Essays in Labor Economics and Child Welfare
MetadataShow full item record
CitationGrimon, Marie-Pascale. 2021. Essays in Labor Economics and Child Welfare. Doctoral dissertation, Harvard University Graduate School of Arts and Sciences.
AbstractThis dissertation comprises of three independent essays on the economics of child welfare. Essays respectively examine the effects of child welfare on parents (Chapter 1), on child deaths (Chapter 2), and the effects of an algorithmic tool to improve child welfare practice (Chapter 3).
One-third of American children's homes are investigated by child welfare agencies for abuse and neglect, yet the impacts of these investigations on parents are not well studied. In the first essay, I establish that months in which a child welfare investigation begins are pivotal moments in parents' lives. I then identify causal effects of child welfare using a novel research design which combines an event study set-up with imperfect random assignments of investigators. Opening a child welfare case considerably increases mothers' enrollment in mental health and substance abuse services, suggesting that child welfare encourages parents to seek help they may need. The effects of opening a child welfare case persist even when no child is removed from the home, and the increase in use of substance abuse services persists for up to six years. In contrast, fathers are much less responsive to child welfare involvement both in the short and long run.
Despite prominent disparities in child mortality, the child protection system's impact on child death disparities has received limited attention. In my second essay, I leverage close to random assignments of workers to families in an American county to provide the first causal estimates of the child protection system on child deaths reported to the medical examiner. I find that opening a child welfare case drastically reduces such child deaths. Extrapolating from these local average treatment effects, the black-white child mortality gap would be two to three times larger without child welfare involvement. Child welfare plays a key role in addressing mortality inequality among children.
Machine learning-based tools have drawn increasing interest from public policy practitioners, yet our understanding of the effectiveness of such tools when paired with human decision makers is limited. In the third essay which is co-authored we use a randomized control trial to evaluate the effects of an established algorithmic decision aid tool implemented by an American child welfare agency. Slightly less than halfway through the trial, this paper presents preliminary evidence on the effects of showing a child’s predicted risk score on child welfare decision outcomes. Even when scores are not shown, workers target case openings and services more to high-scoring children relative to low-scoring children. Workers are thus already sensitive to underlying risk. Still, showing a score further increases the likelihood that high-scoring children have a founded allegation of abuse or neglect. Showing a score also decreases the likelihood that families not visited by child welfare are called back in. These preliminary results suggest that the tool may be improving targeting of child welfare visits.
Citable link to this pagehttps://nrs.harvard.edu/URN-3:HUL.INSTREPOS:37368316
- FAS Theses and Dissertations