laid out in the Data section. Observations are consistently weighted by population.10 Finally, it is important to note that in all cases, EC Pill refers to free provision by the public health system. In Chile, following the passage of the EC pill laws, the pill was also sold at private pharmacies. Unlike public data on EC pill usage, official data on EC pill usage in the private system is not available (Fernández et al., 2016). Thus, all estimates 10We do, however, document unweighted results in Appendix Table A2 and Figure A1, but our preferred estimates always weight by population to ensure that estimates are not driven by municipalities with very few hospitalizations where small total shifts can result in very large proportional changes. 10 refer to the impact of public provision. While we cannot formally assess the impact of private market provision without data on disbursements, to the degree that private provision fills gaps not met by the public health system and availability ‘spills over’ to areas not yet treated by the public system, we may expect our estimates to understate the actual full (public and private) effect of EC pill availability in Chile (Clarke, 2019). DD event study models such as those in equation 1 have a number of significant advantages over standard parametric ‘single-coefficient’ DD models where a binary indicator takes the value of 1 if the EC pill is available, and 0 otherwise.11 However, there is one drawback with such eventstudy models, and this is that they are based on binary treatment adoption indicators. Thus, while this model works well when considering the arrival of the EC pill, it does not work well when considering the intensity of use of the EC pill in each municipality and time period. In order to examine the intensive margin of use the EC pill (ie the intensity of use), we consider two additional models. The first is a single-coefficient DD model, where the EC Pill indicator is replaced with a measure of the rate of pill disbursements in a given municipality per 1000 women: Healthct = α + βPill Disbursementsct + X ′ ctΓ + ϕc + µt + εct. (2) Here, once again any municipal-specific or time-specific factors are captured by respective fixed effects, and β captures the intensive margin impact of EC pill availability. We additionally explore one specification where Pill Disbursementsct is replaced with Pill Rejectionsct, the rate of pill requests per 1000 women which municipalities refuse to provide. In this single-coefficient specification, estimated impacts βbwill capture both the interaction of the supply of the EC pill as 11Namely, (i) they allow us to capture dynamic effects, (ii) they provide a test of parallel pre-trends, and (iii) they take care of recent critiques that single coefficient models may be biased if effects are heterogeneous over time (Goodman-Bacon, 2018; de Chaisemartin and D’Haultfoeuille, 2019). 11 well as the demand for the EC pill, so can no longer be viewed as a pure treatment effect of the policy reform. Finally, to consider a pure policy effect while also allowing for heterogeneous impacts owing to intensity of use we, we provide a fully interacted event-study specification, where we re-estimate equation 1, however we now estimate a series of lags and leads for three municipality types, those with low, medium and high intensity of EC pill usage based on terciles of EC pill disbursements from official MoH disbursement data.12 This model extends equation 1 to examine whether any health impacts are larger in areas with more intensive usage of the EC pill. Note that here we are simply breaking down average impacts from equation 1 into intensity-specific groups in a single model, allowing for treatment heterogeneity where groups are constant over time. In this case heterogeneity is considered by intensity of policy adoption. Such models documenting heterogeneity in a DD setting are frequently estimated, see for example Bhalotra and Venkataramani (2015) (heterogeneity by race/age), and Myers and Ladd (2020) (heterogeneity by exposure time). 4 Results 4.1 Descriptive Statistics We provide descriptive plots of the two principal health outcomes over time in Figure 2. In the case of hemorrhage early in pregnancy (panel A), while there is a general downward trend from 2002–2005 which then flattens until 2009, there is a clear reduction following the rollout of the 12Specifically: Healthct = α0 + ∑ 3 i=1 ∑ 8 j=−8 δi,−j∆ ( EC Pillc,t+j × Pill Intensity = i ) + X ′ ctΓ + ϕc + µt + εct, (3) where all details follow equation 1, however we now estimate a series of lags and leads for three municipality types based on intensity of EC pill usage (indexed with i here). The separation into terciles of intensity of EC pill usage is an arbitrary choice. We could present alternative groups, however this complicates presentation of results and additionally challenges statistical power, as such we limit results here to three policy-specific exposure groups. 12 EC pill in Chile in 2009. In the case of abortion morbidity (panel B), while there is again a sharp drop following the EC pill rollout in 2009, this trend is much noisier, with a sharp rise observed from 2007-2009, and then additionally from 2012 onwards. Plots by quinquennial age group are provided in Appendix Figures A2 and A3. Descriptive statistics of all morbidity and contraceptive measures by municipality and year are provided in Table 1. We observe that morbidity for hemorrhage early in pregnancy is