THE MALE MARRIAGE WAGE PREMIUM IN BRAZIL Janana Teodoro Guiginski 1 - - PDF document

THE “MALE MARRIAGE WAGE PREMIUM” IN BRAZIL Janaína Teodoro Guiginski1 Simone Wajnman2 INTRODUCTION The aim of this paper is to explore earnings differentials of Brazilian men in different marital status who share similar personal and employment characteristics. Literature about the subject unanimously points out the existence of what is known as „male marriage wage premium‟, which means that, when men with similar individual and professional characteristics who differ only in marital status are compared, married men have higher earnings than their single counterparts (Adler and Oner, 2013, Ahituv and Lerman, 2007, Hersch and Stratton, 2000, Killewald and Gough, 2013). Several theories and hypotheses, based on different arguments and methods of analysis have been tested, but there is no consensus on the results, suggesting that the determinants behind male marriage wage premium may be multiple and interrelated. Three hypotheses are commonly put forward to explain why married men have higher earnings than single men: productivity, selectivity, and discrimination. According to hypotheses associated with productivity, specialization provided by marriage allows men to invest more in human capital, which in turn increases productivity and, consequently, wages (Becker, 1991). Hersch and Stratton (2000) included time spent on housework in the wage equation to test whether household specialization may explain the marriage premium for men. Results show that housework has a negative impact on male wages, but does not affect the magnitude of the marriage premium. The authors concluded that marriage seems to make men more productive, but not due to specialization, and suggest alternative explanations such as preferential treatment from employers or changes in behavior derived from marriage or from the decision to get married.

1 Doctoral candidate in the Graduate Program in Demography at CEDEPLAR / UFMG and scholarship recipient of

CNPq-Brasil.

2 Full Professor in the Demography Department at CEDEPLAR / UFMG.

According to selectivity theories, highly productive men, with higher earning potential, are more likely to be married. According to this hypothesis, observable characteristics (such as background, educational attainment, physical appearance, responsibility, etc.) increase the chances of success in the labor market and also in the marriage market (Chiodo and Owyang, 2002). Ahituv and Lerman (2007) show that marital status, work effort, and wage rates are

• interrelated. Work hours and earnings are influenced by marital status, but the opposite is also
• bserved: success in the labor market increases the likelihood of marriage and remarriage and

reduces the likelihood of divorce (Ahituv and Lerman, 28, 2007). The discrimination hypothesis establishes that employers favor married men because they perceive marriage as a sign of greater stability, responsibility, or maturity. Adler and Oner (2013) found higher premium for men who work in more routinized occupations rather than in more creative ones and claim that this result is explained by the discrimination hypothesis. From a stereotyped point of view, service sector and working-class employees would benefit from employers‟ biased perceptions. On the other hand, men in creative occupations would be less adhered to traditional value systems which reward marriage (Adler and Oner, pp. 17-18, 2013). Other hypotheses to explain male marriage wage premium are based on changes in behavior after marriage and also on the role of the wife. Killewald and Gough (2013) suggest that part of the marriage wage premium is a result of changes in work hours, changes in job traits, and

• tenure. They claim that marriage possibly encourages tenure in employment, more effort to paid

labor, changes for better positions, and changes in preferences for financial resources. Moreover, that marriage can provide benefits which affect wages, such as better health conditions and access to spouse‟s human capital. Ashwin and Isupova (2014) emphasize the importance of the breadwinning role for masculine identity and the direct pressure exerted by wives for men to increase their income. In addition, they emphasize that wives influence husbands' potential earnings through monitoring and mentoring, stimulating a “responsible version” of masculinity. They highlight that the wife acts as a co-producer of masculinity within marriage, with direct effects on perspectives and performance for work, positively reflecting on wages. Although recurrent in international literature, this subject remained poorly explored for the case

• f Brazil. Therefore, for this work, data from the 2010 Brazilian Demographic Census, made

available by the IPUMS-I project will be used to examine male marriage premium. Since cohabitation in Brazil is a very common form of union, with same legal status and rights as civil union, we examine whether there are differences between marriage and cohabitation premiums. In addition, based on the hypotheses discussed in the following section, we analyze whether wife/partner traits influence earnings of their respective spouses. Firstly, through quantile regressions, we verify the existence of differences in marriage and cohabitation premiums

across earnings distribution. Secondly, we examine the effect of wife or partner‟s educational attainment and her work hours on male earnings. This work analyzes workers at two different status in employment - formal employee in the private sector and self-employed - in order to verify the employer discrimination hypothesis. If the highest marriage premium is observed for employees rather than for self-employed men, it is possible that employers' preferential treatment of married men plays a significant role. Employer discrimination may also be a plausible hypothesis if the penalty associated with a working wife were greater for employees than for the self-employed. In addition to employer discrimination, other factors can also contribute to explain the negative association between wife/partner‟s work and male earnings, such as household specialization, assortative mating, and income effect. WIVES’ EDUCATIONAL ATTAINMENT AND LABOR SUPPLY In addition to marriage premium, researchers also show possible premiums and penalties associated with productive characteristics of the wives of married men. Spouses‟ educational and work traits may influence (or be influenced by) male earnings in a number of ways, such as through specialization, discrimination, income effect, joint human capital, or may be an effect of the process of marriage formation (Birch and Miller, 2006, Blackaby et al 2007, Jacobsen and Rayack 1996, Mamun, 2012, Song, 2007, Verbakel and de Graaf, 2009, Zavodny, 2008). The positive assortative mating, which is generally observed, is a tendency for individuals with similar educational levels and productive potential to marry each other. If men with higher wage are more likely to marry highly educated women, we can observe a wage premium for the married man associated with his spouse‟s educational attainment. Thus, this premium would be explained by the process of marriage formation (Birch and Miller, 2006). The joint human capital hypothesis suggests that wife/partner‟s high educational attainment positively contributes to a man's stock of human capital, which increases his productivity and earnings (Birch and Miller, 2006; Mamun, 2012). Better educated women can contribute to their spouses‟ work performance, through assistance with job tasks, social and professional relations, influencing decisions and investments in husbands‟ human capital, for example (Mamun, p.55, 2012). Men whose wives work usually earn lower wages than those men whose wives do not work (Jacobsen and Rayack, 1996). According to the employer discrimination hypothesis, men married to non-working women receive higher remuneration because they have greater geographical mobility or because employers favor these men for several reasons. Employer's

favoritism toward men whose spouses do not work may occur via promotions, salary increases

• r even by providing more training opportunities within companies (Jacobsen and Rayack 1996;

Blackaby et al., 2007). The productivity and specialization hypotheses predict that men with working wives have lower productivity than men in more traditional marriages where only the man holds a work position. Reasons for this lower work productivity may be related to lesser effort due to greater time spent on household chores, lesser degree of support for the husband's career offered by working wives, and lesser mobility of couples in which both spouses work (Jacobsen and Rayack, 1996). For Chun and Lee (2001), the highest wage premium for married men whose wives do not work is explained by the positive effects of specialization within marriage. They found that greater degree of specialization within the household results in higher marriage gain. For hypotheses based on selectivity, the wage differential between single earner men and dual earners couples is correlated with, but not caused by, the wife's work hours in the labor market. Marital selectivity or process of marriage formation predicts both premiums and penalties associated with the work of the wife. A male wage premium can be observed for non-working wives when "traditional" marriage selection occurs on the basis of valued characteristics in the labor market (Jacobsen and Rayack, 1996) or when there is negative assortative mating on wage rates (Blackaby et al., 2007). In the latter, men with higher labor-market productivity would be more likely to marry women who work fewer hours - women with high reservation wages, with a higher valuation of leisure, women with high domestic productivity or low potential market wage, for example (Jacobsen and Rayack, 269, 1996). On the other hand, men with high productivity may have better opportunities of marrying equally productive women, leading to a positive correlation between wives‟ work hours and husbands‟

• earnings. This positive correlation may also be observed due to the economic security that a

working wife or a wife with high earning potential may offer her family, allowing the husband to switch jobs and seek better occupations or invest in human capital. Blackaby et al. (2007) show that earnings penalty for British men with working wives were reversed in more recent periods. For the authors, the reversal of penalty and the appearance, in some occupational groups, of a premium for men with working wives are consistent with positive assortative mating by

• education. The strong and positive correlation between spouses‟ educational attainment, the

increase in women‟s labor force participation and greater rewards for women‟s human capital would be sufficient to reverse the negative correlation between wives‟ work and husbands‟ earnings (Blackaby et al., p.120, 2007). The income effect may also be a key factor for the negative association between male earnings and wife's work hours. Most commonly, it is argued that husbands‟ high earnings may allow

wives to work fewer hours or not work at all. Other arguments related with the income effect suggest that if the husband has a low wage, there may be a greater need for the wife's income to support the family. Or, alternatively, a wife with a high earning may make it possible for the husband to work fewer hours or to devote himself to more enjoyable work with a smaller monetary reward (Jacobsen and Rayack, 1996). THE BRAZILIAN CONTEXT There are few papers on the male marriage wage premium in Brazil. Muniz and Rios-Neto (2002) found marriage wage premium for both Brazilian men and women. For men, the wage differential according to marital status was associated with greater potential experience and stability in employment, in addition to favorable discrimination to productive attributes of married men, compared to single men. However, marriage wage premium found for married women would be poorly explained by the earning equation. Souza and França (2013) have found marriage wage premium for men, but marriage wage penalty for women. They observed that in Brazil, the male wage premium is higher in the higher deciles of salary distribution and that female wage penalty is lower in the higher decile. As in Muniz and Rios-Neto (2002), most of the differential is explained by differences in wage return associated with individual attributes, which favors married men. For women, decomposition points to negative discrimination of productive attributes of those who are married. Opice (2010) examines the association between wives‟ work and husbands‟ wages in Brazil. In this case, the focus of the article is not on the marriage premium itself, but on the effects of specialization within marriage on male wages. Results showed that married men whose wives work earn less than married men whose wive do not work for pay. A possible explanation for this result is the income effect. After correcting the endogeneity between the wife's work and the husband's wage, Opice (2010) shows that the coefficient for the woman's work becomes

• positive. This means that men married to women who are more likely to work receive higher

wages than those whose wives present a low likelihood of labor market participation. This article takes into consideration both marriage and cohabitation when analyzing male marriage wage premium. Historically in Brazil, as in the rest of Latin America, both marriage and cohabitation have coexisted. For lower social strata, cohabitation has traditionally been a substitute for marriage, related to economic constraints, inequalities and social exclusion. Generally, traditional cohabitation is associated with high levels of fertility, low educational attainment, and low female independence levels. Currently, there is evidence of a new type of cohabitation, implying a more modern type of union. For the upper classes, scholars suggest

that cohabitation is a result of modernization, economic development, and increased female

• autonomy. Modern cohabitation might also be the result of changes in values and attitudes, and

could be explained by the Second Demographic Transition approach. Cohabitation has tended to increase in the last decades, both among lower classes and better educated groups, and for all age groups. Therefore, in Brazil, the option for cohabitation rather than marriage may be related both to tradition and modernity (Esteve et. al, 2012; Covre-Sussai et. al, 2015). DATA AND METHODOLOGY For the analysis, data from the 2010 Brazilian Demographic Census, gathered by IBGE and made available by the Integrated Public Use Microdata Series project - IPUMS-I3 is used. The sample consists of adult males aged 25 to 49, formally employed in the private sector or self- employed, with declared labor income and hours worked per week. Only urban workers - residents in urban or metropolitan areas and in non-agricultural occupations or activities - were included. For the first set of analyses, 608,118 men in three marital status categories were considered: single (never married), living in consensual union and married. Separated, divorced and widowed men were excluded, due to their high heterogeneity and small proportion, in relation to the other marital status categories. Separate wage/earning equations were estimated for two groups of workers: formal employees in the private sector and self-employed men. The dependent variable is the natural logarithm of monthly earnings. The variable of interest is marital status and the aim is to verify the existence

• f marriage and cohabitation premiums. The independent variables correspond to individual

traits, such as age, educational attainment and race or color; geographical location (urban vs. metropolitan area and state); and, characteristics related to work - work hours, industry, and

• ccupational level.

For the second analysis, only married men or men in consensual union are considered, totaling 361,365 observations. At this stage, single men were not considered, since the analysis is focused on the association between productive characteristics of wives/partners and male

• earnings. The variables of interest are the women‟s educational attainment and work situation

(not working, part-time work or full-time work).

3 Minnesota Population Center. Integrated Public Use Microdata Series, International: Version 6.4. Minneapolis:

University of Minnesota, 2015. Available in: http://doi.org/10.18128/D020.V6.4.

According to Jacobsen and Rayack (1996), women's labor supply is endogenous to the income level of their spouse (income effect). If this endogeneity is ignored, the coefficient associated with the observed work hours of wives will be inconsistent using ordinary least squares. To solve the endogeneity problem, predicted values for women's labor supply are usually employed as an instrumental variable, instead of actual values of worked hours, in the male earning equation. In this article, new equations were estimated by replacing observed work hours by an instrumental variable for the potential work hours of the female partner. In order to allow for the simultaneity of husbands' earning and wives‟ labor supply, thus correcting endogeneity, a two- equation model with instrumental variable technique was used. The strategy and selection of variables followed the approach adopted by Jacobsen and Rayack (1996), Chun and Lee (2001), Blackaby et al. (2005), and Song (2007), among others. First, women's potential work hours were estimated as a function of their characteristics and the productive characteristics of their spouses through the use of Tobit regression model, for employees and self-employed men, separately. Control variables included all variables of the earning equation. The presence

• f children (number of children and presence of children under 7 years of age) and the age and

educational attainment of the women were also considered. Afterwards, predicted values of work hours under the Tobit model were included in the earning equation instead of work hours actually observed. Thus, the income effect is controlled and estimates of male earning equations become consistent. SAMPLE DESCRIPTION TABLE 01 presents the descriptive statistics of the sample, which consists of 608,118 men, almost half of which were married, and 73.1% were employees and 26.9% were self-employed. Among employees, there was a higher proportion of single men (18.7%) and a lower proportion

• f men in consensual union (31.0%) than among the self-employed (13.7% single and 36.7% in

consensual union). Employees are younger, have higher educational attainment, and are more present in metropolitan areas than self-employed men. They also have a higher proportion of full-time workers, in the manufacturing and services industries and in higher occupational levels. When considering marital status, we can observe that single men are consistently younger and better schooled, whereas married men are older, and the majority of men in union have low educational attainment. Men in consensual union have a higher relative proportion of browns and blacks, a higher proportion of workers in the construction industry and a lower proportion in higher occupational levels. Married men have higher average earnings than others, an average

• f R$ 1,785.20 per month for married employees and R$ 1,689.60 for married self-employed

• men. Among employees, the lowest average wage is observed for men in union (R$ 1,198.60)

and, among the self-employed, for the single, with an average earning of R$ 1,162.40 per month. TABLE 1 - Descriptive statistics, according to status in employment and marital status, Brazilian men, 2010.

Employee (n=444.742) Self-employed (n=163.376) ALL Single In Union Married ALL Single In Union Married Frequency 444742 83377 137872 223493 163379 22455 59975 80946 Proportion 100% 18.75% 31.00% 50.25% 100% 13.74% 36.71% 49.54% Monthly Earnings (R$) 1518.7 1356.6 1198.6 1785.2 1440.7 1162.4 1231.5 1689.6

(2003.8) (1769.1) (1331.8) (2374.3) (2095.9) (1656.4) (1787.1) (2390.8)

Age 35.2 30.8 34.7 37.2 37.5 33.5 36.8 39.3

(7.0) (5.7) (6.7) (6.8) (6.9) (6.8) (6.8) (6.5)

Education - No schooling 0.19 0.11 0.25 0.18 0.32 0.25 0.38 0.29 Education – Elementary 0.29 0.22 0.35 0.28 0.36 0.32 0.38 0.36 Education – High school 0.39 0.46 0.34 0.39 0.28 0.35 0.22 0.30 Education - College 0.13 0.20 0.06 0.14 0.05 0.09 0.02 0.06 Color – Brown 0.39 0.37 0.43 0.36 0.42 0.41 0.46 0.38 Color – White 0.52 0.55 0.45 0.55 0.50 0.51 0.44 0.55 Color – Black 0.10 0.08 0.12 0.08 0.08 0.08 0.10 0.07 Residence in Metropolitan Area 0.62 0.64 0.64 0.60 0.53 0.56 0.56 0.50 Residence in Urban Area 0.38 0.36 0.36 0.40 0.47 0.44 0.44 0.50 Part-time work (<40 hours/week) 0.08 0.09 0.07 0.07 0.14 0.20 0.14 0.12 Full time work (40 to 48 hours/week) 0.76 0.79 0.73 0.76 0.58 0.61 0.57 0.58 Excessive Work Hours (>48 hours/ week) 0.17 0.11 0.20 0.17 0.28 0.19 0.29 0.30 Industry - Manufacturing 0.29 0.26 0.27 0.31 0.09 0.08 0.08 0.09 Industry – Construction 0.11 0.07 0.15 0.09 0.31 0.30 0.36 0.27 Industry – Trade 0.20 0.21 0.20 0.20 0.31 0.28 0.28 0.33 Industry – Services 0.41 0.46 0.38 0.40 0.29 0.33 0.27 0.30 Occupational Level I [baseline] -

• Legislators. senior officials and

managers; Professionals 0.13 0.17 0.07 0.15 0.07 0.07 0.05 0.09 Occupational Level II - Technicians and associate professionals; Clerks 0.18 0.23 0.14 0.18 0.08 0.11 0.06 0.08 Occupational Level III - Service workers and shop and market sales 0.19 0.19 0.19 0.18 0.20 0.22 0.18 0.21 Occupational Level IV - Crafts and related trades workers 0.20 0.14 0.23 0.19 0.43 0.36 0.47 0.41 Occupational Level V - Plant and machine operators and assemblers; Elementary occupations 0.31 0.27 0.36 0.30 0.22 0.25 0.24 0.21 Source of basic data: Brazilian Demographic Census 2010 - IBGE / IPUMS-I. Note: for continuous variables, standard deviation between parentheses.

MARRIAGE AND COHABITATION PREMIUMS Results show marriage and cohabitation premiums for both employees and self-employed men (TABLE 02). The control variables present the expected signs4. Both premiums are higher for the self-employed than for employees, which refutes the initial hypothesis that employer discrimination would be the main explanation for the phenomenon. For employees, being married is associated with a 25.6% increase in wages, and for self-employed workers, marriage is associated with a 38.1% increase in earnings, controlling for the other model variables. The cohabitation premium appears much smaller, but still positive and significant, representing a 13.8% increase in the wage of employees and a 19.9% increase in the earning of self- employed, compared to single men. TABLE 2 – OLS coefficients for log of wage/earning, employee and self-employed, Brazilian men, 2010. Log of Wage/Earning Employee Self-employed Marital Status [Consensual Union] 0.129 *** 0.182 *** Marital Status [Married] 0.228 *** 0.323 *** Age 0.034 *** 0.044 *** Age -squared 0.000 *** 0.000 *** Education [Elementary] 0.125 *** 0.207 *** Education [High school] 0.305 *** 0.430 *** Education [College] 0.867 *** 0.838 *** Color [White] 0.104 *** 0.140 *** Color [Black]

• 0.022 ***
• 0.027 ***

Area [Metropolitan] 0.080 *** 0.124 *** State (26 dummies) (omitted) (omitted) Work hours [Part-time]

• 0.032 ***
• 0.224 ***

Work hours [Excessive] 0.098 *** 0.146 *** Industry [Construction]

• 0.075 ***
• 0.070 ***

Industry [Trade]

• 0.126 ***

0.060 *** Industry [Services]

• 0.072 ***

0.087 *** Occupation [Level II]

• 0.370 ***
• 0.214 ***

Occupation [Level III]

• 0.560 ***
• 0.400 ***

Occupation [Level IV]

• 0.489 ***
• 0.417 ***

Occupation [Level V]

• 0.588 ***
• 0.490 ***

Constant 6.264 *** 5.884 ***

• N. Obs.

444742 163376 R-squared 0.4545 0.3022

Source of basic data: Brazilian Demographic Census 2010 - IBGE / IPUMS-I. Levels of significance: *** ≤0.001; ** ≤0.010; * ≤0.050; "." ≤0.100; "ns" not significant at the 10% level.

4 Consolidated theories and numerous studies have already shown that wage or earnings increase with age,

educational attainment and work hours, are greater for whites and residents of metropolitan areas and are also superior for those in higher level occupations (Level I), such as managers, directors and professionals, than in other

• ccupational levels.

Quantile regression results show that the magnitude of marriage and cohabitation premiums increase with quantiles (TABLE 3) 5. Important changes appear in coefficients associated with consensual union and marriage, according to the position in distribution of earnings, especially for employees. For example, for the lower-paid employees (0.10 quantiles), marriage premium is 13.8%, whereas for employees with the highest wages (0.90 quantiles), this premium rises to 28.9%, compared to single employees. For the self-employed, the earnings increase associated with marriage is more stable over earning distribution. Self-employed married men whose earning is in the bottom 10% of the sample receive about 33.8% more than single men, and the marriage premium among those with the highest earnings increases to 37.7%. Irrespective of the quantile observed, marriage and cohabitation premiums are always higher for the self- employed than for employees. TABLE 3 – Quantile regression for log of wage/earning, coefficients on marital status, employee and self-employed, Brazilian men, 2010. Employee (N=444.742) Log of wage Quantiles 0.10 0.25 0.50 0.75 0.90 Marital Status [Consensual Union] 0.078 *** 0.103 *** 0.120 *** 0.133 *** 0.135 *** Marital Status [Married] 0.130 *** 0.181 *** 0.219 *** 0.245 *** 0.254 *** Self-employed (N=163.376) Log of earning Quantiles 0.10 0.25 0.50 0.75 0.90 Marital Status [Consensual Union] 0.155 *** 0.155 *** 0.175 *** 0.188 *** 0.203 *** Marital Status [Married] 0.291 *** 0.293 *** 0.307 *** 0.321 *** 0.320 ***

Source of basic data: Brazilian Demographic Census 2010 - IBGE / IPUMS-I. Levels of significance: *** ≤0.001; ** ≤0.010; * ≤0.050; "." ≤0.100; "ns" not significant at the 10% level. Note: complete results from each quantile are available from the authors.

The results obtained show clear evidence of marriage and cohabitation premiums. The initial hypothesis was that if marriage premium was higher for employees than for the self-employed, employer discrimination would probably be a significant factor explaining the phenomenon. However, the reward for marriage and cohabitation seems to be greater for self-employed workers than for employees, which contradicts the initial hypothesis. In this case, one possible hypothesis to explain the highest premium for self-employed workers lies in the role of the spouse as coproducer of masculinity, which ultimately encourages men to assume the social

5 Full results of the estimates are provided in Appendix 1, TABLE A1.1 and TABLE A1.2.

role of main family breadwinner (Ashwin and Isupova, 2014), positively influencing productivity and, consequently, income. The higher marriage premium than cohabitation premium found would reinforce this hypothesis, because it relates more traditional unions with a higher level of commitment and stability. WIFE/PARTNER’S CHARACTERISTICS In this section, estimates take into account the theories on the role of the wife/partner on male

• income. The aim is to verify whether and how women's educational attainment, work, and

potential labor supply are associated with male incomes. To that end, only married men and men in consensual union are considered. Regression models were separately estimated for employee and self-employed. The dependent variable is the log of monthly wage/earning and the variables of interest are: type of union, wife/partner‟s educational attainment and wife/partner‟s work hours. Control variables are the same as those of previous models. In the first model, the observed work hours of the woman were included. In the second model, the work hours predicted by the Tobit regression6 was included, in order to control the possible endogeneity of wives‟ or partners‟ hours of work and spouses‟ earning (income effect). Results presented in TABLE 4 show that marriage appears to be associated with higher remuneration than consensual union. In addition marital premium is higher for the self-employed than for employees, which is consistent with the previous results. For the self-employed, marriage is related to a 11.4% increase in earning and, for employees, a 7.4% increase in wage levels, in relation to men in a consensual union. In all models and specifications, wives/partners' educational attainment is positively associated to male earnings. The effect is greater for the self-employed than for employees and increases with the woman‟s educational level. For example, being married or in union with a woman with college education is associated with a 49.4% increase in wage for employees and a 64.9% increase in earnings for the self-employed, in comparison to being with a woman without

• schooling. This result may be explained both by positive assortative mating, where high-income

men are more likely to marry highly educated women, and by the joint human capital hypothesis, which suggests that the high educational attainment of the wife/partner increases the man's stock of human capital, and therefore, his productivity and income (Birch and Miller, 2006; Mamun, 2012). This theory seems more adequate to explain the results found, since the rewards are higher as the woman's educational level increases.

6 Results of Tobit estimation for wife/partner's working hours can be found in Appendix 2, TABLE A2.

TABLE 4 – OLS and OLS with instrumental variable coefficients for log of wage/earning, employee and self-employed, Brazilian men, 2010. Log of wage/earning Employee Self-employed OLS OLS + Tobit OLS OLS + Tobit Type of Union [Married] 0.072 *** 0.072 *** 0.108 *** 0.106 *** Wife/partner – Education [Elementary] 0.079 *** 0.077 *** 0.140 *** 0.134 *** Wife/partner – Education [High school] 0.168 *** 0.161 *** 0.286 *** 0.276 *** Wife/partner – Education [College] 0.402 *** 0.391 *** 0.500 *** 0.499 *** Wife/partner – Part time work

• 0.051 ***

0.006 Ns 0.007 ns 0.075 *** Wife/partner – Full-time work

• 0.050 ***
• 0.021 **
• 0.013 **

0.034 * Age 0.038 *** 0.037 *** 0.044 *** 0.042 *** Age -squared 0.000 *** 0.000 *** 0.000 *** 0.000 *** Education [Elementary] 0.095 *** 0.094 *** 0.146 *** 0.144 *** Education [High school] 0.230 *** 0.230 *** 0.293 *** 0.292 *** Education [College] 0.732 *** 0.732 *** 0.608 *** 0.608 *** Color [White] 0.092 *** 0.092 *** 0.125 *** 0.126 *** Color [Black]

• 0.022 ***
• 0.023 ***
• 0.027 ***
• 0.029 ***

Area [Metropolitan] 0.072 *** 0.072 *** 0.121 *** 0.120 *** State (26 dummies) (omitted) (omitted) Work hours [Part-time]

• 0.014 **
• 0.013 *
• 0.205 **
• 0.162 **

Work hours [Excessive] 0.099 *** 0.103 *** 0.146 *** 0.151 *** Industry [Construction]

• 0.074 ***
• 0.073 ***
• 0.058 ***
• 0.055 ***

Industry [Trade]

• 0.123 ***
• 0.124 ***

0.038 *** 0.038 *** Industry [Services]

• 0.068 ***
• 0.070 ***

0.056 *** 0.056 *** Occupation [Level II]

• 0.338 ***
• 0.338 ***
• 0.210 ***
• 0.215 ***

Occupation [Level III]

• 0.533 ***
• 0.534 ***
• 0.369 ***
• 0.373 ***

Occupation [Level IV]

• 0.449 ***
• 0.449 ***
• 0.393 ***
• 0.396 ***

Occupation [Level V]

• 0.547 ***
• 0.546 ***
• 0.447 ***
• 0.449 ***

Constant 6.288 *** 6.279 *** 6.027 *** 5.995 ***

• N. Obs.

361364 140921 R-squared 0.4800 0.4789 0.3171 0.3178

Source of basic data: Brazilian Demographic Census 2010 - IBGE / IPUMS-I. Levels of significance: *** ≤0.001; ** ≤0.010; * ≤0.050; "." ≤0.100; "ns" not significant at the 10% level. Note: OLS = ordinary least squares regression; MQO + Tobit = linear regression with instrumental variable (wife/partner work hours estimated using the Tobit model).

For employees, working wives/partners are associated with a 4.9% reduction in wages, all variables held constant. After correcting for the income effect, part-time work is no longer statistically significant. However, the coefficient for full time work, although reduced, remains negative and significant. This negative effect is due exclusively to married men. For men in union, both coefficients for the estimated working hours (part-time and full-time) become

• insignificant7. A possible explanation is the possibility that, among married couples,

specialization may be higher than among couples in consensual union, as advocated by Becker (1991) in his specialization and comparative advantages model (Zavodny, 2008). For the self-employed, the effect of the wife/partner's work is negative when the woman works full-time, in the first regression. After controlling for the income effect in the second regression, coefficients become positive. For self-employed men, being married or in union with women with predicted part time work and full time work are associated with a 7.8% and 3.5% increase in earnings, respectively, in comparison with women with prediction of no work. Therefore, the income effect seems to explain results for the self-employed, but not for married

• employees. Among employees, especially married ones, even after correcting for income effect,

the correlation remains negative for predicted full time work. For the self-employed, initially the wife's work shows a negative correlation with male earnings. But, once the income effect is controlled, the correlation becomes positive. In addition to the income effect, other mechanisms may be behind the reversal of earning penalty associated with a working wife/partner for the self-employed. There may be complementarity between productivities of spouses or self- employed men may be able to seek better occupations or invest in human capital if his wife or partner can play the role of family provider (Jacobsen and Rayack, 1996). FINAL CONSIDERATIONS This article aimed to verify the existence of a male earning premium associated with marriage and cohabitation in Brazil. Evidence of premium was found both for marriage and cohabitation, with the marriage premium being higher than the cohabitation premium. This issue deserves further investigation, given that the differences between the two types of union in the Brazilian case are not as clear as in other countries. The initial hypothesis of favorable employer discrimination in relation to married men could not be confirmed, since higher premium was

• bserved for self-employed workers than for employees.

The analysis showed that marriage and cohabitation premiums increase throughout earning distribution, in line with Budig‟s (2014) findings when examining paternity (not marriage) premiums and also found higher premium for men with higher income levels. In turn, female maternity penalty is lower at higher income levels and higher for lower income women (Budig, 2014). There is certain consensus on male marriage wage premium, as well as on maternity

• penalty. However, studies analyzing the association between marriage and wages for women

7 Separate estimates for men in consensual union and married men are provided in Appendix 3, TABLES A3.1 and

A3.2.

show much more discrepant results. Some studies have found positive associations between marriage and labor income (Muniz and Rios-Neto, 2002; Killewald and Gough, 2013), whereas

• thers observed no gains for women (Adler and Oner, 2013). Even when positive, the effect of

marriage is not as clear (Muniz and Rios-Neto, 2002), and often presents reduced magnitude and/or statistical insignificance (Adler and Oner, 2013). In any case, when marriage is associated with motherhood, wage penalties associated with the existence of children have a dominant effect and outweigh marriage gains (Killewald and Gough, 2013). The second aim of this article was to examine whether the productive characteristics of the wife

• r partner, namely educational attainment and work hours, were in any way associated to male
• earnings. It was found that wives/partners' educational attainment is positively and increasingly

associated with men's earnings, in accordance with the joint human capital hypothesis, which expects that the wife/partner's educational attainment increases their spouse‟s stock of human capital, raising his productivity and income. The wage penalty associated with the work of the wife of married employees persists even after controlling for the income effect. Similar penalty is initially observed for the self-employed as well, but after controlling the endogeneity of the woman's work hours, the coefficients become positive, suggesting a premium for the potential work hours of the wife/partner. We argue that, for employees, the specialization hypothesis seems to be more adequate to explain these results and, for the self-employed, evidence points to the income effect as well as to a possible complementarity between spouses‟ productivities. However, this issue should also be further investigated in future studies in order to reach a satisfactory conclusion. REFERENCES ADLER, P.; ONER, O. Occupational Class and the Marriage Premium: Exploring Treatment

• Mechanisms. UCLA: The Institute for Research on Labor and Employment, 2013. (Working

Papers). AHITUV, A.; LERMAN, R. I. How Do Marital Status, Work Effort, and Wage Rates Interact? Demography, v. 44, n. 03, pp. 623-647, 2007. ASHWIN, S.; ISUPOVA, O. “Behind Every Great Man…”: The Male Marriage Wage Premium Examined Qualitatively. Journal of Marriage and Family, v. 76, pp. 37-55, fev. 2014. BECKER, G. A treatise on the family. Cambridge: Harvard University Press, 1991. BIRCH, E. R.; MILLER, P. W. How does marriage affect the wages of men in Australia?. Economic Record, v. 82, n. 257, pp. 150-164, 2006. BLACKABY, D. H.; CARLIN, P. S.; MURPHY, P. D. A Change in the Earnings Penalty for British Men with Working Wives: Evidence from the 1980's and 1990's. Labour Economics, v. 14, n. 01,

• p. 119-134, 2007.

BUDIG, M. J. The fatherhood bonus and the motherhood penalty: Parenthood and the gender gap in pay. Washington, DC: Third Way, 2014. CHIODO, A. J.; OWYANG, M. T. For Love or Money: Why Married Men Make More. The Regional Economist, St. Louis: Federal Reserve Bank of St. Louis, abr. 2002. CHUN, H.; LEE, I. Why Do Married Men Earn More: Productivity or Marriage Selection? Economic Inquiry, v. 39, n. 02, pp. 307-319, 2001. COVRE-SUSSAI, M.; MEULEMAN, B.; BOTTERMAN, S.; MATTHIJS, K. Traditional and modern cohabitation in Latin America: A comparative typology. Demographic Research, v.32, article 32, pp. 873-914, 2015. ESTEVE, A.; LESTHAEGHE, R.; LÓPEZ‐GAY, A. The Latin American cohabitation boom, 1970–2007. Population and Development Review, v. 38, n. 01, pp. 55-81, 2012. HERSCH, J.; STRATTON, L. Household Specialization and the Male Marriage Wage Premium. Industrial and Labor Relations Review, v. 54, n. 01, pp. 78-94, out. 2000. (Discussion paper n. 298). JACOBSEN, J. P.; RAYACK, W. L. Do men whose wives work really earn less?. The American Economic Review, v. 86, n. 02, pp. 268-273, 1996. KILLEWALD, A.; GOUGH, M. Does specialization explain marriage penalties and premiums?. American sociological review, v. 78, n. 03, pp. 477-502, abr. 2013. MAMUN, A. Cohabitation premium in men‟s earnings: Testing the joint human capital

• hypothesis. Journal of Family and Economic Issues, v. 33, n. 01, pp. 53-68, 2012.

MUNIZ, J. O.; RIOS-NETO, E. L. G. Diferenciais Salariais por Estado Civil e Sexo: uma Análise de Gênero sobre o Prêmio do Casamento. In: MUNIZ, J. O. Demografia Econômica: Aplicações Macro e Micro ao Caso Brasileiro. 2002. Dissertação (Mestrado em Demografia). Belo Horizonte: UFMG, 2002. pp. 48-79. OPICE, I. B. Os efeitos do trabalho da mulher no salário do marido para o Brasil. 2010. Monografia (Graduação em Economia e Administração) – São Paulo: Insper, 2010. SONG, Y. The working spouse penalty/premium and married women‟s labor supply. Review of Economics of the Household, v. 05, n. 03, pp. 279-304, 2007. SOUZA, P. F. L. de; FRANÇA, J. M. S. Casamento: penalização salarial para as mulheres e prêmio para

• s

homens. UFC: CAEN, 2013. Disponível em: <http://www.caen.ufc.br/attachments/article/251/Casamento_penaliza%C3%A7%C3%A3o_sala rial_para_as_mulheres_e_pr%C3%AAmio_para_os_homens.pdf>. Acesso em: 04 dez. 2016. VERBAKEL, E.; DE GRAAF, P. M. Partner effects on labour market participation and job level:

• pposing mechanisms. Work, employment and society, v. 23, n. 04, pp. 635-654, 2009.

ZAVODNY, M. Is there a „marriage premium‟ for gay men?. Review of Economics of the Household, v. 06, n. 04, pp. 369-389, 2008.

APPENDIX 1 TABLE A1.1 - Quantile regression coefficients for log of wage, employee, Brazilian men, 2010. Employee Log of wage Quantiles 0.10 0.25 0.50 0.75 0.90 Marital status [Consensual union] 0.078 *** 0.103 *** 0.120 *** 0.133 *** 0.135 *** Marital status [Married] 0.130 *** 0.181 *** 0.219 *** 0.245 *** 0.254 *** Age 0.016 *** 0.026 *** 0.033 *** 0.039 *** 0.041 *** Age-squared 0.000 *** 0.000 *** 0.000 *** 0.000 *** 0.000 *** Education [Elementary] 0.064 *** 0.090 *** 0.120 *** 0.146 *** 0.160 *** Education [High school] 0.138 *** 0.214 *** 0.288 *** 0.360 *** 0.410 *** Education [College] 0.503 *** 0.694 *** 0.884 *** 1.037 *** 1.106 *** Color [White] 0.042 *** 0.068 *** 0.092 *** 0.114 *** 0.134 *** Color [Black]

• 0.008 *** -0.012 *** -0.015 *** -0.026 *** -0.034 ***

Area [Metropolitan] 0.047 *** 0.059 *** 0.065 *** 0.071 *** 0.073 *** States (26 dummies) (omitted) Work hours [Part-time]

• 0.054 *** -0.044 *** -0.025 ***

0.006 ns 0.043 *** Work hours [Excessive] 0.018 *** 0.046 *** 0.083 *** 0.130 *** 0.167 *** Industry [Construction]

• 0.037 *** -0.049 *** -0.072 *** -0.096 *** -0.115 ***

Industry [Trade]

• 0.064 *** -0.094 *** -0.116 *** -0.131 *** -0.130 ***

Industry [Services]

• 0.045 *** -0.048 *** -0.049 *** -0.055 *** -0.066 ***

Occupation [Level II]

• 0.250 *** -0.331 *** -0.396 *** -0.419 *** -0.428 ***

Occupation [Level III]

• 0.315 *** -0.448 *** -0.580 *** -0.678 *** -0.712 ***

Occupation [Level IV]

• 0.286 *** -0.380 *** -0.479 *** -0.570 *** -0.649 ***

Occupation [Level V]

• 0.336 *** -0.469 *** -0.590 *** -0.698 *** -0.784 ***

Constant 6.136 *** 6.127 *** 6.262 *** 6.438 *** 6.714 ***

• N. Obs.

444.742 Pseudo R2 0.1176 0.2103 0.2482 0.3068 0.3529

Source of basic data: Brazilian Demographic Census 2010 - IBGE / IPUMS-I. Levels of significance: *** ≤0.001; ** ≤0.010; * ≤0.050; "." ≤0.100; "ns" not significant at the 10% level.

TABLE A1.2 - Quantile regression coefficients for the log of earnings, self-employed, Brazilian men, 2010. Self-employed Log of Earnings Quantiles 0.10 0.25 0.50 0.75 0.90 Marital Status [Consensual Union] 0.155 *** 0.155 *** 0.175 *** 0.188 *** 0.203 *** Marital Status [Married] 0.291 *** 0.293 *** 0.307 *** 0.321 *** 0.320 *** Age 0.033 *** 0.039 *** 0.037 *** 0.046 *** 0.055 *** Age – squared 0.000 *** 0.000 *** 0.000 *** -0.001 *** -0.001 *** Education [Elementary] 0.184 *** 0.201 *** 0.177 *** 0.188 *** 0.195 *** Education [High school] 0.386 *** 0.385 *** 0.394 *** 0.423 *** 0.418 *** Education [College] 0.695 *** 0.783 *** 0.820 *** 0.894 *** 0.927 *** Color [White] 0.106 *** 0.116 *** 0.130 *** 0.151 *** 0.160 *** Color [Black]

• 0.004 ns
• 0.020 *
• 0.025 **
• 0.029 **
• 0.042 **

Area [Metropolitan] 0.139 *** 0.125 *** 0.119 *** 0.105 *** 0.086 *** States (26 dummies) (omitted) Work hours [Part-time]

• 0.418 *** -0.265 *** -0.177 *** -0.125 *** -0.075 ***

Work hours [Excessive] 0.094 *** 0.115 *** 0.134 *** 0.169 *** 0.222 *** Industry [Construction] 0.016 ns

• 0.020 *
• 0.058 *** -0.117 *** -0.188 ***

Industry [Trade] 0.058 *** 0.035 *** 0.045 *** 0.063 *** 0.084 *** Industry [Services] 0.084 *** 0.060 *** 0.077 *** 0.106 *** 0.119 *** Occupation [Level II]

• 0.093 *** -0.140 *** -0.205 *** -0.293 *** -0.390 ***

Occupation [Level III]

• 0.257 *** -0.306 *** -0.403 *** -0.494 *** -0.591 ***

Occupation [Level IV]

• 0.202 *** -0.283 *** -0.406 *** -0.538 *** -0.670 ***

Occupation [Level V]

• 0.352 *** -0.389 *** -0.484 *** -0.590 *** -0.674 ***

Constant 5.290 *** 5.534 *** 6.031 *** 6.314 *** 6.627 ***

• N. Obs.

163376 Pseudo R2 0.1893 0.1543 0.1677 0.1806 0.1821

Source of basic data: Brazilian Demographic Census 2010 - IBGE / IPUMS-I. Levels of significance: *** ≤0.001; ** ≤0.010; * ≤0.050; "." ≤0.100; "ns" not significant at the 10% level.

APPENDIX 2 TABLE A2 - Tobit estimation results – predicted working hours of wife/partner, according to status in employment, Brazilian men, 2010. Independent Variables Employee Self-employed (N=361.365) (N=140.921) Wife/Partner’s Age 3.704 *** 3.822 *** Age – squared

• 0.048 ***
• 0.047 ***

Education [Elementary] 3.373 *** 4.658 *** Education [High school] 10.540 *** 11.806 *** Education [College] 22.699 *** 21.315 *** Color [White] 1.079 *** 1.088 *** Color [Black] 2.192 *** 2.080 *** N° Children [1]

• 2.986 ***
• 2.704 ***

N° Children [2]

• 5.907 ***
• 5.093 ***

N° Children [3]

• 7.626 ***
• 7.709 ***

N° Children [4 or more]

• 9.471 ***
• 9.159 ***

Presence of children <7 y.o.

• 7.412 ***
• 6.842 ***

Area [Metropolitan]

• 0.116 ns

0.132 ns State (26 dummies) (omitted) (omitted) Husband/Partner’s Age 0.250 *

• 0.198 ns

Age-squared

• 0.007 ***
• 0.001 ns

Education [Elementary] 1.320 *** 1.566 *** Education [High school] 0.186 ns 1.520 *** Education [College]

• 3.615 ***
• 1.389 *

Color [White]

• 1.034 ***
• 1.421 ***

Color [Black] 1.111 *** 0.776 ns Work hours [part-time]

• 19.003 ***
• 16.909 ***

Work hours [Excessive] 9.183 *** 13.892 *** Industry [Construction]

• 1.382 ***
• 3.861 ***

Industry [Trade] 3.138 ***

• 0.424 ns

Industry [Services] 3.202 ***

• 1.533 ***

Occupation [Level II] 0.240 ns

• 5.416 ***

Occupation [Level III] 2.594 ***

• 1.048 *

Occupation [Level IV] 0.255 ns

• 7.502 ***

Occupation [Level V] 0.087 ns

• 7.154 ***

Constant

• 52.357 ***
• 43.111 ***

Sigma 32.504 *** 34.296 ***

Source of basic data: Brazilian Demographic Census 2010 - IBGE / IPUMS-I. Levels of significance: *** ≤0.001; ** ≤0.010; * ≤0.050; "." ≤0.100; "ns" not significant at the 10% level.

APPENDIX 3 TABLE A3.1 – OLS and OLS with instrumental variable coefficients for log of wage, employee, separated by marital status, Brazilian men, 2010. EMPLOYEE Log of wage Married In Union OLS OLS + Tobit OLS OLS + Tobit Wife/partner – Education [Elementary] 0.080 *** 0.077 *** 0.078 *** 0.077 *** Wife/partner – Education [High school] 0.164 *** 0.154 *** 0.175 *** 0.172 *** Wife/partner – Education [College] 0.399 *** 0.382 *** 0.405 *** 0.401 *** Wife/partner – Part time work

• 0.077 ***

0.009 ns

• 0.007 ns

0.006 ns Wife/partner – Full-time work

• 0.070 ***
• 0.024 *
• 0.016 ***

0.001 ns Age 0.044 *** 0.043 *** 0.033 *** 0.033 *** Age -squared 0.000 *** 0.000 *** 0.000 *** 0.000 *** Education [Elementary] 0.100 *** 0.099 *** 0.087 *** 0.086 *** Education [High school] 0.243 *** 0.244 *** 0.211 *** 0.211 *** Education [College] 0.740 *** 0.742 *** 0.704 *** 0.704 *** Color [White] 0.103 *** 0.104 *** 0.072 *** 0.072 *** Color [Black]

• 0.028 ***
• 0.030 ***
• 0.016 ***
• 0.017 ***

Area [Metropolitan] 0.089 *** 0.090 *** 0.040 *** 0.040 *** State (26 dummies) (omitted) (omitted) Work hours [Part-time]

• 0.009 ns
• 0.010 ns
• 0.022 **
• 0.014 .

Work hours [Excessive] 0.108 *** 0.115 *** 0.086 *** 0.086 *** Industry [Construction]

• 0.091 ***
• 0.090 ***
• 0.048 ***
• 0.048 ***

Industry [Trade]

• 0.133 ***
• 0.135 ***
• 0.101 ***
• 0.102 ***

Industry [Services]

• 0.088 ***
• 0.091 ***
• 0.030 ***
• 0.030 ***

Occupation [Level II]

• 0.342 ***
• 0.342 ***
• 0.314 ***
• 0.314 ***

Occupation [Level III]

• 0.557 ***
• 0.558 ***
• 0.476 ***
• 0.476 ***

Occupation [Level IV]

• 0.471 ***
• 0.471 ***
• 0.393 ***
• 0.393 ***

Occupation [Level V]

• 0.566 ***
• 0.565 ***
• 0.497 ***
• 0.497 ***

Constant 6.192 *** 6.172 *** 6.240 *** 6.237 ***

• N. Obs.

223493 137872 R-squared 0.4860 0.4840 0.3944 0.3942

Source of basic data: Brazilian Demographic Census 2010 - IBGE / IPUMS-I. Levels of significance: *** ≤0.001; ** ≤0.010; * ≤0.050; "." ≤0.100; "ns" not significant at the 10% level. Note: OLS = ordinary least squares regression; MQO + Tobit = linear regression with instrumental variable (wife/partner work hours estimated using the Tobit model).

TABLE A3.2 – OLS and OLS with instrumental variable coefficients for log of earning, self- employed, separated by marital status, Brazilian men, 2010. SELF-EMPLOYED Log of earnings Married In Union OLS OLS + Tobit OLS OLS + Tobit Wife/partner – Education [Elementary] 0.116 *** 0.107 *** 0.160 *** 0.156 *** Wife/partner – Education [High school] 0.250 *** 0.238 *** 0.322 *** 0.312 *** Wife/partner – Education [College] 0.462 *** 0.462 *** 0.556 *** 0.548 *** Wife/partner – Part time work

• 0.015 ns

0.096 *** 0.027 ns 0.061 *** Wife/partner – Full-time work

• 0.032 ***

0.032 ns 0.011 ns 0.061 ** Age 0.050 *** 0.049 *** 0.041 *** 0.039 *** Age -squared

• 0.001 ***
• 0.001 ***

0.000 *** 0.000 *** Education [Elementary] 0.153 *** 0.151 *** 0.135 *** 0.133 *** Education [High school] 0.292 *** 0.292 *** 0.296 *** 0.295 *** Education [College] 0.596 *** 0.595 *** 0.651 *** 0.653 *** Color [White] 0.135 *** 0.136 *** 0.112 *** 0.113 *** Color [Black]

• 0.019 ns
• 0.020 ns
• 0.033 **
• 0.035 **

Area [Metropolitan] 0.125 *** 0.125 *** 0.116 *** 0.115 *** State (26 dummies) (omitted) (omitted) Work hours [Part-time]

• 0.202 ***
• 0.155 ***
• 0.205 ***
• 0.167 ***

Work hours [Excessive] 0.141 *** 0.156 *** 0.153 *** 0.147 *** Industry [Construction]

• 0.078 ***
• 0.075 ***
• 0.040 **
• 0.037 **

Industry [Trade] 0.039 *** 0.038 *** 0.033 * 0.033 * Industry [Services] 0.034 ** 0.033 ** 0.075 *** 0.075 *** Occupation [Level II]

• 0.209 ***
• 0.217 ***
• 0.206 ***
• 0.205 ***

Occupation [Level III]

• 0.382 ***
• 0.386 ***
• 0.342 ***
• 0.343 ***

Occupation [Level IV]

• 0.435 ***
• 0.441 ***
• 0.333 ***
• 0.331 ***

Occupation [Level V]

• 0.425 ***
• 0.430 ***
• 0.454 ***
• 0.452 ***

Constant 5.951 *** 5.881 *** 5.940 *** 5.927 ***

• N. Obs.

80946 59975 R-squared 0.3002 0.3010 0.2937 0.2940

Source of basic data: Brazilian Demographic Census 2010 - IBGE / IPUMS-I. Levels of significance: *** ≤0.001; ** ≤0.010; * ≤0.050; "." ≤0.100; "ns" not significant at the 10% level. Note: OLS = ordinary least squares regression; MQO + Tobit = linear regression with instrumental variable (wife/partner work hours estimated using the Tobit model).