Gender under Distress: Essays on Climatic and Familial Determinants of Gendered Outcomes in African Populations

Abstract

This dissertation consists of five essays on how gendered outcomes change when stressors put pressure on resources in Sub-Saharan Africa. Three essays consider how deviations in rainfall from what is normally observed in a location affect outcomes for girls and women. Rainfall strongly influences household income and local and national economic activity in Sub-Saharan Africa, as most households engage in agriculture which often lacks irrigation. As yearly variation in relative rainfall is independent of household characteristics, these papers use rainfall to uncover how changes in income affect gender differences in infant mortality and violence against women in Sub-Saharan Africa, and female-headed households in South Africa. Two essays consider stressors coming from within the family, namely failure to achieve a desired gender composition of offspring and restrict fertility to desired levels. Specifically, the articles consider how these stressors affect infant mortality and women’s perceptions of the ideal family size and gender composition. Written together with Andreas Kotsadam, the first paper uses data from Demographic and Health Surveys (DHS) on 1.5 million infants born 1980-2010 in 29 countries and finds that extreme droughts cause a gender difference in infant mortality with 12 more female than male infant deaths per 1000 live births. We argue that the results are too large to be explained by biological differences between the sexes, which is further supported by the finding that differences are larger in areas where the level of son preference is high and desired fertility rates are low. We also find that the effect is larger for nonworking women and in locations where female labour force participation is low, but we do not find different effects by level of education. The second essay is co-authored by Sara Cools and Andreas Kotsadam and uses variation in rainfall to study how income shocks affect intimate partner violence in Sub-Saharan Africa. We use DHS data with up to 149,000 women in 17 countries who where asked about whether they had experienced violence from their partners. The analysis proves the presence of spatial autocorrelation in weather shocks when using cross-sectional data, and corrects for this. We find no evidence supporting our theory that unusually low levels of rainfall would increase intimate partner violence. This holds both for the probability of experiencing violence in a given year, and for the risk of experiencing violence for the first time during the course of a marriage. We speculate that the collective nature and slow onset of droughts trigger less aggression and more adherence to social norms than other income shocks that might increase violence against women in Sub-Saharan Africa and elsewhere. In the third essay which is co-authored with Raya Muttarak and André Pelser, we show that an already marginalised group - female-headed households in South Africa - is differentially affected by relatively modest levels of variation in rainfall. Data from three waves of the National Income Dynamics Survey in South Africa allow us to follow incomes of 4,162 households from 2006-2012. We find that households where a single head can be identified based on residency or work status are more vulnerable to climate variability than households headed by two adults. Single male-headed households are more vulnerable because of lower initial earnings and, to a lesser extent, other household characteristics that contribute to economic disadvantages. However, this can only explain some of the differential vulnerability of female-headed households. This suggests that there are traits specific to female-headed households, such as limited access to protective social networks or other coping strategies, which makes this an important dimension of marginalisation. Households headed by widows, never-married women, and women with a non-resident spouse (e.g. ‘left-behind’ migrant households) are particularly vulnerable. The fourth essay studies how being born at a parity or of a sex which is undesired by the mother relates to infant mortality in Sub-Saharan Africa, and how the differential mortality attributable to these preferences changes across the demographic transition. Using data from 79 DHS surveys, I find that being undesired according to the mother is associated with a differential mortality which is not due to constant maternal factors, family composition, or factors that are correlated with maternal preferences and vary continuously across siblings. As a share of overall infant mortality, differentials according to undesiredness of children explain 3.3% of male and 4% of female infant mortality. Undesiredness can explain a larger share of infant mortality among mothers with lower fertility desires, and a larger share of female than male infant mortality for children of women whose fertility desire is between 1-3 children. Undesired gender composition is more important for infant mortality than undesired childbearing and may also lead couples to increase family size beyond the maternal desire, in which case infants of the surplus gender are particularly vulnerable. Data from the same surveys are used in the fifth essay which asks whether mothers adjust their fertility preferences when having more children. To that end, this paper uses giving birth to twins at first parity and the two firstborn children being of the same sex as natural experiments which increase the number of children ever born and the number of live children at interview. I find that having an additional child increases the ideal family size by 0.1 - 0.8 child and that the effects are much stronger for women who have fewer children than they desire than for women who have more children than they desire. These findings are not consistent with the prevailing theory that rationalisation (i.e. the upwards adjustment in preferences to disguise unwanted childbearing) explains changes in fertility preferences. Rather, I propose that a process of learning about costs and benefits of childbearing could lead to the continuous updating of fertility preferences that we observe. Furthermore, the gender of the first child is used to estimate that having a son or daughter leads to an increase of 0.1 child in the desire for children of that gender.