Womens Labor Force Participation and Length of Breastfeeding in Four - - PDF document

1

Women’s Labor Force Participation and Length of Breastfeeding in Four African Countries Paper prepared for presentation at the IUSSP Meetings, Cape Town, South Africa October 31, 2017 Elizabeth Heger Boyle, JD, PhD, boyle014@umn.edu Greta Gangestad, gange026@umn.edu Miriam L. King, PhD, kingx025@umn.edu Sula Sarkar, PhD, sanb0027@umn.edu Minnesota Population Center University of Minnesota, Minneapolis, USA While increases in women's labor force participation might be expected to reduce breastfeeding, prior research results for low and middle income countries are mixed. This study advances knowledge in this area by linking data on women’s employment from IPUMS-International (Minnesota Population Center 2017) with data on breastfeeding from IPUMS-DHS (Boyle, King, and Sobek 2017). We utilize a multilevel ordered logit of data on children drawn from four African DHS: in Kenya, Malawi, Morocco, and Zambia. We test the effect not only of mother's own employment but also the level of female employment overall and in agriculture at the second administrative level. We find that women’s individual employment has no effect on length of breastfeeding, but greater levels of female employment are associated with increases in the length

• f breastfeeding within regions.

Introduction The World Health Organization (2001) recommends, for the health of infants, exclusive breastfeeding in the first six months of life, and continued breastfeeding to two or more years of

• age. Studies in wealthy countries, such as the United States, Australia, and Singapore, have shown

that female employment can be associated with a decision to not take up breastfeeding or make the period of breastfeeding shorter (Ryan, Zhou, and Arensberg 2006; Scott 2006; Sivakami 2003). We might expect a very different pattern in Africa, however, because prolonged breastfeeding is

2

normative in most parts of the continent (Østergaard and Bula 2010). Prior research results for low- income countries are mixed, with earlier cessation of breastfeeding found for working women in some studies (Pakistan, Nair, Ariana, and Webster 2014), for non-working women in other studies (Bangladesh, Akter and Rahman 2009), and varying by employment sector in still other studies (DeRose 2007). Broad regional differences in breastfeeding practices in Malawi further complicate the picture, suggesting the importance of contextual influences (Kazembe 2008). Hypotheses We hypothesize that women will breastfeed their infants longer in districts where female employment is widespread. Given the normative context of breastfeeding in the countries in our analysis, more women in the labor market will force employers to accommodate childcare

• demands. We also hypothesize that the contextual effect of women's work may be different when

considering levels of female employment in agriculture. We further hypothesize that the effect of an individual woman's work will vary according to her household's resources, following Helen Ware's logic that "one would wish to divide mothers who work outside the home into two groups: those who work because of the driving pressure of poverty and those whose work is more a source

• f interest and a higher standard of living."

Data and Methods The Demographic and Health Surveys have been an invaluable source of information about the health of women of childbearing age and young children in low- and middle-income countries

• ver the past thirty years, but the relatively small samples for these surveys has limited their use

for constructing contextual variables. DHS geographic regions are very large (at the first administrative level or larger), while sample clusters incorporate too few women to provide reliable indicators of local population characteristics. Furthermore, the DHS include limited, if

3

any, information on important community characteristics, such as the percentage of migrants or features of the labor market. We overcome these limitations by linking census data to the DHS, to examine contextual as well as individual effects of women's work on the duration of breastfeeding. Note that IPUMS-DHS (Boyle, King & Sobek 2017) will soon include these and other contextual variables on the project website so that they can be easily included in the harmonized, integrated data files that researchers create. Data Linking Strategy Here, we combined individual-level data from IPUMS-DHS, GPS data for the DHS samples (i.e., information on the latitude and longitude of each sample cluster), and summary data for the second-level administrative districts where DHS clusters are located, based on tabulations

• f very large census microdata samples from IPUMS-International (Minnesota Population Center

2017). We 1) mapped DHS clusters (using the GPS coordinates) onto the census second-level administrative districts, 2) associated each GPS cluster with the corresponding census-based value in the surrounding second-level administrative unit, and 3) linked this contextual information to individual children in the DHS by merging based on cluster number. Figure 1 illustrates our linking

• strategy. Figure 1 shows the percentage of women age 15-59 who are employed within census
• regions. Each dot represents a DHS cluster. Each child in our sample is assigned the value

associated with the census region that contains his or her cluster. Sample We paired census and survey data for the four DHS surveys: Kenya 2008-9, Malawi 2010, Morocco 2003, and Zambia 2013. In each of these countries, a DHS survey was fielded within 1 to 3 years of a national census, and the census was available through IPUMS-International. Our final sample includes 24,729 children, aged 2-4, nested within 59 regions. Variables

4

Breastfeeding duration. Due to digit preference at 6 month intervals, we recoded breastfeeding duration, our dependent variable, into 5 categories (0 months, 1-11 months, 12-17 months, 18-23 months, and 24+ months). To avoid right censoring of the length of breastfeeding, we include only children ages 2 to 4 in our sample. Contextual variables. Using the censuses, we calculated the percent of women 15-59 employed and the percent of women employed in agriculture in xx administrative two regions. Figure 1 shows the level of female employment in each of these regions by category (the contextual variables themselves are continuous). In most regions in Morocco, less than 25% of women are participating in the labor market. Female labor force participation is more common in the other three countries. Other independent variables. Our statistical models include the following additional independent variables child's sex (male 0/female 1), child's birth order (topcoded at 8+), mother's level of education, whether mother is currently in union, household wealth (middle category excluded), urban residence (no 0/yes 1), and mother's current employment (no 0/yes 1). Statistical Model We used a multilevel ordered logit models for the duration of breastfeeding for surviving children age 2 to 4, nested with DHS subnational regions. We include a random intercept and a random coefficient for percent of women working, which substantially improved the models’ fit. Results: Effect of Women's Labor Force Participation on the Length of Breastfeeding Table 1 presents the results of our multivariate analysis. Model 1 includes only individual- level variables from the DHS unrelated to employment. Male children are breastfed longer than female, which suggests that mothers have a sex preference. Later-born children are breastfed longer than children early in the birth order. This likely because women with more children are interested in deferring their next pregnancy. Children in the wealthiest households (compared to

5

• thers in their country) and in urban areas are breastfed for shorter periods than children from the

middle wealth quintile and rural areas. Model 2 adds mother’s employment status, which has no statistically significant effect on length of breastfeeding. Models 3 and 4 consider the contextual factors. We find that, all else being equal, the more women working in a region, the longer women in that region are breastfeeding. We also discovered an interaction effect: the effect of more women working is diminished in regions where a greater percentage of women are working in agriculture. To test the robustness of our results, we ran several modified models. The increase in health benefits from breastfeeding is likely smaller between 18 and 24 months than it is between 6 and 12 months. We therefore tested models that combined the two longest categories of breastfeeding, making the breastfeeding variable four rather than five categories. The odds were not notably

• different. The direction of relationships did not change nor did any relationship lose or gain

statistical significance. We also ran the models that dropped Morocco from the analysis, reasoning that north African countries may represent a population different from sub-Saharan African

• countries. This did change the models some, so we are currently assessing the relative fit to

determine the most appropriate approach. Our analysis of four DHS surveys and associated censuses demonstrates the feasibility and usefulness of combining individual-level DHS data with contextual data for census regions. In our multi-country sample, a mother's own employment did not appear significant, but contextual data illuminated the potential (and positive) effect of female employment on the length of breastfeeding at the regional level. Next steps

Breastfeeding for two years is only one WHO guideline. WHO also recommends taking up breastfeeding within the first hour of life and exclusive breastfeeding for the first six months of life. We are

6 currently investigating whether there are proxies for additional dimension of healthy breastfeeding in the DHS data. These other dimensions may show different associations with women working and the other independent variables. For example, Webb-Girard et al. (2012) found that food insecurity may prompt resistance to exclusive breastfeeding in Kenya. If broadly true, this would suggest a stronger effect for household wealth than we are seeing with our length-of-breastfeeding models. Family structure, such as polygamy, may influence both female work and breastfeeding behavior (compare Omariba and Boyle 2007). We are thus also exploring whether we can capture such differences in our analyses. Finally, breastfeeding is only one dimension of child nutrition. We hope to extend our analyses to other aspects of nutrition that affect child health.

Conclusions Our first set of substantive analyses showed that, along with sex, birth order, household wealth, and household location (urban or rural), higher prevalence of female employment was associated with longer breastfeeding, even when individual mother's employment had no significant effect. We conclude that, within the four countries studied, when female employment is more normative, infant care and breastfeeding are more integrated into work environments. Women working in agriculture, which may signal greater levels of poverty, appeared to suppress the otherwise positive effect of more women working. Our analysis was also motivated not only by the important topic of breastfeeding but also by the desire to test the feasibility of adding census-based contextual variables to DHS microdata, using GPS coordinates to locate sample clusters in second administrative level districts. The test was successful. Over the next year, we intend to construct census-based contextual variables on

• ther topics (including education, gender inequities, work, media exposure, and household

utilities), add them to more DHS samples, and disseminate these contextual variables through the free online IPUMS-DHS database (idhsdata.org). The authors of this paper work on the IPUMS-

7

International census and IPUMS-DHS data integration projects, and thus are well-placed to facilitate such analyses for the broad research community. References Akter, S. and M. M. Rahman. 2009. “The Determinants of Early Cessation of Breastfeeding in Bangladesh.” World Health & Population 11(4):5–12. Boyle, Elizabeth Heger, Miriam L. King, and Matthew Sobek. 2017. “IPUMS-DHS.” Retrieved April 15, 2016 (https://www.idhsdata.org/idhs/). DeRose, Laurie F. 2007. “Women’s Work and Breastfeeding Simultaneously Rise in Ghana.” Economic Development and Cultural Change 55(3):583–612. Kazembe, L. N. 2008. “Spatial Modelling of Initiation and Duration of Breastfeeding: Analysis

• f Breastfeeding Behaviour in Malawi-I.” World Health and Population 10(3):14–31.

Minnesota Population Center. 2017. IPUMS-International. https://international.ipums.org/. Retrieved (https://international.ipums.org/). Nair, Manisha, Proochista Ariana, and Premila Webster. 2014. “Impact of Mothers’ Employment

• n Infant Feeding and Care: A Qualitative Study of the Experiences of Mothers

Employed through the Mahatma Gandhi National Rural Employment Guarantee Act.” BMJ Open 4(4):e004434. Omariba, D. and Michael H. Boyle. 2007. “Family Structure and Child Mortality in Sub-Saharan Africa: Cross-National Effects of Polygyny.” Journal of Marriage and Family 69(2):528–543. Østergaard, Lise Rosendal and Agatha Bula. 2010. “‘They Call Our Children „Nevirapine Babies‟’: A Qualitative Study about Exclusive Breastfeeding among HIV Positive Mothers in Malawi.” African Journal of Reproductive Health 14(3):213–222. Ryan, Alan S., Wenjun Zhou, and Mary Beth Arensberg. 2006. “The Effect of Employment Status on Breastfeeding in the United States.” Women’s Health Issues 16(5):243–51. Scott, J. A. 2006. “Predictors of Breastfeeding Duration: Evidence From a Cohort Study.” PEDIATRICS 117(4):e646–55. Sivakami, M. 2003. “The Impact of Maternal Work Participation on Duration of Breastfeeding among Poor Women in South India.” Asia-Pacific Population Journal 18(3):69–90. Webb-Girard, Aimee et al. 2012. “Food Insecurity Is Associated with Attitudes towards Exclusive Breastfeeding among Women in Urban Kenya: Food Insecurity and Exclusive Breastfeeding.” Maternal & Child Nutrition 8(2):199–214.

8

World Health Organization. 2001. “Report of the Expert Consultation of the Optimal Duration of Exclusive Breastfeeding, Geneva, Switzerland, 28-30 March 2001.” Retrieved September 28, 2017 (http://www.who.int/entity/nutrition/publications/infantfeeding/optimal_duration_of_exc _bfeeding_report_eng.pdf). Ware, Helen. 1984. "Effects of Maternal Education, Women's Roles, and Child Care on Child Mortality," Population and Development Review, vol. 10: 191-214.

9

Table 1. Context of Women Working and Duration of Breastfeeding in Kenya, Malawi, Morocco, and Zambia (1) (2) (3) (4) VARIABLES Model 1 Model 2 Model 3 Model 4 Female child

• 0.080***
• 0.080***
• 0.080***
• 0.080***

(0.024) (0.024) (0.024) (0.024) Birth order of child 0.097*** 0.096*** 0.096*** 0.096*** (0.006) (0.006) (0.006) (0.006) Mother's education level

• 0.067**
• 0.069**
• 0.069**
• 0.069**

(0.021) (0.021) (0.021) (0.021) Mother in union

• 0.012
• 0.010
• 0.010
• 0.008

(0.036) (0.037) (0.037) (0.037) Poorest wealth quintile

• 0.023
• 0.024
• 0.024
• 0.022

(0.038) (0.038) (0.039) (0.036) Poorer wealth quintile 0.020 0.020 0.020 0.020 (0.037) (0.037) (0.037) (0.037) Richer wealth quintile 0.018 0.018 0.022 0.021 (0.040) (0.040) (0.040) (0.040) Richest wealth quintile

• 0.112*
• 0.113*
• 0.100*
• 0.107*

(0.047) (0.047) (0.048) (0.048) Urban

• 0.432***
• 0.431***
• 0.403***
• 0.400***

(0.038) (0.038) (0.039) (0.039) Mother working 0.003 0.003 0.003 (0.003) (0.003) (0.003) Percent of working women in agriculture (region) 0.221 (0.114) 0.704** (0.240) Percent of women working (region) 1.031* (0.427) 2.417** (0.740) Working women in agriculture*Women working (region)

• 1.576*

(0.689) Observations 24,729 24,729 24,729 24,729 Number of groups 59 59 59 59 Standard errors in parentheses; cut points not shown *** p<0.001, ** p<0.01, * p<0.05

10

Figure 1. Census regions showing ordinal categories of percent women working with associated DHS clusters.