Associations of changes in age-education structure with earnings of female and male workers in Brazil

(1)

Associations of changes in age-education structure with earnings of female and male workers in Brazil

Ernesto Friedrich de Lima Amaral1

Eduardo Luiz Gonçalves Rios-Neto2

Samantha Haussmann Rodarte Faustino3

Guilherme Quaresma Gonçalves4

Abstract

This study aims to estimate associations of demographic and educational changes with earning of female and male workers in Brazil. Previous studies considered earnings of male workers, but they did not include earnings of female workers. The 1970, 1980, 1991, 2000, and 2010 Brazilian Demographic Censuses can be utilized for this analysis. Preliminary results using the 1991 and 2000 Censuses suggest that the share of male population had stronger negative effects on earnings in the lowest (0–4) and middle (5–8) education groups. For women, these negative associations are more pronounced in the middle (5–8) and highest (9+) education groups. Changes in age and educational compositions have significant associations with earnings profiles of female and male workers. Future improvements of this model should consider the remaining correlation that is present between female and male shares to better estimate trends on earnings. We intend to continually update this study, including the analysis of the 2010 Census.

Keywords

Demographic transition. Educational transition. Labor market. Earnings. Female labor force participation. Brazil.

1_{Assistant Professor of the Department of Sociology of Texas A&M University.}

2_{Professor of the Department of Demography of Federal University of Minas Gerais – Cedeplar/UFMG.}

3_{Ph.D. Student of the Department of Demography of Federal University of Minas Gerais – Cedeplar/UFMG. Research Fellow} of the Foundation for Research Support of the State of Minas Gerais (Fundação de Amparo à Pesquisa do Estado de Minas Gerais – FAPEMIG).

4 _{Ph.D. Candidate of the Department of Demography of Federal University of Minas Gerais}_{– Cedeplar/UFMG. Research} Fellow of the Coordination for the Improvement of Higher Education Personnel (Coordenação de Aperfeiçoamento de Pessoal

(2)

1. Introduction and background

Associations of demographic and educational transitions with earnings of Brazilian workers is the main concern of this study. Previous research has considered

earnings of male workers (Amaral et al. 2013a, 2013b, 2015). The present research includes earnings of female workers in the estimations. However, when including female workers in models that estimate the impact on male earnings, one must consider the two-way relationship between fertility and female labor force

participation. Women who are not in the labor force do not compete with women who are in the labor force. Fertility explains labor force participation. At the same time, the decision to participate in the labor market influences fertility levels. Thus, the aim of the present study is to analyze how the age-education structure of female and male workers affected female and male earnings in Brazil between 1970 and 2010.

The increasing integration of women into the job market is a phenomenon that has been occurring in several countries worldwide, with a spike in participation occurring during the period after World War II (Souza, 2009). This trend has shaped the

international job market, but it has also affected the traditional family structure. The relationship between economic structure and an increase in female labor force

participation is a key factor for understanding the gender changes that have occurred in several countries. In general, when women enter the workforce and receive paid employment, an overall restructuring can occur, which often includes an increase in employment services, more flexible work hours, and new patterns of industrial relations (Daly and Rake, 2003).

In addition to women entering the job market at an escalated rate, the post-World War II period was marked by a series of transformations that affected family structure and gender relationships. In several countries, there have been trends towards reducing fertility rates, an escalated divorce, and increased participation of married women with young children in the job market (Souza, 2009). In general, the male breadwinner model (male head-of-household model) has increasingly been abandoned. Women have been giving birth to fewer children, and the age at which women have their first child has increased.

(3)

The increase in female participation in the job market is a phenomenon that

contributes to the process of demographic transition (Soares and Falcão, 2008). The transition is initially characterized by lower mortality and fertility rates. The changing role of women in society can be partially attributed to the impact of reduced mortality on family decisions. In this context, the increase in life expectancy and the decrease in family size have a positive effect on female participation in the workforce, and these factors contribute to reducing the income gap between men and women.

In Brazil, the increase in female participation in the labor market has been occurring since the 1970s (Costa, 1990; Rios-Neto and Wajnman, 1994; Souza, 2009;

Wajnman, Queiroz, and Liberato, 1998). According to the Brazilian Institute of

Geography and Statistics (Instituto Brasileiro de Geografia e Estatística – IBGE), the proportion of economically active women ranged from 47.17 percent in 1997 to 50.27 percent in 2002, and 52.35 percent in 2007 (Souza, 2009). Women with varying characteristics have had higher workforce participation rates since the 1980s. In other words, the increase in female participation in the Brazilian job market occurred regardless of age, race, marital status, socioeconomic status, or region of residence (Rios-Neto and Batista, 1998; Souza, 2009). However, despite their increasing participation in the labor market, women continue to earn less than men, even when their education levels are equal (Alves and Corrêa, 2009). Despite the increase in the participation of female labor being extremely important for the growth of Brazil, a tradition of economic instability and gender inequalities still remain. While differences in income have not been eliminated, they have been significantly reduced, especially in the formal labor market.

2. Data and methods

The 1970, 1980, 1991, 2000, and 2010 Brazilian Demographic Censuses were collected by the Brazilian Institute of Geography and Statistics (Instituto Brasileiro de

Geografia e Estatística – IBGE) and were used for the present study. We aggregated

the microdata in cells by micro-region, sex, age, education, and year. Taking into account the sample design (frequency weights), we estimated hourly earnings and the distribution of men and women by age-education groups in each year and micro-region.

(4)

Much of the increase in female participation in the labor market was related to part-time work. In more recent decades, this scenario has changed, and women are increasingly involved in full-time work. Increasingly more women are committed to paid employment. This phenomenon is particularly true for more educated women. In order to control for part-time work, our dependent variable indicates the logarithm of mean real hourly earnings from main occupation of women and men. This

information is not available prior to the 1991 Census. We estimated some

preliminary results using the 1991 and 2000 Censuses, which are illustrated in the following section.

Because our estimates utilized 502 micro-regions, 24 sex-age-education groups, and 2 Censuses, the maximum number of possible observations in the regressions was 24,096. However, our models included only cells that had at least 25 individuals, in order to avoid issues of heteroscedasticity. Thus, we reduced the estimates to 20,865 micro-region/sex/age/education/year cell observations. We introduced area fixed-effects in order to take into account regional disparities around the country. We plan on utilizing state of residence in future estimations, in order to avoid small cell sizes generated by micro-regions.

We expect to see a positive impact when considering age within each education group. This positive impact is also anticipated with increased education within each age group. In order to control for these effects, we included sex-age-education indicators.

The models also included effects of changing the sex-age-education composition on female and male earnings. Thus, a set of variables was added to the model to estimate the impact of own-cohort size on earnings. Our main hypothesis is that cohort size has a negative impact on earnings. The following equation illustrates these estimations:

log(Yijrt) = si + xj + γr + πt + Xijrt + (si * xj) + (si * πt) + (xj * πt) + (γr * πt) + (Xijrt * πt)+ (si

* xj * πt) + εijrt,

where log(Yijrt) is the logarithm of mean real hourly earnings from main occupation of

(5)

years of schooling) and age group j (j=15–24, 25–34, 35–49, 50–64 years of age). Earnings are observed by micro-region r (r=1,…., 502) at time t (t=1991, 2000). Finally, si is a vector of fixed effects that indicates the group’s educational

attainment, xj is a vector of fixed effects that indicates the age group, γr is a vector of

fixed effects that indicates the micro-region, and πt is a vector of fixed effects that

indicates the time period. The linear fixed effects control for differences in earnings across sex, education groups, age groups, and micro-regions over time. We added the distribution of females and males by sex-age-education group from each micro-region and year (Xijrt). The interaction (si * xj) accounts for the age profiles of

earnings across educational groups. The interactions (si * πt), (xj * πt), and (γr * πt)

control for the changing impact of education, age, and micro-region over time. The term (Xijrt* πt) is the interaction between the distribution of the female and male

populations and time. The interaction (si * xj * πt) accounts for variation in the age

profile of earnings by education group and time.

3. Preliminary results

Income increases with age and education for men and women (Table 1). Furthermore, the earnings of women are lower than those of men in all age-education groups between 1970 and 2000. By taking into account both men and women, it is possible to try to understand the trends of earnings. The distribution of the female population by age and education is highly correlated with the distribution of the male population (Table 2). We use distributions of men and women by age-education groups as independent variables, as a strategy to verify the associations of changes in the age-education structure on female and male earnings. Regression results for 1991 and 2000 (Table 3) indicate that within each age group, an increase in education has a positive effect on female and male earnings. Similarly, within each education group, an increase in age also has a positive effect on earnings. These coefficients confirm the income trends exposed in Table 1.

>>> Table 1 <<< >>> Table 2 <<< >>> Table 3 <<<

(6)

Distribution of males by age-education groups had negative effects on male earnings (Table 3). More specifically, males in 0–4 and 5–8 education groups and 25–34, 35– 49, and 50–64 age groups experienced these negative associations. The strongest negative magnitudes are observed for coefficients in the middle education group. Male proportions for those aged between 15 and 24 years also had negative associations with earnings for 0–4 and 9+ education groups.

Proportions of women by age-education groups also indicated negative associations with female earnings (Table 3). Women in 15–24, 25–34, and 35–49 age groups within 5–8 and 9+ education groups are examples of negative associations with earnings. For older women (50–64 age group), these associations are observed in the 5–8 education group.

In summary, preliminary results suggest that the share of male population had

stronger negative effects on earnings in the lowest (0–4) and middle (5–8) education groups. For women, these negative associations are more pronounced in the middle (5–8) and highest (9+) education groups. Considering changes in age and

educational compositions makes a sizeable difference on the estimation of earnings profiles among female and male workers.

4. Final considerations

The estimation of models with the 24 gender-age-education groups was an

improvement on previous studies. However, there are many non-employed women who do not compete with employed women in the job market. Our present model has an endogenous estimate because the proportion of women in the sex-age-education groups does not consider women who are not competing in the job market. Future improvements of this model should consider the remaining correlation that is present between female and male shares to better estimate trends on earnings. The number of children ever born, the level of education, and socioeconomic status influence women’s decisions to participate in the labor market. Thus, it is necessary to use instrumental variables to correct the information on females who are economically active within the population before these independent variables are included in the models. The new distribution of employed women, estimated with instrumental variables, would not be influenced by individual characteristics of women.

(7)

Some strategies could be used to estimate existing information regarding the female population in the labor force. In order to take into account women without children (zero to one parity), the sex ratio between women aged 15–24 years and men aged 20–29 years could be used as an instrumental variable in the models. Information about stillbirths could also be used as a tool, as stillbirths have an impact on the number of children that a woman has (in this case, from zero to one child). At the same time, stillbirths do not directly affect women’s participation in the labor force, and they are not linked to other female characteristics (Souza 2009). In relation to women increasing parity from one to two children, information on twins could be used to infer the participation of women in the labor market. This methodology involves measuring the exogenous fertility of women who have two children when they were planning on having only one child (Souza 2009; Verona 2004). A strategy to control for a woman’s decision to have a third child is to utilize information on whether she has two children of the same sex and wants a child of the opposite sex (Souza 2009). This information can be used to remove the endogeneity that exists between socioeconomic variables and women who have multiple births (at least three children). After correcting for the population of working-age women, the regression models that assess the impact of the distribution of men and employed women in the age-education groups would be re-estimated. Thus, the new models would not have the same endogeneity problems. We intend to continually update this study, including the analysis of the 2010 Census.

5. References

Alves JED, Corrêa S. 2009. “Igualdade e desigualdade de gênero no Brasil: um panorama preliminar, 15 anos depois do Cairo.” In Brasil, 15 anos após a

Conferência do Cairo. ABEP; UNFPA. Campinas.

Amaral EFL, Almeida ME, Rios-Neto ELG, Potter JE. 2013a. “Effects of the age-education structure of female workers on male earnings in Brazil.” Poverty &

Public Policy, 5(4): 336-353.

Amaral EFL, Potter JE, Hamermesh DS, Rios-Neto ELG. 2013b. “Age, education, and earnings in the course of Brazilian development: Does composition matter?” Demographic Research, 28(20): 581-612.

(8)

Amaral EFL, Queiroz BL, Calazans JA. 2015. “Demographic changes, educational improvements, and earnings in Brazil and Mexico.” IZA Journal of Labor &

Development, 4(23): 1-21.

Costa L. “Aumento da participação feminina: uma tentativa de explicação.” 1990. In

Encontro da Associação Brasileira de Estudos Populacionais, 7., Caxambú. Anais... Caxambú: ABEP. v. 2, p.231-243.

Daly M, Rake K. 2003. Gender and the welfare state: care, work and welfare in

Europe and USA. Cambridge: Polity Press.

Rios-Neto ELG, Wajnman S. 1994. “Participação feminina na população economicamente ativa no Brasil: alternativas para projeções de níveis e padrões.” Pesquisa e Planejamento Econômico, Rio de Janeiro, v. 24, n. 2, p. 203-234, Maio.

Rios-Neto ELG, Batista DBDA. 1998. “Segregação ocupacional entre solteiras e casadas: o possível impacto da licença maternidade.” In Encontro da

Associação Brasileira de Estudos Populacionais, 11. Belo Horizonte. Anais...

Belo Horizonte: ABEP. p. 2663-2686.

Soares R, Falcão B. 2008. “The Demographic Transition and the Sexual Division of Labor.” Journal of Political Economy, 116(6).

Souza L. 2009. “O efeito dos filhos sobre a participação feminina no mercado de trabalho brasileiro: explorando diversas fontes de variação exógena na

fecundidade.” PhD Dissertation. Belo Horizonte: Centro de Desenvolvimento e Planejamento Regional (CEDEPLAR), Universidade Federal de Minas Gerais (UFMG).

Verona APA. 2004. “A relação entre fecundidade e educação dos filhos: um experimento natural utilizando dados de gêmeos.” PhD Dissertation. Belo Horizonte: Centro de Desenvolvimento e Planejamento Regional (CEDEPLAR), Universidade Federal de Minas Gerais (UFMG).

Wajnman S, Queiroz BL, Liberato C. 1998. “O crescimento da atividade feminina nos anos noventa no Brasil.” 1998. In Encontro da Associação Brasileira de

(9)

Table 1. Male and female mean real monthly earnings from main occupation+

by age-education group, Brazil, 1970–2000 Age-education

group

1970 1980 1991 2000

Males Females Males Females Males Females Males Females 15–24 years; 0–4 years of schooling 158.54 100.57 276.29 146.56 196.05 123.47 213.23 146.87 15–24 years; 5–8 years of schooling 285.87 171.14 359.01 188.65 261.94 153.93 250.15 163.19 15–24 years; 9+ years of schooling 530.31 312.25 641.81 361.47 428.00 265.46 361.56 260.98 25–34 years; 0–4 years of schooling 227.16 151.49 439.51 207.90 289.52 162.82 303.49 180.29 25–34 years; 5–8 years of schooling 585.02 268.05 818.99 316.88 472.21 227.73 459.94 256.07 25–34 years; 9+ years of schooling 1,183.87 463.03 1,562.22 638.97 894.31 482.70 834.13 511.05 35–49 years; 0–4 years of schooling 273.56 172.48 551.83 242.43 381.59 188.68 394.58 208.30 35–49 years; 5–8 years of schooling 845.03 384.19 1,316.54 474.19 755.74 335.18 668.49 317.34 35–49 years; 9+ years of schooling 1,661.43 574.40 2,348.69 792.60 1,557.74 684.45 1,482.51 761.92 50–64 years; 0–4 years of schooling 275.46 179.59 553.77 270.25 380.52 202.94 436.81 234.93 50–64 years; 5–8 years of schooling 978.13 505.29 1,587.19 675.32 918.25 413.44 913.82 409.95 50–64 years; 9+ years of schooling 1,724.94 664.49 2,823.26 940.49 1,826.73 751.04 2,080.80 912.38 Grand mean 711.47 314.40 1,093.98 430.86 694.56 331.22 699.24 363.49

(10)

Table 2. Percentage distribution of the male and female working-age population by age-education group, Brazil, 1970–2000

Age-education group

1970 1980 1991 2000

Males Females Males Females Males Females Males Females 15–24 years; 0–4 years of schooling 28.19 29.14 20.59 19.51 14.61 11.98 9.04 6.51 15–24 years; 5–8 years of schooling 5.38 5.19 10.53 10.49 12.09 12.13 12.46 11.44 15–24 years; 9+ years of schooling 2.74 3.07 5.87 6.79 5.97 7.55 10.24 12.47 25–34 years; 0–4 years of schooling 19.71 20.29 16.39 16.60 12.41 11.95 8.82 7.56 25–34 years; 5–8 years of schooling 1.98 1.67 3.90 3.57 6.82 6.45 7.63 7.21 25–34 years; 9+ years of schooling 2.00 1.79 4.77 4.89 7.40 8.35 8.12 9.72 35–49 years; 0–4 years of schooling 22.66 22.81 19.02 19.82 17.11 17.57 13.32 13.17 35–49 years; 5–8 years of schooling 1.62 1.29 2.39 2.04 3.67 3.47 6.73 6.58 35–49 years; 9+ years of schooling 1.59 1.13 2.84 2.33 5.54 5.52 8.46 9.33 50–64 years; 0–4 years of schooling 12.84 12.70 11.72 12.40 11.49 12.52 10.36 11.31 50–64 years; 5–8 years of schooling 0.65 0.48 0.94 0.80 1.16 1.11 1.99 2.00 50–64 years; 9+ years of schooling 0.62 0.43 1.05 0.76 1.72 1.41 2.84 2.72 Grand total 25,760,60 0 26,037,53 7 32,613,94 7 33,695,90 4 43,434,54 6 45,265,54 8 53,177,95 3 55,440,18 3

(11)

Table 3. Fixed-effects estimates on the logarithm of mean real hourly earnings from main occupation+_{of women and men (dependent variable), Brazil, 1991}

and 2000 Variables Coefficients (standard errors) Constant –0.1222*** (0.0155) Year 1991 ref. 2000 0.0873*** (0.0037) Male age-education indicators

15–24 years; 0–4 years of schooling ref.

15–24 years; 5–8 years of schooling –0.0312

(0.0399)

15–24 years; 9+ years of schooling 0.7949***

(0.0243)

25–34 years; 0–4 years of schooling 0.2466***

(0.0272)

(0.0296)

(0.0252)

(0.0412)

(0.0233)

(0.0217)

(0.0419)

(0.0372)

(0.0255) (continue)

(12)

Variables Coefficients (standard errors) Female age-education indicators

(0.0200)

(0.0328)

(0.0223)

(0.0265)

(0.0255)

(0.0230)

(0.0389)

(0.0217)

(0.0210)

(0.0408)

(0.0261)

(0.0235) Distribution of males by age-education

groups

15–24 years; 0–4 years of schooling –2.5553***

(0.2045)

15–24 years; 5–8 years of schooling 1.3495*

(0.6130)

15–24 years; 9+ years of schooling –3.5306***

(0.3997)

(0.3795)

(0.7754)

25–34 years; 9+ years of schooling 1.5404**

(0.5284)

(0.4342)

(0.7139)

(0.4963)

50–64 years; 0–4 years of schooling –1.5477**

(0.5694)

50–64 years; 5–8 years of schooling –13.2141**

(3.8535)

(1.9653) (continue)

(13)

Variables Coefficients (standard errors) Distribution of females by age-education

groups

15–24 years; 0–4 years of schooling 0.1228

(0.1599)

(0.4828)

(0.4387)

(0.3340)

(0.5993)

(0.5648)

35–49 years; 0–4 years of schooling 0.1281

(0.4030)

(0.6124)

(0.5269)

(0.5860)

(2.7318)

50–64 years; 9+ years of schooling 0.1225

(1.7735)

Number of observations 20,865

Number of groups 502

Fraction of variance due to area-time fixed

effects 0.6477

F (48; 20,315): all coefficients=0 4.340.25***

F (501; 20,315): area fixed effects=0 47.68***

*_{Significant at p<.05;}**_{Significant at p<.01;}***_{Significant at p<.001} Sources: 1991 and 2000 Brazilian Demographic Censuses.