While growth has taken off in sub-Saharan Africa (SSA) in recent decades, the structural transformation of its economies is still lagging far behind that observed in other previously low-and-middle-income regions across the world. At the same time, SSA is undergoing a unique fertility transition: the decline in the number of children born per woman is decreasing slower than in any other region of the world, even conditional on income (Bongaarts 2017). This project proposes to study to what extent emerging employment opportunities in SSA, especially salaried positions for low-skilled women, may affect women’s fertility choices, thereby helping to speed up SSA’s fertility decline. We know very little about whether and how new types of jobs (e.g. through the modernization of agriculture) shift demand for female labour, and how this affects women’s occupations and fertility choices, especially in SSA contexts. This study proposes to fill this gap in the literature by exploiting the recent emergence of the cut-flower industry in Kenya — which employs 60-70% women (Mitullah, Kamau and Kivuva 2017), most of whom have little education, on salaried positions — since the 1990s as a natural experiment to study the effects of the expansion of wage employment opportunities for women on fertility.
Recent evidence from South Asia (Jensen 2012; Heath & Mobarak 2015) and historical evidence from Sweden (Schultz 1985) and the US (Ager et al. 2020) suggest that the expansion of female job opportunities – or an increase in the relative return to women’s labour – can reduce fertility. Evidence is altogether much scarcer in SSA, and this literature does not study the differential fertility impacts of women’s self- employment vs. wage/salaried work opportunities. Our hypothesis is that, in contexts where organised childcare is largely absent, jobs that are less compatible with looking after young children (here, wage jobs) will increase the time cost associated with children, leading to lower fertility.
To test this hypothesis, we propose to create a new database documenting the arrival and expansion of flower-processing plants in Kenya, in particular greenhouses growing cut-flowers for export, which we can match to detailed, geo-identified micro data: two rounds of multi-topic household surveys, three rounds of DHS surveys, and three decennial population censuses.
To analyse the effect of cut-flower export industry on wage employment on fertility, we need two pieces of detailed information: first, the exact time and location of new flower-processing plant arrival across Kenya to understand which villages and locations were “treated”. Second, “stock” and “flow” measures of fertility as well as employment outcomes in the vicinity of new flower-processing plants. Our first step is to construct a geographical census of all flower-processing plants in Kenya. In order to obtain the historic time of entry of each flower plant, we plan the following triangulation exercise: first, searching backwards through historic satellite images of the same location allows for a rough determination of when a given plant first appeared.
Second, a phone survey of all flower plants for which contact details can be found online will enable enumeration of information on the firm’s start of operations in a given location. Third, we plan on making use of customs records to understand when a given flower plant first appeared as exporter, as a reasonable proxy for the flower plant’s year of entry.
In step 2, the set of outcomes we will focus on is as follows: first, were women’s employment status and earnings indeed affected by the arrival of new flower processing plants (i.e. greenhouses) in the labour market? Did the share of women in salaried employment increase in affected locations, compared to “control” locations that would have been suitable for hosting a flower processing plant, but did not (yet) receive one? Did women subsequently change their fertility decisions, and are the women who do statistically more likely to have engaged in wage work? The publicly available micro data listed above will allow us to derive and harmonize outcome variables of interest. All three data sources include fertility questions allowing us to derive “stock” measures of fertility, and, for our main “flow” measure, we will leverage the detailed birth histories in the DHS to construct yearly birth rates at the ward (sublocation) level.
Since the timing of entry of flower-processing plants in a given location is non-random, and different geographical units get “treated” at different times, the empirical setting lends itself well to a staggered difference-in-differences design. The identifying assumption is that, in absence of flower plant entry in the treatment locations, the average outcomes for all groups (where each group gets treated at a different time) would have evolved in parallel. This assumption is unlikely to hold when areas that are unlikely to ever see the arrival of a flower-processing plant (such as the arid parts of Kenya, which are too dry and hot for flowers to bloom, or urban centres, since all greenhouse locations are exclusively rural) are included in the comparison group. Part of the identification work here will thus include the construction of a valid comparison group, namely by identifying the set of non-treated locations that could have plausibly been eligible for treatment, but have not (yet) seen the arrival of flower-processing plants. Our baseline specification will be a two-way fixed effects regression, controlling for year and ward fixed effects, where we will restrict the comparison group to locations that are deemed geographically suitable to flower production. To identify the key predictors of this geographic suitability, we will run best subset selection models to identify the set of geographic covariates that best predict plant entry among variables such as rainfall, temperature, elevation, solar intensity, soil nutrients, and proximity to water bodies.
SSA’s slow fertility decline puts pressure on the sub-continent to generate millions of jobs for increasingly large cohorts of labouicr market entrants. Identifying factors that may help curb fertility rates is thus of crucial interest to policymakers in the region. If salaried jobs for poorer women in rural areas can decrease fertility by lowering demand for children where it is highest (Zipfel 2022), current policies of supply-side policies such as expanding access and removing information barriers to contraceptive use would need to be complemented by demand-side policies that boost salaried employment growth, especially for women.