIempirically assesses the importance of several theories in explaining thewage variance for selected years from the 1980s and 1990s. Section IVconcludes.

II . THEORIES OF WAGE DIFFERENTIALS

The empirical literature has shown that wage differentials are a pervasivephenomena in market economies, and are strongly associated with theindustry the worker is affiliated to, which is the critical issue in the duallabour market theory from Doeringer and Piore [1971]. There are threestylised facts that challenge the traditional theories of wage differentials: (i)temporal stability of wage structure; (ii) similarity of wage structuresbetween countries in different stages of economic development; (iii) certainindustries pay high wages to all workers, while other industries pay low-wages to all workers for given occupations and human capitalcharacteristics. The next paragraphs present theories that attempt to shedlight on these issues.

Transitory Differences

Inter-industry wage differentials can be a consequence of inter-sectoralshifts in labour demand that follow the changing demand for specificproducts. The changes in product demands can be due to technologicalinnovations, external competition, changes in tastes, alterations in relativeprices or any other type of exogenous (and unexpected) changes.Accordingly, industries which face increasing demands for their products

111WAGE DIFFERENTIALS IN BRAZIL

382jds04.qxd 30/11/2001 12:28 Page 111

tend to pay relatively higher wages, since they have a tendency to makeexcess profits during the transitory period. In the presence of imperfectinformation or adjustment costs, the change in demand may generate notonly frictional unemployment, but also wage differentials during the periodof adjustment. Although the process is not instantaneous, wage differentialstend to decrease and converge in the long run toward the market clearingwage.

Many authors find high temporal stability and high correlation of theinter-industry wage differentials of many countries [Krueger andSummers, 1987; Gittleman and Wolff, 1993]. Krueger and Summers[1988], Abuhadba and Romaguera [1993], Arai [1994], Morrison [1994],Pinheiro and Ramos [1994] among others, find a high temporal wagestability for countries in which labour markets and institutions governingthe wage bargain are very different (such as Sweden, the United States,Ecuador and Brazil). These findings cast doubts on the relevance of thetransitory difference model to explain wage differentials. Krueger andSummers [1987] address this empirical evidence to the market andtechnology structures which are intrinsically associated with industry,even in less developed countries. The transitory differences theory doesnot explain the stylised facts that employers pay similar wage premia toworkers with different levels of human capital, and in jobs with differentlevels of risk.

Unmeasured Abilities

The unmeasured ability hypothesis is based upon the idea that employersscreen workers not only by their measured abilities, like schooling andexperience at work, but also by their unobserved skills. This sorts them intoindustries according to the demand for measured and unmeasuredcharacteristics.2 Since unmeasured abilities are not, by definition, directlycaptured by datasets, these productive characteristics are not accounted forin ordinary econometric estimations. In order to justify the inter-industryaverage wage differentials, we may have to assume that a high averagewage industry – that is an industry that pays relatively high wages toworkers whatever their human capital characteristics – has adisproportionate number of workers with high measurable andunmeasurable abilities.

One possible way to explain the common empirical evidence that wagesare correlated across occupations is that there is an endogenous labourdistribution according to employers’ demand for workers’ abilities. Thiswould lead to a positive relationship between technology and measured andunmeasured workers’ abilities, thus explaining the differences betweenindustries’ average wages [Groshen, 1991].


382jds04.qxd 30/11/2001 12:28 Page 112

Firm-specific training may play an important role in the explanation ofwage differentials. The argument is that a worker who receives specifictraining increases his/her skills and abilities demanded by a particular firmor industry. As a consequence, his/her marginal productivity and relativewage are higher than in any other firm or industry, thus causing wagedifferentials between workers who are ‘supposed’ to be equally skilled[Abowd et al., 1999]. In these circumstances, despite having firm-specifictraining, the worker’s opportunity cost is still the ongoing wage, sincehis/her firm-specific skills are not well recognised or valued by otheremployers.

Compensating Differentials

The compensating differentials theory argues that high wages are simplycompensating differences for job conditions within and among firms.Employers who offer different working conditions in terms of safety (forexample, injury rate), uncommon risk of lay-off, physical exertion, overtimeworked, and undesirable environment (for example, noise, humidity andventilation), should appropriately compensate workers for their jobattributes. In order to explain wage differentials, one could argue that jobattributes vary from industry to industry or even from firm to firm, and thusordinary controls for job conditions would not properly capture theparticularities and effects of job attributes on wages. The compensatingdifferentials theory does not explain why some industries pay high (low)wages to all occupations, since there is no a priori reason to pay a whitecollar worker the same premium paid to a blue-collar worker who usuallyworks in harder conditions.

Monitoring and Shirking Model

The shirking model has received extensive treatment in the literature[Shapiro and Stiglitz, 1984; Bulow and Summers, 1986]. Its basic idearelies on the assumption that workers have some discretion concerningtheir work effort, and there is a continuous monotonic cost to monitorthem. In order to motivate equally able workers to strive hard in theirperformance, employers pay a wage higher than the worker’s opportunitycost, which is given by the market wage or by the level of unemploymentbenefit. The higher the difference between the paid wage and thealternative wage, or the higher the unemployment rate, the higher the fearof being fired, since in these circumstances the job becomes moreattractive, encouraging the worker to try harder. The shirking modelpredicts that wage differentials appear as a consequence of the differencein monitoring costs between firms or industries. Although verysuggestive, this theory is not able to explain why wages are correlated


382jds04.qxd 30/11/2001 12:28 Page 113

across occupations, since one would not expect that different occupationswould require the same monitoring level.

Turnover Costs Model

The general assumption of this model is that turnover is costly to firms. Thisleads them to create wage policies to avoid such costs from quits, trainingand hiring new workers, which are, by supposition, basically provided bythe firm [Salop, 1979; Stiglitz, 1986; Weiss, 1980]. The microeconomicfoundations of these costs are explored in Oi [1962], and his main argumentis that labour may be a fixed, rather than a variable cost. Each occupation ina specific firm requires specific training to enable an ordinary worker toincrease his productivity, since the firm and the occupation possesspeculiarities that require a complementarity between capital and the labourforce. A trained worker is defined as a quasi-fixed production factorbecause the firm incurs costs for hiring and training, which are consideredto be an investment of the firm in its labour force. The higher the fixed costsin recruitment and training to improve labour’s productivity, the longer theexpected period the worker will be engaged in the firm. Thus, in order tominimise turnover, the firm offers a wage premium. These arguments areused to justify the positive relation between job tenure and wagedifferentials. In order to explain inter-industry wage differentials, theturnover models have to assume that an industry that pays on average apositive wage premium should have a high number of positions that requirefirm-specific trained workers, and that fixed costs cut across occupations,which would justify the correlation of wage premiums for say, white andblue collar workers.

Sociological Model

Human resource managers have two choices to elicit productivity ofworkers: through a rigid monitoring system and disciplinary policy, or anincentive system, which motivates employees through a mix of pecuniaryand non-pecuniary inducements. The sociological model relies on thesecond choice and believes that industrial relations are governed bymotivations. Akerlof [1982, 1984] introduces the idea that there is morethan a pure bilateral exchange of labour for money between firm andworker. Akerlof stresses that social conventions and norms of the workgroup have a strong effect on the workers’ attitude. Workers acquiresentiment for each other and for the firm, and as consequence, theydevelop a commitment and loyalty to the firm’s aims, which motivatesthem to work hard. But they expect a return or reciprocity from the firm,which is hoped to be given in a ‘fair’ package that includes not only ahigher relative wage compared with what they would receive elsewhere,


382jds04.qxd 30/11/2001 12:28 Page 114

their past wages, and unemployment rate but also non-pecuniarycompensations such as lower pressure of supervisors, and the firm’sattitude to laid-off workers. The notion of fairness also relies on howprofitable the firm is, and how essential the worker is to the firm and to itsgoals.

The most relevant contribution of this model to the understanding ofwage differentials is its explanation for the common empirical finding onthe correlation of wage premia across occupations, which would be due toworkers’ collective attitudes. Although this theory has an interestingjustification for the correlation of wage structures within a firm, it is hard tobelieve that workers in different occupations have sufficient discretionarypower over their work to lead them to affect significantly their ownproductivity in order to establish an exchange relationship with the firm.Also, it is not clear why a productive (and altruistic) worker who is amember of a team would look after a team-mate who is less productive, butbenefits from his/her effort.

III . ASSESSING THE THEORIES OF WAGE DIFFERENTIALS

Our empirical analysis on wage determination is based on micro-data fromthe Pesquisa Nacional por Amostras por Domicílio (PNAD), from theInstituto Brasileiro de Geografia e Estatística (IBGE), and on aggregate datafrom the Relatório Anual de Informações Sociais (RAIS), from the Ministryof Labour, for various years between 1984 and 1998.3 The investigation ofthe wage structure is carried out for the 2-digit manufacturing industrylevel.4 The micro-data sample is composed of (labour market active)individuals, non-employers, between the ages of 18 and 65, working 21 ormore hours a week (full-time) in the main job.

Wage Structure Stability and Transitory Differences

The transitory difference theory is particularly appealing in the analysis ofthe Brazilian labour market due to the economic reforms carried out in the1980s and 1990s. In order to assess this theory, we compute the temporalSpearman rank correlation coefficient for raw inter-industry wagedifferentials based on PNAD data. To estimate the wage premia weemployed the Haisken-DeNew and Schmidt [1997] methodology. Theresults in table one show that the wage structure is quite stable. This resultis striking, since one could expect that the stabilisation of inflation, tradeliberalisation and the other reforms undertaken in the 1980s and 1990swould change the wage structure. This however does not seem to be thecase, suggesting that the wage structure is not responsive to short runeconomic reforms.


382jds04.qxd 30/11/2001 12:28 Page 115

TA

BL

E 1

SP

EA

RM

AN

RA

NK

CO

RR

EL

AT

ION

OF

NO

N-C

ON

TR

OL

LE

D W

AG

E D

IFF

ER

EN

TIA

LS

. 2

-DIG

IT I

ND

US

TR

Y L

EV

EL

1984

1985

1986

1987

1988

1989

1990

1992

1993

1995

1996

1997

1998

1984

1.9

71.8

86.9

52.9

01.9

38.9

47.9

31.9

69.9

23.9

57.9

26.9

1319

851

.860

.973

.894

.891

.940

.919

.955

.925

.960

.919

.912

1986

1.9

03.9

35.8

96.9

31.9

06.9

26.9

42.8

87.9

26.9

1619

871

.926

.906

.965

.921

.957

.947

.926

.936

.927

1988

1.8

94.9

79.9

31.9

42.9

45.8

91.9

27.9

3019

891

.921

.948

.944

.913

.865

.939

.858

1990

1.9

65.9

75.9

77.9

32.9

62.9

6419

921

.960

.973

.917

.970

.953

1993

1.9

71.9

42.9

70.9

4719

951

.935

.987

.984

1996

1.9

13.9

4719

971

.962

1998

1

Not

e:A

ll co

effi

cien

ts a

re s

igni

fica

nt a

t the

fiv

e pe

r ce

nt le

vel.

382jds04.qxd 30/11/2001 12:28 Page 116

TA

BL

E 2

SP

EA

RM

AN

RA

NK

CO

RR

EL

AT

ION

OF

CO

NT

RO

LL

ED

WA

GE

DIF

FE

RE

NT

IAL

S.

2-D

IGIT

IN

DU

ST

RY

LE

VE

L

1984

1985

1986

1987

1988

1989

1990

1992

1993

1995

1996

1997

1998

1984

1.9

38.8

66.9

04.8

75.8

68.8

78.9

14.8

96.8

56.8

83.8

18.8

0919

851

.862

.923

.860

.843

.879

.835

.851

.801

.860

.768

.739

1986

1.8

84.8

73.8

86.8

99.8

64.8

51.8

66.8

35.8

58.8

3419

871

.931

.85

5.8

82.9

00.9

35.8

91.9

06.8

45.8

3619

881

.866

.856

.916

.891

.878

.862

.812

.838

1989

1.8

84.8

32.8

14.7

38.7

47.7

70.6

7119

901

.817

.830

.735

.770

.719

.671

1992

1.9

35.9

09.8

96.8

57.6

7719

931

.906

.934

.875

.868

1995

1.9

45.9

06.9

7119

961

.873

.909

1997

1.9

2319

981

Not

e:A

ll co

effi

cien

ts a

re s

igni

fica

nt a

t the

fiv

e pe

r ce

nt le

vel.

Con

trol

var

iabl

es a

re f

ive

regi

on d

umm

ies,

edu

catio

n, e

xper

ienc

e, e

xper

ienc

e sq

uare

, hea

dof

fam

ily, g

ende

r, w

ork-

card

, rac

e, m

etro

polit

an a

rea,

urb

an/r

ural

are

a.

382jds04.qxd 30/11/2001 12:28 Page 117

To some extent, the results from table one should be regarded carefully,since the inter-industry wage differentials are not controlled by workerscharacteristics. Table 2 shows an alternative estimation of temporalcorrelation that accounts for control variables. The results now present aslightly different picture in quantitative terms as compared to Table 1, butthe qualitative results remain the same. The correlation coefficients are asstable as those in the previous table, supporting the finding that the wagestructure seems not to be affected by economic changes, even after controlvariables are included in wage equations.

Our results support the stylised fact that the wage structure is stable overtime. The temporal stability of the wage structure casts doubt on the effectsof transitory labour demand and short-run shocks on wages. Overall, theresults indicate that the wage structure is permanent, and that the wagedispersion is not explained by movements in relative wages across themanufacturing sector.

Unmeasured Abilities

Considering that the variables in our micro-data sets may not satisfactorilycapture all productive individual attributes, one could argue that theunexplained wage differentials reflect unmeasured labour quality. As we aredealing with individual data, the issue is then whether differences in averagewages paid in different industries are related to differences in average levelsof unmeasured abilities that are captured by employers but not by theeconometrician.

Since there are no longitudinal data in Brazil to carry out first-differenced estimations to test unmeasured abilities as in Krueger andSummers [1988], Murphy and Topel [1990], and Gibbons and Katz [1992],we adopt an alternative approach to assess the relevance of this theoryfollowing closely the weak versus strong version of the screeninghypothesis [Psacharopoulos, 1979; Cohn et al., 1987].5 We estimate wageequations for a sub-sample of workers with tenure less than one year in thepresent job, and for a sub-sample with tenure larger than ten years, othervariables held constant. The first group of workers is labelled as‘beginners’, while the second is labelled as ‘long-tenure workers’. Weassume that the labour assignment of beginner workers is given byscreening. Thus, education works as an informational device for identifyingworkers’ abilities. If we take workers with similar productive observablecharacteristics, then we hypothesise that firms may not identify workers’differences in abilities and capabilities, at least at the very beginning of theworkers’ contract.6 Therefore, beginners tend to be paid the industry’saverage wage for a given job. However, as time goes by, the workersacquire firm-specific training, and unmeasured abilities and talents for the


382jds04.qxd 30/11/2001 12:28 Page 118

given job do appear and become recognisable through their differences inproductivity. As a consequence, earnings of workers with similarmeasurable productive characteristics tend to diverge rather than toconverge over time. The direct implication is that beginner workers’ wagesare less dispersed than long-tenure workers’ wages.

In order to test the above proposition, Table 3 shows the weighted andadjusted inter-industry wage dispersion of controlled and non-controlledwage equations for beginners and long-tenure workers. Seeking to have asensible sample size, we pooled data from PNAD of 1990 to 1998. PNADsfrom the 1980s do not have the variable ‘tenure’, which prevents anassessment for this period. The results in table three show that the wagedispersion of long-tenure workers estimated with and without controls areabout two times higher than that of beginners, thus supporting ourhypothesis. This finding contradicts the strong screening hypothesis.

If we assume that ability is complementary to schooling [Rosen, 1989],then the return rate of schooling will be higher for the more able workers,since this variable is capturing the effect of non-observed abilities [Card,1998]. Correspondingly, the rate of return of education of long-tenureworkers is 0.115, while it is 0.081 for beginners, thus suggesting that tosome extent, unmeasured abilities and unrecognised talents are captured bythe education variable.

Compensating Wage Differentials

Compensating wage differentials is an appealing theory for assessingindustry wage differentials in Brazil due to the substantial heterogeneity inthe technological level of its establishments and therefore to substantial


TABLE 3WAGE DISPERSION OF BEGINNER AND LONG-TENURE WORKERS

Data Weighted and adjusted SD

Non-controlled(1) < one-year tenure .145(2) > ten -year tenure .293Controlled(3) < one-year tenure .070(4) > ten -year tenure .147

Notes: Data for 1990 to 1998 are pooled. Samples sizes are 15,451 for beginner and 12,321 forlong-tenure workers. The control variables are education, experience, experiencesquared, work card, head of family, region dummies, metropolitan area, urban/rural, race,gender. F-tests that wage differentials jointly equal zero were rejected at the .001 level.The inter-industry wage differentials were estimated using the Haisken-DeNew andSchmidt [1997] approach.

382jds04.qxd 30/11/2001 12:28 9

heterogeneity in job conditions. There are national and multinationalcompanies in the same industry using high technology and modernmanagerial systems, alongside low-scale production firms employinginformal workers in hazardous and insecure workplace environments. Sincework condition variables are not included in our wage equation estimations,it could be argued that wage dispersion for equally productive workerscould result from compensating differentials.

Unfortunately, there is no survey on individual job attributes in Brazil.We therefore adopt an aggregate analysis approach. We regressmanufacturing inter-industry wage differentials on industry injury rate,firing rate, and temporary contract rate. All these variables expressunpleasant job conditions. Injury, firing and temporary contract rates weredrawn from RAIS. We expect to find a positive correlation, thus implyingthat wage premia compensate for unpleasant job conditions.

Table 4 reports the results for OLS cross-section regressions.7 Apartfrom the injury rate for 1986, all other statistically significant coefficientsare negatively correlated with the wage premium variable, thus opposingthe theoretical prediction. Thus, industries that experience higher levels offiring and temporary job contracts are also the ones that pay lower wagepremium.8 These results seem to be in accordance with LDC’s labourmarket characteristics. Because good job opportunities are scarce and socialsecurity is limited, most workers are forced to take the available rather thanthe desired job. Polacheck and Siebert [1993] point out that not only tastesaffect attitudes of workers to risky jobs, but also economic circumstancessuch as poverty, which push poor people to take unpleasant and riskier jobswithout being appropriately compensated. To some extent, such arguments


TABLE 4REGRESSION OF INTER-INDUSTRY WAGE DIFFERENTIALS

ON WORK CONDITIONS

Variable 1985 1986 1987 1988 1989 1990 1992 1993

On-the-job .145 .381* –.144 .880 –.014 .118 .102 –.149injury rateFiring rate –3.61* –.432 –2.07 –1.87** –3.78* –2.55* –3.09 –4.41*Temporary .078 –.046 –.103** –.138 –.043 –.045 –.053 –.031job rateNo. Obs. 21 21 21 21 21 21 21 21R-square .73 .83 .79 .73 .66 .68 .77 .77F-statistics 5.19 9.57 7.35 5.05 3.73 4.09 6.34 6.58

Notes: Other variables in the models are education, gender, urban and work card. (*) significantat the five per cent level. (**) significant at the ten per cent level. The inter-industry wagedifferentials were estimated using the Haisken-DeNew and Schmidt [1997] approach.

382jds04.qxd 30/11/2001 12:28 0

tend to support the negative signs. The compensating differentials theorydoes not appear to explain the wage determination in Brazil, at least at thetwo-digit manufacturing industry level. Further research however is neededusing both more detailed industry-level and occupation-level data.

Turnover Costs

The main assumption of the labour turnover model is that turnover is costlyto firms, which leads them to create wage policies to avoid the costsassociated with quits, training and hiring. To test this model’s predictionsthat different firms have different incentives to pay wage premia seeking toprevent turnover costs, we regress wage differentials on turnover. FollowingPencavel [1980], Dickens and Katz [1987], and Katz and Summers [1989],we expect to find a negative correlation, which would mean that high-wageindustries have lower turnover levels, while low-wage industries havehigher turnover levels.

Table 5 shows the results of such regressions. The coefficients ofturnover are always negative, as predicted, although they are statisticallysignificant in only five out of 12 years.9 Figure 1 provides more insights onthe issue showing a negative relation between turnover and wage premia.The figure was built using pooled data from 1985 to 1998. These evidencessuggest that turnover seems to play a role in the determination of wages inBrazilian manufacturing.

To evaluate the turnover model from another perspective, we followKrueger and Summers [1988] and Arai [1994] and regress inter-industrywage differentials on job tenure. Their argument is that if workers fromhigh-paying industries earn wage premia, we should expect to find apositive relation between job tenure and wage differentials, since a lowerturnover rate is reflected in longer job tenure.10 Table 6 reports thecoefficients on job tenure and show that, apart from 1990, the relationbetween the wage premium and tenure is always positive as expected,suggesting that the length of job tenure affect the wage premium. Figure 2shows the correlation between wage premium and tenure using pooled datafrom 1990 to 1998. The relation is positive as expected, supporting theprevious finding that turnover is a source of wage determination in Brazil.This result might be related to the high severance pay and the high costs oftraining associated with the low supply of skilled workers.

Monitoring and Shirking

In this model, a dispersion in wages between comparable work arises as aresult of the differences in monitoring costs between firms or industries. Inthis respect, it has been assumed that monitoring is more costly in largefirms due to the need to employ higher-skilled workers to match the capital-


382jds04.qxd 30/11/2001 12:28 Page 121

TA

BL

E 5

RE

GR

ES

SIO

N O

F I

NT

ER

-IN

DU

ST

RY

WA

GE

DIF

FE

RE

NT

IAL

S O

N T

UR

NO

VE

R

1985

1986

1987

1988

1989

1990

1992

1993

1995

1996

1997

1998

Tur

nove

r–.

029

–.04

2**

–.02

2–.

050*

*–.

107*

–.05

8*–.

027

–.06

7–.

010

–.09

2–.

054*

.018

No.

Obs

.20

2020

2020

2020

2020

2020

20R

-squ

are

.55

.72

.68

.71

.61

.66

.73

.67

.73

.77

.90

.78

F-st

atis

tics

3.42

7.31

6.18

7.04

4.43

5.44

7.70

5.87

7.91

9.55

26.8

110

.15

Not

es:

Oth

er v

aria

bles

in th

e m

odel

s ar

e ed

ucat

ion,

gen

der,

urba

n, a

nd w

ork

card

. (*)

sig

nifi

cant

at t

he f

ive

per

cent

leve

l. (*

*) s

igni

fica

nt a

t the

ten

per

cent

leve

l. T

he in

ter-

indu

stry

wag

e di

ffer

entia

ls w

ere

estim

ated

usi

ng th

e H

aisk

en-D

eNew

and

Sch

mid

t [19

97]

appr

oach

.

382jds04.qxd 30/11/2001 12:28 Page 122

intensive requirements, which are more common in larger plants[Hamermesh, 1980; Dickens and Katz, 1987]. Since skilled workers aremore difficult to monitor, there would have to be a positive relation betweenfirm-size and wages. In order to assess the prediction that larger firms payhigher wage premia, we compute the Spearman rank correlation coefficientbetween inter-industry wage differentials and industry average firm size (interms of employees). We expect to find high and positive correlation, asfound by Mellow [1982], Dickens and Katz [1987], Krueger and Summers[1988] and Brown and Medoff [1989].

To run the test, we use firm size data aggregated in industry level fromthe RAIS-Establishments, which accounts for the number of employeesfrom about 180,000 manufacturing firms each year for 1988, 1992 and 1993(see Appendix for details). We did not assess other years of the 1990s dueto changes on the 2-digit industry definition. The correlation results are0.43, 0.64, and 0.53, respectively, which are positive and statisticallysignificant as expected, thus providing some evidence of the monitoring andshirking models’ prediction for the explanation of wage differentials.However, since other theories also predicts this relation, our results on themonitoring and shirking theory should be regarded with caution.


FIGURE 1WAGE DIFFERENTIALS VERSUS TURNOVER

382jds04.qxd 30/11/2001 12:28 Page 123

Sociological Model

One important prediction of the sociological model is that firms pay high orlow-wages to all their workers because of norms, loyalty feeling, andfairness considerations. The literature has been relying on this model toexplain the high and stable correlation of wages among skilled and unskilledworkers, and among workers with different levels of seniority, given by jobtenure and age [Krueger and Summers, 1988; Gatica et al., 1995;Romaguera, 1991; Moll, 1993].


TABLE 6

REGRESSION OF INTER-INDUSTRY WAGE DIFFERENTIALS ON JOB TENURE

1990 1992 1993 1995 1996 1997 1998

Tenure –.041 .056* .030 .036* .009 .037 .061*No. Obs 21 21 21 21 21 21 21R-square .56 .76 .64 .80 .73 .88 .86F-statistics 3.88 9.95 5.37 12.16 8.16 22.5 19.86

Notes: Other variables in the models are education, gender, urban and work card. (*) significantat the five per cent level. The inter-industry wage differentials were estimated using theHaisken-DeNew and Schmidt [1997] approach.

FIGURE 2WAGE DIFFERENTIALS VERSUS TENURE

382jds04.qxd 30/11/2001 12:28 Page 124

We assess the above prediction by estimating the correlation coefficientof the inter-industry wage premia of workers belonging to different groupsof age, education and tenure. If the prediction applies, then we should findhigh and positive coefficients throughout. Table 7 reports the results forselected years in the period 1984–98. All coefficients are statisticallysignificant, suggesting the relevance of the sociological model to theexplanation of wage differentials. These results imply that the sociologicalmodel seems to play an important role in the determination of wages.Therefore, an industry that pays high (low) wages to some workers, say, themore educated, does pay high (low) wages to all workers, thus creating anelement of segmentation and rigidity in the wage structure.

IV. CONCLUSIONS

The investigation of the sources of wage determination and wage dispersionin developing countries has concentrated on the effects of both humancapital variables and different sources of segmentation associated with theinstitutional arrangements and structural characteristics of these economieson earnings. In this article, however, we test several competitive models,and labour market segmentation theories as described by efficiency wagemodels for Brazil. Our main results are the following. The wage structurehas not changed over 1984–98, a period marked by successive inflationstabilisation plans and market-oriented economic reforms. This findingstrongly support the stylised fact that the wage structure is stable over time.We found some evidence that unmeasured abilities affect wagedetermination, thus explaining the wage dispersion for workers withcomparable measurable productive traits. We found no evidence in favour


TABLE 7

INTER-INDUSTRY WAGE DIFFERENTIALS CORRELATION COEFFICIENTS

Samples 1984 1989 1993 1998

Age[less than 30 and more than 45] .678 .623 .781 .804Education[less than 4 and more than 10 years] .745 .509 .800 .660Tenure[less than 2 and more than 8 years] NA NA .540 .745

Notes: Variables in wage equations are race, female, head of family, education (whenappropriate), urban, metropolitan area, region. All coefficients are significant at the fiveper cent level. The inter-industry wage differential were estimated using the Haisken-DeNew and Schmidt [1997] approach.

382jds04.qxd 30/11/2001 12:28 Page 125

of the compensating wage differentials theory, which may have to do withthe low opportunity costs of the large majority of workers. The efficiencywage theory, represented by the turnover, monitoring and sociologicalmodels, seem to play an important role on the wage formation inmanufacturing.

The above findings explain, at least in part, the inter-industry wagedifferentials not explained by ordinary wage equations, and highlightsources of rigidities and segmentation in the Brazilian labour market.Regarding the policy implications, it appears that economic policies aimedat affecting different aspects associated with the manufacturing labourmarket might achieve better results if they considered the fact that thelabour market is rigid, along the lines of efficiency wages, while notneglecting the impacts of unmeasured abilities on wage determination.

final revision accepted January 2001

NOTES

1. Between 1986 and 1994, six stabilisation plans were introduced: Plano Cruzado, 1986,Plano Bresser, 1987, Plano Verão, 1989, Planos Collor I e II, 1990, and Plano Real, 1994,but only this last one successfully reached its aims. For details on the trade liberalisation andother reforms of the 1990s, see Moreira and Correa [1998] and Loser and Guerguil [1999].

2. One major strand of literature, the screening hypothesis, contends that education is used byemployers primarily as a screen to sort out workers with high ability. Thus, a certain rule isused to select workers. It may happen that uneducated individuals are passed over even ifthey are able in favour of those who are educated.

3. For detailed information on the datasets, see Appendix.4. We recognise that this level of aggregation misses a lot of variation. The aggregation level

is due to reliability of data and compatibility of PNAD and RAIS.5. The weak version of the screening hypothesis says that employers offer higher starting

salaries to the more educated relative to the less educated in the absence of any otherinformation on the new employees expected productivity. The strong version says thatemployers will continue to pay higher wages to the more educated, after the employee hasbeen with them for some time [Psacharopoulos, 1979: 181].

6. Layard et al. [1991] point out that the longer a worker stays in a job, the more their talentsare recognised.

7. We assessed the period 1985–93 only due to commensurability of job conditions data.8. We run fixed effects models (not reported), but the results suggested that industry fixed

effects are not appropriate to capture the relationship of job conditions and inter-industrywage differentials. The estimated coefficients of job conditions appeared very small andinsignificant, and the R-square and F statistics dropped dramatically, reaching very smallfigures.

9. Fixed effect models were estimated, but the results suggest very little variance of turnover-industry.

10. This procedure may perhaps be subject to an endogeneity problem. Tenure data are availablefrom 1990 only.


382jds04.qxd 30/11/2001 12:28 Page 126

REFERENCES

Abowd, J.M., Kramarz, F. and D.N Margolis, 1999, ‘High Wage Workers and High Wage Firms’,Econometrica, Vol.67, No.2, pp.251–333.

Abuhadba, M. and P. Romaguera, 1993, ‘Inter-Industrial Wage Differentials: Evidence fromLatin American Countries’, The Journal of Development Studies, Vol.30, No.1, pp.190–205.

Akerlof, G.A., 1982, ‘Labor Contracts as Partial Gift Exchange’, Quarterly Journal ofEconomics, Vol.97, No.4, pp.543–69.

Akerlof, G.A., 1984, ‘Gift Exchange and Efficiency-Wage Theories: Four Views’, AmericanEconomic Review, Vol.74, No.2, pp.79–83.

Amadeo, E.J. and J.M. Camargo, 1993, ‘Labour Legislation and Institutional Aspects of theBrazilian Labour Market’, Labour, Vol.7, No.1 , pp.157–80.

Arai, M., 1994, ‘An Empirical Analysis of Wage Dispersion and Efficiency Wages’,Scandinavian Journal of Economics, Vol.96, No.1, pp.31–50.

Arbache, J.S., 1999, ‘Do Unions Always Decrease Wage Dispersion? The Case of BrazilianManufacturing’, Journal of Labor Research, Vol.20, Summer, pp.425–36.

Arbache, J.S. and F.G. Carneiro, 1999, ‘Unions and Interindustry Wage Differentials’, WorldDevelopment, Vol.27, No.10, pp.1875–83.

Bacha, E. and L. Taylor, 1978, ‘Brazilian Income Distribution in the 60’s: Facts, Model Resultsand Controversy’, The Journal of Development Studies, Vol.14, No.3, pp.271–97.

Blanchflower, D.G.; Oswald, A.J. and P. Sanfey, 1996, ‘Wages, Profits, and Rent-Sharing’,Quarterly Journal of Economics, Vol.111, No.1, pp.227–251.

Brown, C. and J. Medoff, 1989, ‘The Employer Size-Wage Effect’, Journal of Political Economy,Vol.97, No.5, pp.1027–59.

Bulow, J.I. and L.H Summers, 1986, ‘A Theory of Dual Labor Market with Application toIndustrial Policy, Discrimination, and Keynesian Unemployment’, Journal of LaborEconomics, Vol.4, No.3, pp.376–414.

Card, D., 1998, ‘The Causal Effects of Education on Earnings’, in O. C. Ashenfelter and D. Card(eds.), Handbook of Labor Economics, Vol.3A, Amsterdam: Elsevier.

Carneiro, F., 1997, ‘The Changing Informal Labour Market in Brazil: Cyclicality versusExcessive Intervention’, Labour, Vol.11, No.1, pp.3–22.

Castells, M. and A. Portes, 1989, ‘World Underneath: The Origins, Dynamics, and Effects of theInformal Economy’, in A. Portes, M. Castells and L.A. Benton (eds.), The InformalEconomy. Studies in Advanced and Less Developed Countries, Baltimore, MD: JohnsHopkins University Press.

Cohn, E., Kiker, B.F. and M.M. de Oliveira, 1987, ‘Further Evidence on the ScreeningHypothesis’, Economics Letters, Vol.25, pp.289–94.

Corbo, V. and M. Stelcner, 1983, ‘Earnings Determination and Labor Markets: Gran Santiago,Chile – 1978’, Journal of Development Economics, Vol.12, No5.1–2, pp.251–66.

Cowell, F.A., Ferreira, F.H.G. and J. Litchfield, 1996, ‘Income Distribution in Brazil 1981–1990:Parametric and Non-Parametric Approaches’, Discussion Paper No. DARP 21, LondonSchool of Economics.

Daneshvary, N., 1993, ‘Earnings Differentials Between Natives and Immigrants with CollegeDegree’, The American Economist, Vol.37, No.1, pp.46–52.

Dickens, W.T. and L.F. Katz, 1987, ‘Inter-industry Wage Differences and IndustryCharacteristics’, in K. Lang, and J. Leonard (eds), Unemployment and the Structure ofLabor Markets. Oxford: Basil Blackwell.

DiNardo, J. Fortin, N.M. and T. Lemieux, 1996, ‘Labor Market Institutions and the Distributionof Wages, 1973–1992: A Semiparametric Approach’, Econometrica, Vol.64, No.5,pp.1001–44.

Doeringer, P.B. and M.J. Piore, 1971, ‘Internal Labor Markets and Manpower Analysis’,Lexington: D.C. Health & Co.


382jds04.qxd 30/11/2001 12:28 Page 127

Fields, G. and H. Wan, 1989, ‘Wage-setting Institutions and Economic Growth’, WorldDevelopment, Vol.17, No.9, pp.1471–83.

Gatica, J., Mizala, A. and P. Romaguera, 1995, ‘Interindustry Wage Differentials in Brazil’,Economic Development and Cultural Change, Vol.43, No.2, pp.315–31.

Gibbons, R. and L. Katz, 1992, ‘Does Unmeasured Ability Explain Inter-Industry WageDifferentials?’ Review of Economic Studies, Vol.59, No.3, pp.515–35.

Gittleman, M. and E.N. Wolff, 1993, ‘International Comparisons of Inter-Industry WageDifferentials’, Review of Income and Wealth, Vol.39, No.3, pp.295–312.

Groshen, E.L., 1991, ‘Five Reasons Why Wages Vary Among Employers’, Industrial Relations,Vol.30, No.3, pp.351–81.

Haisken-DeNew, J.P. and C.M. Schmidt, 1997, ‘Inter-Industry and Inter-Region Differentials:Mechanics and Interpretation’. Review of Economics and Statistics, Vol.79, No.3, pp.516–521.

Hamermesh, D.S., 1980, ‘Commentary’, in J.J. Siegfried (ed.), The Economics of Firm Size,Market Structure, and Social Performance, Washington, DC: Federal Trade Committee.

Harris, J.R. and M.P. Todaro, 1970, ‘Migration Unemployment and Development: A Two SectorAnalysis’, American Economic Review, Vol.60, March, pp.126–42.

Katz, L.F. and L.H. Summers, 1989, ‘Industry Rents: Evidence and Implications’, BrookingsPapers on Economic Activity, Microeconomics: 209–275.

Krueger, A.B. and L.H. Summers, 1987, ‘Reflections on Inter-Industry Wage Structure’, in K.Lang, and J. Leonard, Unemployment and the Structure of Labor Markets, Oxford: BasilBlackwell.

Krueger, A.B. and L.H. Summers, 1988, ‘Efficiency wages and the Inter-Industry WageStructure’, Econometrica, Vol.56, No.2, pp.259–93.

Lam, D. and D. Levison, 1992, ‘Age, Experience, and Schooling: Decomposing EarningsInequality in the United States and Brazil’, Social Inquiry, Vol.62, No.2, pp.221–45.

Langoni, C.G., 1973, Distribuição de Renda e Crescimento Econômico no Brasil, Rio de Janeiro:Editora Espressão e Cultura.

Layard, R., Nickell, S. and R. Jackman, 1991, Unemployment, Macroeconomic Performance andthe Labour Market, Oxford: Oxford University Press.

Lindauer, D. and R. Sabot, 1983, ‘The Public/Private Wage Differential in a Poor UrbanEconomy’, Journal of Development Economics, Vol.12, No.5, 1–2 , pp.137–52.

Loser, C. and M. Guerguil, 1999, ‘Trade and Trade Reform in Latin America and the Caribbeanin the 1990s’, Journal of Applied Economics, Vol.2, No.1, pp.61–96.

Macedo, R., 1985, ‘Diferenciais de Salários entre Empresas Estatais e Privadas no Brasil’,Revista Brasileira de Economia, Vol.39, No.4, pp.437–48.

Macpherson, D.A. and B.T. Hirsch, 1995, ‘Wages and Gender Composition: Why Do Women’sJobs Pay Less?’ Journal of Labor Economics, Vol.13, No.3, pp.426–71.

Marshall, A., 1996, ‘Weakening Employment Protection in Latin America: Incentive toEmployment Creation or to Increasing Instability?’ International Contributions to LabourStudies, Vol.6, No.6, pp.29–48.

Mellow, W., 1982, ‘Employer Size and Wages’, Review of Economic and Statistics, Vol.64, No.3,pp.495–501.

Moll, P.G., 1993, ‘Industry Wage Differentials and Efficiency Wages. A Dissenting View withSouth Africa Evidence’, Journal of Development Economics, Vol.41, No.2, pp.213–46.

Moreira, M.M. and P.G. Correa, 1998, ‘A First Look at the Impacts of Trade Liberalization onBrazilian Manufacturing Industry’, World Development, Vol.26, No.10, pp.1859–74.

Morrison, A.R., 1994, ‘Are Institutions or Economic Rents Responsible for Interindustry WageDifferentials?’ World Development, Vol.22, No.3, pp.355–68.

Murphy, K.M. and R.H. Topel, 1990, ‘Efficiency Wages Reconsidered: Theory and Evidence’, inY. Weiss, and G. Fishelson (eds.), Advances in Theory and Measurement of Unemployment,London: Macmillan.


382jds04.qxd 30/11/2001 12:28 Page 128

Oi, W.Y., 1962, ‘Labor as a Quasi-Fixed Factor’, Journal of Political Economy, Vol.70, No.6,pp.538–55.

Pencavel, J., 1980, An Analysis of the Quit Rate in the American Manufacturing Industry,Princeton Industrial Relations Section, Princeton University.

Pinheiro, A.C. and L. Ramos, 1994, ‘Inter-industry Wage Differentials and Earning Inequality’,Estudios de Economia, Vol.21, No.1, pp.79–111.

Polacheck, S.W. and W.S. Siebert, 1993,‘The Economics of Earnings, Cambridge: CambridgeUniversity Press.

Psacharopoulos, G., 1979, ‘On the Weak versus the Strong Version of the Screening Hypothesis’,Economics Letters, Vol.4, pp.181–5.

Psacharopoulos, G. and E. Velez, 1992, ‘Schooling, Ability, and Earnings in Colombia, 1988’,Economic Development and Cultural Change, Vol.40, No.3, pp.629–44.

Rama, M., 1995, ‘Do Labor Market Policies and Institutions Matter? The Adjustment Experiencein Latin America and the Caribbean’, Labour, Special Issue, pp.S243–69.

Rama, M., 1997, ‘Determinants of Hourly Earnings in Ecuador: The Role of Labor MarketLegislation’, Journal of Labor Economics, Vol.15, No.3, pp.S137–65.

Romaguera, P., 1991, ‘Wage Differentials and Efficiency Wage Models: Evidence From theChilean Economy’, Working Paper 153, Santiago: Corporacion de InvestigacionesEconómicas para América Latina.

Rosen, S., 1989, ‘Human Capital’, in J. Eatwell, M. Milgate and P. Newman (eds.), The NewPalgrave – Social Economics, London: Macmillan Press.

Salop, S., 1979, ‘A Model of Natural Rate of Unemployment’, American Economic Review,Vol.69, No.1, pp.117–25.

Shapiro, C. and J.E. Stiglitz, 1984, ‘Equilibrium Unemployment as a Worker Discipline Device’,American Economic Review, Vol.74, No.3, pp.433–44.

Stiglitz, J., 1986, ‘Theories of Wage Rigidity’, in K. Koford, J. Butziewicz, and J. Miller (eds.),Keynes: Economic Legacy, New York: Praeger.

Teal, F., 1996, ‘The Size and Sources of Economic Rents in a Developing Country ManufacturingLabour Market’, Economic Journal, Vol.106, No.473, pp.963–76.

Thomas, M. and L. Valleé, 1996, ‘Labour Market Segmentation in Cameroonian Manufacturing’,The Journal of Development Studies, Vol.32, No.6, pp.876–98.

Tokman, V.E., 1983, ‘The Influence of the Urban Informal Sector on Economic Inequality’, in F.Stewart (ed.), Work, Income and Inequality. Payments System in the Third World, London:Macmillan Press.

Weiss, A. 1980, ‘Job Queues and Layoffs in Labor Markets with Flexible Wages’, Journal ofPolitical Economy, Vol.88, No.3, pp.526–38.

Yamada, G., 1996, ‘Urban Informal Employment and Self-Employment in DevelopingCountries: Theory and Evidence’, Economic Development and Cultural Change, Vol.44,No.2, pp.289–314.

DATA APPENDIX

The Pesquisa Nacional por Amostras de Domicílio is conducted by the official statistical bodyInstituto Brasileiro de Geografia e Estatística. It collects a huge range of variables once a year(normally in the third week of September) across the country. The data are not collected on alongitudinal basis. About 100,000 households are interviewed each year, and about 350,000individuals have their information recorded. There are no PNAD for 1991 and 1994 due topopulation census and shortage of funds, respectively.

The Relatório Anual de Informações Sociais (RAIS) is an annual administrative survey withlots of workers’ information conducted and published by the Ministry of Labour. Data are


382jds04.qxd 30/11/2001 12:28 Page 129

collected the following way: every employer must send the RAIS forms containing acomprehensive set of information on each of its workers such as monthly wages all over the pastyear, nature of labour contract, schooling, age, sex, place of birth, nationality, job tenure, jobclassification, among others to the Ministry of Labour once a year. A special aspect of RAIS isits very large employment coverage, which is about 22 million workers a year. RAIS can beroughly considered as a labour and establishment census, although the Ministry of Labourestimate that around 35 per cent of the formal employment is not recorded due to lack ofcompletion of the questionnaire. Self-employed, employers, informal employment and illegalactivities are not recorded by the RAIS census. RAIS is developed with the aim of providinginformation to identify workers entitled to receive social benefits and to monitor the labourmarket. Despite the fact that RAIS is a longitudinal data base, published data are aggregated onone-, two- and three-digit industry levels.


382jds04.qxd 30/11/2001 12:28 Page 130

wage differentials in brazil: theory and...

Documents