ON-LINE: Zoe McLaren - Stormborn: Evaluating the Impact of Facility-Based Delivery on Health and Economic Outcomes Using a Machine Learning Approach
Event details
Author: Zoe McLaren (Uni of Maryland Baltimore County)
Abstract: There is wide consensus among social scientists about the importance of early life investments in determining outcomes later in life. Therefore, encouraging women to give birth in health facilities rather than at home remains a key policy objective in low-and middle-income countries. Despite this focus, there is little unconfounded evidence on whether facility-based delivery leads to improved child health and economic outcomes relative to delivering at home. This paper uses a rigorous instrumental variables approach with a machine learning component to examine the impact of facility-based delivery on child, maternal and household outcomes in Africa, including spillovers between siblings.
My identification strategy uses exogenous weather shocks such as rain and high temperatures in the days leading up to birth as instruments for facility-based delivery since weather shocks may hinder travel and therefore induce at least some women who considered delivering in a health facility to instead deliver at home. I use rich data from the 2012-2019 Demographic and Health Surveys that includes women's detailed birth histories as well as family, household and geographic data. I link births with high-
resolution and high-frequency weather data from the ERA5-Land database. I use machine learning methods (lasso, ridge regression and elastic net) to select the strongest instruments from a large set of 400 potential instruments for facility-based delivery.
Results from the elastic net instrumental variables estimation show that facility-based delivery is statistically significantly associated with an increase in under-five survival, an increase in the height-for-age percentile, an increase in the weight-for-age percentile, and a shorter subsequent birth interval relative to children who were delivered at home. These children are more likely to be vaccinated, however this is not statistically significant. I found a large positive spillover effect of having an elder sibling deliver in a health facility on the likelihood of receiving at least one vaccination. The local average treatment effect estimate from this novel instrumental variables approach identifies large overall benefits of facility-based delivery that accrue to children of women who are dissuaded from traveling to a health facility by relatively small, potentially surmountable barriers. There is therefore a need for better interventions to encourage, incentivize and enable facility-based delivery for this key target group: women who may be undecided about delivering in a facility or for whom small barriers are constraining. Policy interventions targeting this marginal population are therefore likely to yield a large return on investment.
Host: Cheti Nicoletti (York)