SHORT ABSTRACT Understanding the links between womens employment and - - PDF document

SHORT ABSTRACT Understanding the links between women’s employment and nutritional status in rapidly developing economies: the cases of Brazil and India Stephanie Bispo, Amos Channon, University of Southampton Maternal work changes household dynamics, what can affect child nutrition, particularly in places with high rates of stunting. The association between maternal work and child nutrition has not been well understood in the context of emerging countries, where opportunities for women to work in different sectors are growing rapidly. This study aimed to explore changes in female labour force participation in Brazil and India, and to investigate how such changes affect child stunting. Two waves of the Demographic and Health Survey were used for Brazil and India. Models were performed based on hypotheses regarding the trade-off between earnings from work and lack of time for childcare. Models for India were stratified by rural/urban areas, and family structure, since the first affect the types of work offered for women, and the second affects availability for childcare. Random-intercept models showed that in Brazil, most combinations of work led to positive effects for child growth, while in India, maternal work was overall associated to higher likelihood of having a stunting child. When maternal work was interacted with education and higher status, the positive effect of maternal work was also observed in India. The differences in findings between both countries occurred mainly because of the stage of country’s development, and the conditions of work offered for women. Findings suggest that improving maternal education, and policies protecting the working mother can affect child nutrition positively.

INTRODUCTION Malnutrition affects many children in low and middle income countries, with highlights on its prevalence in India and sub-Sahara Africa. It is estimated that at least 165 million children are stunted worldwide, from which 69 million live in south-central Asia, and 51 million live in Sub Sahara Africa (Black et al., 2013). Solid evidence shows that children from different areas and ethnic backgrounds can experience normal and similar patterns of weight and height when raised under appropriate conditions (de Onis et al., 2007). Therefore, high rates of stunting are unacceptable, especially when it refers to countries experiencing accentuated economic growth. Economic growth can lead to direct investments on nutrition, what is an opportunity for improving rates of malnutrition, economic inequality and poverty. It also can affect child nutrition indirectly through improving opportunities of work for women. There is an ongoing concern in the literature about how female labour force participation (FLFP) affects child health and nutrition, based on the important role of women in performing household tasks and as the main child caregiver (Anderson et al., 2003, Brown et al., 2010, Lamontagne et al., 1998). The association between maternal work and child nutrition has been extensively studied in low income countries during the eighties and nineties (Abbi et al., 1991, Desai and Jain, 1994, Engle, 1991). The research was driven by the concern regarding social and economic roles of women in times of economic recession and high rates of child mortality. Results from various studies from this time came with some limitations, where in most cases, analyses were not controlled by covariates, no characteristics of maternal work were investigated, and no consistent findings or mechanisms of how work impacts on child nutrition were provided (Leslie, 1988). Many studies discussed that in low income countries maternal work was positively associated with child nutritional outcomes due to payments in cash or kind, suggesting that the more the mother earned, the better the nutritional status of the child (Engle, 1993, Lamontagne et al., 1998, Leslie, 1988). Indeed, in situations of poverty, maternal work is a chance to contribute to household income and increase household expenditure, as demonstrated in many developing countries, such as the Dominic Republic, Guatemala and the Philippines (Johnson and Rogers, 1988, Popkin and Solon, 1976, Engle and Pedersen, 1989). Other studies did not find this association, suggesting that the income benefit did not counter the deleterious effects of the limited time available for child care and breastfeeding (Toyama et al., 2001, Ukwuani and Suchindran, 2003, Abbi et al., 1991). The main argument was that feeding and taking care of the child is very time consuming and nutritional

deficiencies or morbidity during childhood can have enormous effect on the child, especially in the context of poverty (Lamontagne et al., 1998, Glick and Sahn, 1998, Popkin and Solon, 1976). The complexity of this association is evidenced when considering work and household dynamics, both influenced by the context where the woman lives in. While women’s work for earnings is a dynamic process that can change according to demands, non-paid work can occupy as much time of mothers without being adequately captured by data as “work” (Engle, 1989; Desai & Jain, 1994). This generates a constraint in time available for child care, not necessarily bringing benefits for child health and nutrition. Furthermore, having a good quality substitute caregiver, or having other adults in the household to share the responsibility for childcare can attenuate negative effects of the absence of the mother due to working duties (Tucker, 1989, Lamontagne et al., 1998). While changes in FLFP and women’s role occurred slowly in developed countries, this experience in emerging countries1 have been much quicker in the midst of poverty and high rates of child

• malnutrition. The concern with child nutrition in these countries highlights the importance of

understanding the nature of work offered to women during economic growth and whether it worsens or improves child outcomes. Both topics, female work and child nutrition, are considered key priorities in international development, and have been included in the sustainable development goals, targeting to improve nutrition, to achieve gender equality, to empower girls and women, and free women from unpaid work (Hawkes and Popkin, 2015). Moreover, it is at the most interest of emerging countries to have continuous economic growth, what is proportionated by FLFP and improved nutrition. This paper explores maternal work and child nutrition in Brazil and India, two countries with very different backgrounds that have experienced accelerated economic growth since the end of 1990s (O'Neill, 2001), aiming to investigate how maternal work affects child height-for-age before and during economic growth, and how the nature of work can influence this association. Theoretical Framework The theoretical framework used in this study follows two main bodies of the literature, the first focusing on the economics of household decision making, and the other accessing women’s labour

1 “Emerging countries” is a term used to describe countries based in six criteria: economic growth rate;

economic liberalization (market-oriented systems); export growth rate; financial market development; level and velocity of IT development; and political influence. These countries are very heterogeneous, and despite of economic development, they are often facing different levels of inequalities in income distribution, education and health (see OECD, 2011).

force participation and child-care decisions2. Those theories served as ground for understanding how maternal work affects child nutrition, and hypothesis were raised in the attempt of evaluating associations for different characteristics of maternal work. Since Becker (1965), a child has been presented as a consumer durable in the family, and a time- intensive consumer that can affect FLFP. Assuming that the woman is a mother of a child younger than five years old, a standard utility-maximizing framework models her time as: 𝑉 = 𝑉(𝑌, 𝑅, 𝑀), Where 𝑌 represents a composite market good, 𝑀 is the mother’s leisure and 𝐷 is child care. Both, child care and maternal work are seen as antagonists, as they create opposite time and income

• constraints. Time constraints are imposed when the mother allocates her time between hours

worked in the market (𝐼), leisure, and child care, while the child’s time is divided between maternal (𝐷𝑛) and non-maternal care (𝐷𝑜). 𝐼 + 𝐷𝑛 + 𝑀 = 1 𝐷𝑛 + 𝐷𝑜 = 1 The income constraint comes from the fact that maternal earnings from work (𝐹), which adds income to the household (𝑍) can also be spent on non-maternal care, or in other needs of the child (N). 𝑂 + 𝐷𝑜 = 𝐹𝐼 + 𝑍 Child nutrition status (𝐷𝑂𝑇) at time 𝑢 can be written as a function of the income inputs invested in the child’s health (𝐽), time inputs invested in the child (𝐷), and a set of exogenous variables that directly and indirectly affect child nutrition, such as biological and environmental characteristics (𝑎), 𝐷𝑂𝑇 = 𝐺(𝐽𝑗𝑢, 𝐷𝑗𝑢, 𝑎𝑗𝑢) Since maternal work can affect all parts of this function, the positive effect of work on child nutrition can be seen when the work function encounters a positive or balanced trade-off between the maternal time and income. 𝑋 ∗> 0 𝑗𝑔 (𝐹𝐼 + 𝑍) + (𝐷𝑛 + 𝐷𝑜) > 0

2 See Connelly (1992), and Becker (1965).

Translating this into a regression equation, child nutrition status is a result of the income generated by work (𝐽𝑗𝑘𝑙), the quality maternal care and non-maternal care offered to the child (𝐷𝑗𝑘𝑙) and the biological and environment characteristics in which the child is raised (𝑎𝑗𝑘𝑙), all for each child 𝑗, from each mother 𝑘 and region 𝑙 . Error components are represented by 𝜈jk, as a random intercept for mother j and state k, 𝜀𝑙 as the random intercept for state k, and 𝜗𝑗𝑘𝑙 as specific to each child i from mother j and state k .

(1) 𝐼𝐵𝑎𝑗𝑘𝑙 = 𝛾1 + 𝛾2𝐽𝑗𝑘𝑙 + 𝛾3𝐷𝑗𝑘𝑙 + 𝛾4𝑎𝑗𝑘𝑙 + 𝜈𝑘𝑙 + 𝜀𝑙 + 𝜗𝑗𝑘𝑙,

This equation illustrates the ideal evaluation of the effect of work on child nutrition. However, such a detailed data is hardly ever obtained in the literature. Therefore, this study raised hypothesis regarding income and time availability for different characteristics of maternal work based on the

• literature. Those are discussed separately for Brazil and India.

Brazil Despite the modest economic growth when compared to other emerging countries, Brazil has undergone many social changes in recent years culminating in poverty reduction, improved education and better opportunities of work for women (van Klaveren et al., 2009). Following the same pattern found in developing countries, Brazil had significant increase in rates of FLFP, in which women are highly educated (Agenor and Canuto, 2015). The literature for developed countries highlight that self-employment is related to higher availability for parental child care than other types of employment (Hundley, 2000, Bianchi and Casper, 2000, Wellington, 2006). However, the Brazilian labour market has a great number of non-professional self-employed activities in which women spend more time outside home, such as itinerant business, selling products, and self-employed housemaids (Wajnman et al., 1998). Considering that self- employed mothers working at home and away from home are distinct and should be evaluated separately, the following hypotheses were raised: 1- Self-employed mothers working at home have more time with the child, what adds to the additional income generated, potentiating the benefits for the chid: CNS self-employed/home = 𝑌𝑗𝑘𝑙 + 𝐷𝑗𝑘𝑙(𝐽𝑗𝑘𝑙) + 𝑣𝑗 2- For employed mothers, the additional income, alongside with working benefits (e.g. regular income, unemployment insurance, and maternity leave) can overcome the limited time for childcare, having a positive effect on child nutrition: CNS employed = 𝑌𝑗𝑘𝑙 + 𝐽𝑗𝑘𝑙 𝑦 (𝐽𝑗𝑘𝑙 − 𝐷𝑗𝑘𝑙) + 𝑣𝑗

3- Self-employed mothers working away from home have the most negative effect on child nutrition due to the lack of time available for child care, lower job security and lower payment: CNS self-employed/away= 𝑌𝑗𝑘𝑙 + 𝐽𝑗𝑘𝑙 – 𝐷𝑗𝑘𝑙 + 𝑣𝑗 The growing female education in Brazil also suggests that an interaction between maternal work and education could result in better income and better choices of childcare, according to the hypotheses below: 1- For non-working mothers with lower education, the income and education effect disappears, and child nutrition is a combination of all the identified and non-identified covariates related to child nutrition and time available for childcare (𝐹𝑗𝑘𝑙 is indicating maternal education): 𝐷𝑂𝑇 = 𝑌𝑗𝑘𝑙 + 𝐷𝑗𝑘𝑙 + 𝐽𝑗𝑘𝑙(𝐹𝑗𝑘𝑙/𝐽𝑗𝑘𝑙 ∗ 𝐹𝑗𝑘𝑙) + 𝑣𝑗 2- For working mothers with lower education, the income effect can be wiped out by the lack

• f time available for childcare, as lower education is often associated to lower wages:

𝐷𝑂𝑇 = 𝑌𝑗𝑘𝑙 + 𝐹𝑗𝑘𝑙(𝐽𝑗𝑘𝑙/𝐹𝑗𝑘𝑙) − 𝐷𝑗𝑘𝑙 + 𝑣𝑗 3- For those with higher education, non-working mothers have the positive effect of more time available for childcare summed to the effect of better education. Better education is often associated to a better socioeconomic status, and healthier choices: 𝐷𝑂𝑇 = 𝑌𝑗𝑘𝑙 + 𝐷𝑗𝑘𝑙 + 𝐽𝑗𝑘𝑙(𝐹𝑗𝑘𝑙/𝐽𝑗𝑘𝑙) + 𝑣𝑗 4- Working mothers with higher education have healthier choices for the child that can extend to child care even when the mother is absent, added to the benefits of better wages, affecting child nutrition positively: 𝐷𝑂𝑇 = 𝑌𝑗𝑘𝑙 + 𝐽𝑗𝑘𝑙 ∗ 𝐹𝑗𝑘𝑙 − 𝐷𝑗𝑘𝑙 + 𝑣𝑗 India In India, rates of female education improved slowly from 1998-99 to 2005-06, and FLFP remained stagnant (Klasen and Pieters, 2015). Socioeconomic inequality has not decreased much across years, and it seems to be rooted on differences across urban and rural areas and cultural restraints to empower women. The assumptions on income and childcare are problematic, as a number of women are not paid in cash, and the positive effect of earnings from work in kind are not well

• measured. The time available for child care is also very difficult to be estimated, as a number of

women in rural areas work seasonally, but no details are provided about the hours and intensity involved in this work, and many parts of India follow a culture of extended patriarchal families. The literature shows that children in societies using this type of family structure are cared for multiple individuals, as adults can share responsibilities toward child care (Bordia Das & Zumbyte, 2016).

The hypotheses raised in this context need to consider the many differences of female work in rural and urban areas, affecting the type of payment and time invested at work; and the family structure that can affect the type of childcare received. In rural areas, hypotheses were raised according to who the mother works for, and whether the work was in agriculture or not. In urban areas hypotheses were raised on women’s occupation. In rural areas of India, where the majority of families work in agriculture, it is common to work for the family with no wages associated to it (Desai and Jan, 1994). Still, the fact of belonging to a joint family, and working for the family is often associated with greater economic status, since pooling income from men provides more assets and better living conditions for the household (Niranjan et al., 2005). Oppositely, paid employment overepresent poorer women and those belonging to scheduled castes (Srivastava and Srivastava, 2010, Desai and Jain, 1994). Self-employed agriculture is composed by farmers, what suggests a higher status. Then, hypotheses here suggest that: 1- Women working for the family, living in a joint family have the benefit of a pooling income

• r production at the household, as well as people to help caring for their child, what can

affect child nutrition in a more positive way. For nuclear families, working for the family suggests that the couple have members of the family around, having a less extent of the benefits seen in joint family. 2- Mothers working for others, and living in nuclear families do not have enough earnings to compensate the absence of the mother as a carer, affecting child nutrition negatively. For joint families, this indicates a lower socioeconomic status, but the shared responsibility for the child implies in less negative effect for the child. 3- Self-employed mothers have higher status and flexibility, what affects child nutrition positively. Although the majority of the working population in rural areas are inserted in the agriculture sector, jobs in other sectors are very heterogeneous, and tend to require better training and education. For instance, there are jobs within manufacturing, which requires informal training, teachers in lower grades, paramedics and public administration; and there are those involved in unpaid family work (Nceus, 2012). Comparing agriculture workers with non-agricultural workers in rural areas, the following hypotheses are raised: 1- Non-agricultural workers have higher education/training and more regular earnings, suggesting a positive effect of income that overcomes the lack of time available for child care.

2- The positive income effect is very low in agriculture sector compared to non-agriculture sector, and then, the effect of maternal work is negative for nuclear families and minimal for joint families due to the shared responsibility over the child. Population studies from India show that urban workers are in general better off than rural workers irrespectively of gender (Srivastava & Srivastava, 2009). Estimates for urban areas also show expansion of education for females, decline in education gender gap, and increase in labour market returns to education in the past two decades (Pieters, 2010). Highly educated women had more

• pportunities in white collar jobs, such as public administration, finance and business, and also in

education and health sector, while lower education provided jobs as manufacturer, construction, utilities, trade, hotels and restaurants (Klasen & Pieters, 2015). Therefore, the hypothesis in these areas is that better occupation brings a higher positive effect of income that offset the absence of the mother in the household. METHODS Data for this study was obtained from the two last rounds of the Demographic and Health Surveys (DHS) conducted in Brazil (1996 and 2006) and India (1998-99 and 2005-06). The DHS is a repeated cross-sectional survey conducted in various developing countries , aiming to evaluate population and maternal and child health through representative and standardized data (Rutstein and Rojas, 2006). Sampling design In general, the DHS follows a multistage design, with a two stage stratified sample. It uses an existing sampling frame, that is stratified in homogeneous groups, where a fixed number of households is selected by equal probability systematic sampling (ORC Macro, 1996, IFC, 2012). For both datasets the sampling covered all states of the country, considering differences between rural and urban

• areas. In Brazil, the partition between rural and urban areas was done during the strata division,

before the household selection (Estatística and Estatística, 2010). For India, where differences between rural and urban areas are more remarkable, the division was done in the first stage. After distinguishing between rural and urban, the samples were drawn separately for each area, and allocated proportionally to the size of the state’s urban and rural populations (IIPS, 2000, IIPS, 2007). Data collection The main instruments to collect data in both countries were the household and individual questionnaires by trained interviewers. The first provides a list of household members and basic demographic information about each member. This information is used for selecting women of

reproductive age for answering the individual questionnaire who also provide information about their young children (ORC Macro, 1996, Rutstein and Rojas, 2006). The DHS also collects data on anthropometric measures, such as weight and height of women, children and men. While in Brazil, the datasets included anthropometric measures for all children, in India, the 1998-99 round measured anthropometric data of children under three years of age of ever-married women who were interviewed with the Women’s Questionnaire. Therefore, children

• lder than 36 months, or those whose mothers were absent, dead, or under age 15 were not

measured (IIPS, 2000). Sampling weights are used to extrapolate the sample to the target population using design weights and sampling weights for both households and individuals (IFC, 2012). Variables Data on children who were not resident in the household were excluded as were children who had no anthropometric measures available, or missing information on maternal work. The outcome variable used in this study was the height-for-age z-score (HAZ) calculated based on the WHO Child Growth Standards (de Onis et al., 2007). Children with HAZ below -2 were categorized as stunted. The main independent variable was maternal work. This was a binary variable where mothers answered whether they had done any work in the last seven days that was paid in cash or kind. For those answering yes, data on the characteristics of work were collected, including question about the occupation, whether the woman worked from home or away from home and the type of earnings as cash or kind. To ensure comparable results between Brazil and India, controlling variables for statistical analysis were chosen aiming consistency over all datasets where possible. Variables were chosen based on what is reported in the literature to be associated to child nutritional status (CNS), and presented in the majority of datasets. Those were grouped in four main groups:

• 1. Child Characteristics: sex and age of the child; birth order; reported birth size; reported

recent disease (diarrhoea or fever);

• 2. Maternal characteristics: age; marital status; BMI; work; education;
• 3. Socio-economic/household characteristics: wealth index of the household; number of de

jure residents; partner’s education; religion; area of residence (rural/urban)

The wealth index was created for each dataset based on household ownership of durable goods and dwelling characteristics, types of water access, sanitation facilities, and cooking fuel. Durable goods included information on whether the household had electricity, television, telephone, refrigerator, vehicles, and whether the family owned the dwelling. Dwelling characteristics included information

• n type of material used for construction of floors, walls, roof and cooking stove. The wealth index

was generated from principal component analysis, taking into account differences between rural and urban areas (Rutstein and Johnson, 2004). Households were split into wealth quintiles, varying from lowest values (poorest) to highest values (richest). Education was created based on the highest level of education. It was categorized as no education, primary, secondary, higher secondary and tertiary education, following the categorization used in previous study in India using the NFHS (Subramanyam et al., 2011). Data analysis Random-intercept models were used to evaluate the association between maternal work and child nutrition considering the structure of the data in which children are nested in mothers, who are nested in regions. For Brazil, the response variable was the continuous variable HAZ. For India, due to the high prevalence of stunting and skewed distribution of HAZ, the variable was categorized as stunted or non-stunted. The models were performed in four steps. The first model consisted on univariate model, including only the working-related variable and error components. The second model added child characteristics, followed by the third model added of maternal characteristics and the fourth/full model added of household characteristics. RESULTS Descriptive statistics and the number of individuals participating in this study are presented in Table

• 1. The percentage of working mothers in Brazil was higher than in India, with significant increase

between the two waves, while in India, the low percentage of mothers inserted in the labour force remained the same. Both countries had opposite figures regarding maternal education and percentage of people living in rural areas. Brazil had a more educated population, where the percentage of mothers with higher secondary education or more doubled in 10 years. In India, there was a small progress in education, but almost half of the mothers were still not educated, and the majority of people lived in rural areas. Despite the significant decrease in the prevalence of child stunting, India still had almost half children in this condition. The results of the univariate analysis for the first hypothesis in Brazil (Table 2) confirmed that for 1996, the positive impact of income was null for self-employed mothers working away from home,

and no differences on HAZ were found when comparing to non-working mothers. For self-employed working from home, the impact was positive and significant, but it was higher for mothers who were

• employed. In 2006, self-employed mothers working away had a negative impact on child nutrition,

which was no longer significant when covariates were added.

Table 1. Sociodemographic characteristics of children, mothers and household in Brazil and India

Table 2. Results from multilevel linear modelling on type of work (employed/self-employed) and child’s HAZ

in Brazil, 1996 and 2006

* <0.05; **<0.01; ***<0.001 Full model – Model added of child, maternal and household characteristics

Variables Brazil 1996 Brazil 2006 India 1998-99 India 2005-06 N children 3833 4104 23077 23297 Mean HAZ

• 0.54
• 0.19
• 1.98
• 1.59

% stunting 13.4 5.9 53.1 46.4 N mothers 3050 3427 21214 21278 % working mothers 39.9 42.9 29.9 27.0 Maternal education No education 6.7 2.6 52.0 47.7 Primary 39.5 17.2 15.6 13.7 Secondary 31.2 34.5 23.8 27.7 Higher secondary or more 22.5 45.6 8.5 10.9 Wealth quintiles Q1 30.5 21.9 19.4 20.4 Q2 22.3 22.1 22.6 20.7 Q3 18.7 21.0 20.5 20.2 Q4 14.9 21.4 20.1 20.1 Q5 13.6 13.6 17.4 18.7 Area Rural 24.6 18.5 76.5 75.1 Urban 75.5 81.5 23.5 24.9

1996 2006 Univariate Full model Univariate Full model Intercept

• 0.49
• 0.70
• 0.29
• 0.18

Self-employed/away

• 0.05

0.10

• 0.12*
• 0.07

Self-employed/home 0.18* 0.11 0.14* 0.12 Employed 0.23*** 0.10 0.23*** 0.06 Variance state 0.30 0.13 0.23 0.12 Variance mother 0.80 0.66 0.60 0.53 Variance children 1.15 1.12 1.08 1.05 Log likelihood

• 6702
• 5579
• 6678
• 6203
• 2LogLikelihood 2

267.7*** 73.4*** 151.8*** 45.09***

The next model, interacting working and education (Table 3) showed that better education affected child nutrition positively regardless of the work status of the mother, but when work interacted with education, the impact on child linear growth was significantly higher, even when controlled for socioeconomic status. This observed effect of working was higher in 1996 than in 2006.

Table 3- Results for multilevel linear regression model on the effect of the interaction term among work and education on child’s HAZ in Brazil 1996 and 2006.

* <0.05; **<0.01; ***<0.001 Full model – Model added of child, maternal and household characteristics

For India, analysis in rural areas showed that being employed increased the likelihood of having a stunted child, with a greater effect for those working for others. These results only remained significant in full models for joint families in 1998-99. Being a self-employed mother only affected the likelihood of having a stunted child for nuclear families in the most recent data. Similar results were observed for those working in agriculture, where the likelihood of having a stunted child was higher only in univariate models, but non-significant when controlled for covariates (Table 4). In urban areas, all types of work, but professional work, increased the likelihood of having a stunted

• child. Working in the service sector was associated with child stunting only for nuclear family,

although it was no longer significant when added of covariates. Skilled work was also associated to child stunting, only for the most recent data. Being a professional decreased the likelihood of child stunting, what seemed to have a major effect on nuclear families.

1996 2006 Univariate Full model Univariate Full model Intercept

• 0.78
• 0.69
• 0.60
• 0.22

Working characteristics W/LE 0.04 0.05 0.09 0.11 NW/HE 0.51*** 0.15* 0.45*** 0.26*** W/HE 0.77*** 0.34** 0.56*** 0.36*** Where Away

• 0.08
• 0.03
• 0.12
• 0.12

Variance state 0.26 0.13 0.21 0.12 Variance mother 0.75 0.66 0.57 0.53 Variance children 1.15 1.12 1.08 1.05 Log likelihood

• 6567
• 5579
• 6651
• 6213
• 2LogLikelihood 2

212.5*** 73.6*** 122.3*** 45.7***

Table 4. Logistic multilevel modelling on the impact of different work characteristics on stunting in rural India.

1998-99 2005-06 Nuclear Joint Nuclear Joint Univariate Full model Univariate Full model Univariate Full model Univariate Full model Employed family 1.22** 0.96 1.27** 0.93 1.31*** 1.06 1.54*** 1.14 Employed others 1.36 1.11 1.74*** 1.31* 1.47*** 1.14 1.62*** 1.28 Self-employed 1.1 0.97 1.03 0.74 1.36* 1.2 0.95 0.82 Variance state 0.13 0.18 0.21 0.09 0.14 0.09 0.12 0.14 Variance mother 0.6 1.18 0.31 1.1 0.62 1.22 0.85 1.36 Agriculture 1.30*** 0.95 1.46*** 1.02 1.51*** 1.16' 1.60*** 1.22’ Non agriculture 1.09 1 1.16 0.99 1.1 1.01 1.13 1.15 Variance state 0.32 0.30 0.21 0.09 0.14 0.09 0.12 0.15 Variance mother 0.59 1.07 0.31 1.1 0.63 1.29 0.84 1.48 ‘ <0.10; * <0.05; **<0.01; ***<0.001 Full model – Model added of child, maternal and household characteristics

Table 5. Logistic multilevel modelling on the impact of different work characteristics on stunting in urban India.

1998-99 2005-06 Nuclear Joint Nuclear Joint Univariate Full model Univariate Full model Univariate Full model Univariate Full model Services 1.67** 1.01 1.82’ 1.39 1.75*** 1.26 1.45 1.43 Skilled 1.26 0.92 1.33 1.06 1.46* 1.10 1.89* 1.18 Professionals 0.43*** 0.78 0.55* 1.27 0.32*** 0.55* 0.48** 0.91 Variance state 0.13 0.09 0.10 0.10 0.12 0.11 0.10 0.15 Variance mother 0.14 0.32 1.49 2.31 1.04 1.48 1.57 2.09 ‘ <0.10; * <0.05; **<0.01; ***<0.001 Full model – Model added of child, maternal and household characteristics

DISCUSSION This study investigated how different characteristics of maternal work were associated with child stunting in Brazil and India, considering important aspects of female work in each country. Findings from both countries went on opposite directions, what was expected considering cultural discrepancy and differences in the stage of social development in each country. In Brazil, maternal work had positive effects on child linear growth overall, in agreement with the hypothesis raised. But, significance was lost particularly when maternal education was added to the

• models. The interaction between education and work remained significant after controlling for all

covariates, suggesting that jobs requiring more years of study or training and providing higher wages

have a positive trade-off for the child, while in other types of work, child height-for-age is more effectively influenced by the context in which the child is raised. In India, maternal work was overall associated with higher likelihood of having a stunted child. Univariate results confirmed the hypothesis raised previously that being employed for others was detrimental for child nutritional status in rural areas, and working in agriculture had worse effects than working in other sectors. But, most results were no longer significant when added of covariates. Other studies also found the work effect to be lost when added of maternal and household covariates (Ukwuani and Suchindran, 2003, Bernal, 2008). First of all, education is important for child nutrition and it has a direct impact on choices and behaviours of the mother regarding child care which reflects on better child nutrition. Although it is expected a gradient of education among these types of work, working variables might not be capturing well behavioural choices from the mother as much as maternal education does. Moreover, education, wealth and work are three variables interacting and affecting each other, and then, it is expected that when adding such variables that explain better behaviours and environment characteristics, the type of work has no effect on child

• nutrition. Having socioeconomic status and maternal education as confounders is particularly

important for India, where it is widely cited in the literature that working women are often found in lower castes and poorer conditions than other women, seen the cultural constraints regarding female work (Klasen and Pieters, 2015, Bhalla and Kaur, 2011, Lahoti and Swaminathan, 2013). Some differences between results in Brazil and India are interesting to be noted. Results from Brazil suggested that being a self-employed mother working from home had positive effects for the child, while working away from home had negative effects. This goes in agreement with the history of female employment in Brazil. The slow progress of Brazil towards women’s participation in the formal labour force encouraged women to be included in the informal sector for many years, mainly working as housemaids and sales. FLFP increased in Brazil, but self-employment is still higher for women than men, and there are still high rates of women working in low paid sectors (Leone and Baltar, 2008). Oppositely, in India, being self-employed in rural areas had low or null effect on the likelihood of having a stunted child. This could be explained by the high status of having the own land and being able to insert products in the market (Srivastava, 2009), what increases earnings from work and improves household socioeconomic status. These differences in Brazil and India show how the context is important when exploring associations between female work and child’s health.

The models performed in this study intended to consider the importance of time available for child care using proxies of this measure. In Brazil, a binary variable of whether the mother worked at home or away from home was included in the models, but none was significant. This is difficult to interpret with no measures of how long the mother spends at home with the child if working or not. Working at home not necessarily means the mother is able to spend time with the child and to provide childcare (Glick & Sahn, 1998), and women working away from home can have part-time jobs, and have more time with the child. Studies from high income countries have suggested that time available for child care might not be as important as researchers try to acknowledge. Instead, maternal education is what determines the quality of mother’s activities with the child (Bianchi and Robinson, 1997, Behrman and Rosenzweig, 1999). Because relatively little is known about women’s non-market activities, time spent with the child when mothers are not working is mostly overestimated (Bianchi, 2000), leaving aside two key points about time of the mother for child care: 1) At the same time opportunities for female work increased, home appliances helped to decrease maternal use of time on domestic activities, therefore, the impact of work on time available for the child care might be exaggerated in previous studies when mothers had to spend more time on household activities; 2) Nowadays children tend to be at school in early ages irrespectively of whether the mother works or not. Studies have also been reporting that father’s time with children increases in response of their partner’s hours at work, possibly offsetting the lack of child care from the mother (Baxter, 2007). In India, the context is totally different, where it is not common to have pre-scholars attending schools and child centres are almost non-existent (Palriwala and Neetha, 2011). This study tried to capture differences of childcare using family structure as a proxy. No conclusive differences could be

• bserved among nuclear or joint families, but the Odds ratios were slightly higher for joint families

when the type of employment was related to a lower socioeconomic status. For instance, in disagreement with the hypothesis suggesting that the effect of being employed by others would be more detrimental for nuclear families, the results showed a higher effect size for joint families, which remained significant after controlling for covariates in 1998-99. This might be indicating that joint families in poorer conditions are in disadvantage when comparing with nuclear families, having more difficulties for providing for all members of the family. Another explanation found in the literature is about the status of women as daughter-in-law in a patriarchal family, claiming that those receive less investment in health during pregnancy, they are engaged in more hours of work including onerous task, and receive less quality-food (Barua, 2004; Chorgade et al, 2006; Das Gupta, 1996). This could affect the child indirectly through maternal health, and directly by applying the

same neglect to the child. Further analyses are needed to explore these associations controlling for confounding factors. In urban areas, being a professional was the only category of work with lower chances of having a stunted child. This association remained significant only for nuclear families in the most recent wave. This brings the discussion of whether the significance of this association in joint families is due to the effect of better socioeconomic status due to higher wages from maternal work, or simply because better-educated women belong to wealthier families. CONCLUSION This study comes with limitations. Although using two datasets from a different period of time, the data cannot show the dynamics of maternal work. It is possible that some women were not working in the first years of the child, but intended to go back to work later on. Longitudinal analysis would be best to describe work trajectories and evaluate maternal decisions on child nutrition. There is also a lack of information on whether the mother works part-time or full time, hours at work, and secondary employment -variables that are important for the hypotheses raised here. Still, this study has the strength of bringing a comparative work between two emerging countries with very different outcomes, shedding light on how the context affects the association between female work and child nutrition. This analysis concludes that the relationship between maternal work and child nutrition depends on the stage of a country’s development, types of work provided for women and policies protecting the child and the mother who works. Such characteristics have an impact on the trade-off between income and time availability, and when maternal work improves household environment, it tends to be more positive for the child. REFERENCES ABBI, R., CHRISTIAN, P., GUJRAL, S. & GOPALDAS, T. 1991. The impact of maternal work status on the nutrition and health status of children. Food and Nutrition Bulletin, 13, 20-25. AGENOR, P.-R. & CANUTO, O. 2015. Gender equality and economic growth in Brazil: a long-run

• analysis. Journal of Macroeconomics, 43, 155-172.

ANDERSON, P. M., BUTCHER, K. F. & LEVINE, P. B. 2003. Maternal employment and overweight

• children. J Health Econ, 22, 477-504.

BAXTER, J. 2007. When dad works long hours: How work hours are associated with fathering 4-5- year-old children. Family Matters, 60. BECKER, G. S. 1965. A Theory of the Allocation of Time. The economic journal, 493-517. BEHRMAN, J. R. 1990. The action of human resources and poverty on one another: what we have yet to learn. Living standards measurement study working paper. BEHRMAN, J. R. & ROSENZWEIG, M. R. 1999. “Ability” biases in schooling returns and twins: a test and new estimates. Economics of Education Review, 18, 159-167. BERNAL, R. 2008. The effect of maternal employment and child care on children's cognitive

• development. International Economic Review, 49, 1173-1209.

BHALLA, S. & KAUR, R. 2011. Labour force participation of women in India: some facts, some queries. LSE Asia Research Centre Working Paper. BIANCHI, S. M. 2000. Maternal employment and time with children: Dramatic change or surprising continuity? Demography, 37, 401-414. BIANCHI, S. M. & CASPER, L. M. 2000. American families, Population Reference Bureau. BIANCHI, S. M. & ROBINSON, J. 1997. What did you do today? Children's use of time, family composition, and the acquisition of social capital. Journal of Marriage and the Family, 332-344. BLACK, R. E., VICTORA, C. G., WALKER, S. P., BHUTTA, Z. A., CHRISTIAN, P., DE ONIS, M., EZZATI, M., GRANTHAM-MCGREGOR, S., KATZ, J., MARTORELL, R. & UAUY, R. 2013. Series: Maternal and child undernutrition and overweight in low-income and middle-income countries. The Lancet, 382, 427- 451. BROWN, J. E., BROOM, D. H., NICHOLSON, J. M. & BITTMAN, M. 2010. Do working mothers raise couch potato kids? Maternal employment and children's lifestyle behaviours and weight in early

• childhood. Soc Sci Med, 70, 1816-24.

DE ONIS, M., ONYANGO, A. W., BORGHI, E., SIYAM, A., NISHIDA, C. & SIEKMANN, J. 2007. Development of a WHO growth reference for school-aged children and adolescents. Bulletin of the World Health Organization, 85, 660-667. DESAI, S. & JAIN, D. 1994. Maternal employment and changes in family dynamics: The social context

• f women's work in rural South India. Population and Development Review, 115-136.

ENGLE, P. L. 1991. Maternal work and child-care strategies in peri-urban Guatemala: nutritional

• effects. Child Dev, 62, 954-65.

ENGLE, P. L. 1993. Influences of mothers' and fathers' income on children's nutritional status in

• Guatemala. Social Science & Medicine, 37, 1303-1312.

ENGLE, P. L. & PEDERSEN, M. E. 1989. Maternal work for earnings and childrens nutritional status in urban Guatemala. Ecology of food and nutrition, 22, 211-223.

ESTATÍSTICA, I. B. D. G. E. & ESTATÍSTICA, I. B. D. G. E. 2010. Pesquisa Nacional por Amostra de Domicílios. Um panorama da saúde no Brasil: acesso e utilização dos serviços, condições de saúde e fatores de risco e proteção à saúde 2008. IBGE Rio de Janeiro. GLICK, P. & SAHN, D. E. 1998. Maternal labour supply and child nutrition in West Africa. Oxf Bull Econ Stat, 60, 325-55. HAWKES, C. & POPKIN, B. M. 2015. Can the sustainable development goals reduce the burden of nutrition-related non-communicable diseases without truly addressing major food system reforms? BMC medicine, 13, 143. HUNDLEY, G. 2000. Male/female earnings differences in self-employment: The effects of marriage, children, and the household division of labor. Industrial & Labor Relations Review, 54, 95-114. IFC, I. 2012. Demographic and Health Survey Sampling and Household Listing Manual. MEASURE DHS. IIPS 2000. National Family Health Survey (NFHS-2), 1998-99. India. Mumbai. IIPS 2007. National Family Health Survey (NFHS-3), 2005–06: India: Volume I. JOHNSON, F. & ROGERS, B. Nutritional status in female-headed households in the Dominican

• Republic. International Conference on Women, Development, and Health, Michigan State University,

1988. KLASEN, S. & PIETERS, J. 2015. What explains the stagnation of female labor force participation in urban India? The World Bank Economic Review, lhv003. LAHOTI, R. & SWAMINATHAN, H. 2013. Economic growth and female labour force participation in

• India. IIM Bangalore Research Paper, 414.

LAMONTAGNE, J. F., ENGLE, P. L. & ZEITLIN, M. F. 1998. Maternal employment, child care, and nutritional status of 12-18-month-old children in Managua, Nicaragua. Soc Sci Med, 46, 403-14. LEONE, E. T. & BALTAR, P. 2008. A mulher na recuperação recente do mercado de trabalho brasileiro. Revista Brasileira de Estudos de População. LESLIE, J. 1988. Women's work and child nutrition in the Third World. World Development, 16, 1341- 1362. NCEUS, N. 2012. Report on Conditions of Work and Promotion of Livelihoods in the Unorganised Sector. NIRANJAN, S., NAIR, S. & ROY, T. 2005. A socio-demographic analysis of the size and structure of the family in India. Journal of Comparative Family Studies, 623-651. O'NEILL, J. 2001. Building better global economic BRICs. ORC MACRO 1996. Sampling manual. DHS-III Basic Documentation. PALRIWALA, R. & NEETHA, N. 2011. Stratified familialism: the care regime in India through the lens

• f childcare. Development and Change, 42, 1049-1078.

PIETERS, J. 2010. Growth and inequality in India: analysis of an extended social accounting matrix. World Development, 38, 270-281. POPKIN, B. & SOLON, F. S. 1976. Income time the working mother and child nutriture. Journal of Tropical Pediatrics and Environmental Child Health, 22, 156-66. RUTSTEIN, S. O. & JOHNSON, K. 2004. The DHS wealth index. DHS comparative reports no. 6. Calverton: ORC Macro. RUTSTEIN, S. O. & ROJAS, G. 2006. Guide to DHS statistics. Calverton, MD: ORC Macro. SRIVASTAVA, N. & SRIVASTAVA, R. 2010. Women, work, and employment outcomes in rural India. Economic and political weekly, 49-63. SUBRAMANYAM, M. A., KAWACHI, I., BERKMAN, L. F. & SUBRAMANIAN, S. V. 2011. Is economic growth associated with reduction in child undernutrition in India? Plos Medicine, 8, e1000424- e1000424. TOYAMA, N., WAKAI, S., NAKAMURA, Y. & ARIFIN, A. 2001. Mother's working status and nutritional status of children under the age of 5 in urban low-income community, Surabaya, Indonesia. J Trop Pediatr, 47, 179-81. TUCKER, K. 1989. Maternal employment differentiation and child health and nutrition in Panama. UKWUANI, F. A. & SUCHINDRAN, C. M. 2003. Implications of women's work for child nutritional status in sub-Saharan Africa: a case study of Nigeria. Social Science & Medicine, 56, 2109-2121. VAN KLAVEREN, M., TIJDENS, K., HUGHIE-WILLIAMS, M. & MARTIN, N. R. 2009. An Overview of Women’s Work and Employment in Brazil - Decisions for Life MDG3 Project, Country Report No. 12. Amsterdam, Netherlands: Amsterdam Institute for Advanced Labour Studies. WAJNMAN, S., QUEIRÓZ, B. & LIBERATO, V. 1998. O crescimento da atividade feminina nos anos noventa no Brasil. Anais do XI Encontro Nacional de Estudos Populacionais. Abep, 2, 2.429-2.454. WELLINGTON, A. J. 2006. Self-employment: the new solution for balancing family and career? Labour Economics, 13, 357-386.