Health Womanpower: The Role of Federal Policy in Women's Entry into Medicine (Job Market Paper)
During the 1970s, women’s representation in medical schools grew rapidly from 9.6% of all students in 1970 to 26.5% in 1980. This paper studies the role of federal policy in increasing women’s access to medical training through two distinct channels: pressure to curb sex discrimination in admissions and a massive expansion in total enrollment through Health Manpower policy starting in 1963. To study this, I construct a novel school-by-year data set with enrollment and application information from 1960 through 1980. Using a continuous difference- in-differences design, I find that medical schools respond to the threat of losing federal contracts by increasing first year enrollment of women by 4 seats at the mean, which explains 27% of women’s gains between 1970 and 1973. Further, I provide OLS evidence that year-to-year expansions explain around 40% of women’s gains from 1970 to 1980; I verify this result in a synthetic control case study to identify the increase in women’s enrollment resulting from large jumps in capacity.
Continuous Treatment Difference-in-Differences with Unknown Controls: A Data-Driven Approach
(with Elird Haxhiu)
This paper studies difference-in-differences (DD) research designs where all observations receive a continuous treatment (or dose) in response to an aggregate policy, so there is no group that is ex post unexposed. This setting stands in contrast to the recent literature re-examining DD estimators which typically requires that a subset of observations never receive the treatment to identify the Average Treatment Effect on the Treated (ATT). We develop a framework to estimate the treatment effect when the dose takes effect only after a cutoff value, the Minimum Effective Dose (MED), and introduce the average treatment effect on the effectively treated (ATET) as our target estimand. We propose a sample splitting estimator of the ATET and MED under non-parametric assumptions on the dose response function. First, in a hold-out sample, we borrow methods from the pharmacological literature to estimate the MED in a model selection step. This is then used to estimate the ATET with the remaining observations in a second step. This estimator is asymptotically conservative: it does not erroneously identify any treated units as untreated in the limit even if the MED is on the boundary of the parameter space, but as a result, it provides an attenuated estimate of the ATET. We use the bootstrap procedure in Efron (2014) to construct standard errors for the ATET estimate that reflect uncertainty over the value of the MED. Our simulations suggest that this estimator performs well in finite samples.