UC Santa Barbara

This dissertation consists of three chapters. In the first chapter I analyze the impact of unemployment insurance on personal bankruptcy filings. In response to the COVID-19 pandemic, the US federal government expanded unemployment benefits and extended coverage to previously ineligible workers through Pandemic Unemployment Assistance (PUA) and Federal Pandemic Unemployment Compensation (FPUC). I study the impact of pandemic era unemployment benefit terminations on personal bankruptcy filings in states that ended these federal programs early. Using nonlinear difference-in-differences, I find that Chapter 13 filings increased between 14-15% on average. These findings are consistent with previous work that documented an increase in employment and a decrease in household financial security associated with the termination of pandemic unemployment insurance programs.

In the second chapter, which is joint work with Molly Schwarz, we investigate the extent to which the provision of Medicaid to previously ineligible, low-income childless adults affects their household composition. Using a staggered adoption difference-in-differences design on an urban sample of individuals with less than a high school degree, we find that 26 to 39 year olds experience a significant 4.2% decline in the number of individuals living in the household, which is due to living with fewer extended family members. At the same time, 26 to 39 year olds experience a relatively smaller decline in the number of rooms (1.8%), leading to a 3.1% reduction in the level of household crowding, as measured by persons per bedroom. These reductions in household crowding are strongest for Hispanic individuals and those living in areas with above-median housing costs. In comparison, there are no significant impacts on household composition for 40 to 64 year olds as a consequence of the policy.

In the third chapter I explore the estimation and interpretation of the coefficient on a treatment variable in two-way fixed effects regressions with group-by-time fixed effects. Using the decomposition from de Chaisemartin and D'Haultfoeuille (2020) I show that while this design can be successful in leveraging within group variation to estimate the parameter of interest, in cases with heterogeneous treatment effects it leads to the well-documented weighting issues that arise in the canonical two-way fixed effects regression when there is variation in treatment timing. As opposed to the canonical two-way fixed effects regression, however, when group-by-time fixed effects are included weighting issues can arise when there is variation in treatment timing within groups, across groups, or both. I also show that in this setting it is possible to include groups in which all units share the same treatment sequence or that include only subgroups with constant treatment, and that observations in these groups do not contribute to the estimation of the parameter of interest. In this case, under heterogeneous treatment effects the average treatment effect on the treated cannot be obtained, but rather in certain cases the average treatment effect on the treated for groups that have within group variation in treatment sequences is obtained under the assumption that parallel trends hold in every group. While this is typically still a parameter of interest, including groups in which all units share the same treatment sequence or that include only subgroups with constant treatment incorrectly reduces the size of the standard errors and therefore biases inference.