Measures of EC pill rollout and usage We generate two measures of rollout and usage of the EC pill in Chile, which is our principal independent variable of interest. For the rollout, our first indicator is a dummy variable, coming from a series of telephone surveys conducted by FLACSO between 2009–2011 (Dides et al., 2009; Dides C. et al., 2010, 2011). In these surveys, local health centres in each municipality were asked whether they prescribe the EC pill, and under what circumstances. If the centres respond that they do, the municipality is classified as providing EC. If they report that they do not, or that they only provide it in the limited case of rape, then they are classified as non-pill municipalities. Secondly, for measuring intensity of use, we generate a rate of pill disbursements and a rate of pill rejections, by harmonizing previously unused administrative data provided by the MoH. In the case of rejected pill requests, these only began being recorded in 2010, so data on this is not observed in 2009. A graph of the number of municipalities recorded to allow EC pill disbursement, and the actual disbursements according to MoH Data is provided as Figure 1. In Appendix Table A1 we document that these sources are in general agreement. 8We observe the universe of 26.2 million hospital visits occurring over this period, 99.74% of which are correctly matched to the municipality of residence of the patient. These data unfortunately don’t include richer covariates such as religion or ethnicity. 9While we do not have access to microdata for ambulatory visits, macro-level information shows that there are many such visits in the country each year. For example, in 2016 nationwide there were 755,547 ambulatory visits classified as gynecological check-ups. This compares to 299,855 hospitalizations for reasons related to maternal health in this period. Thus, our microdata captures only one (important) margin of healthcare utilization. 8 Other municipal level records Finally, we collected or generated a number of other data sources at the municipal×year level. These are (a) the population of fertile-aged women (provided by the National Statistical Institute); (b) the identity, sex and party of each municipal mayor and his/her vote share (from the Electoral Service); (c) administrative records on all other contraceptive disbursements through the public health service, and (d) placebo health outcomes generated from the same administrative health records (male morbidity, and morbidities in the puerperium period). Combining these data sources, we were able to ensemble a dataset of a maximum of Chile’s 346 municipalities over 15 years of data, or 5190 observations/registers. A small number of observations have missing measures in certain periods. In particular, our measure of EC pill availability has 103 missing observations for years in which municipalities did not provide information on their pill disbursement status. Similarly, the measure of refused pills is not recorded in 2009 only. We document summary statistics in the following section. 3.2 Methods We exploit the staggered arrival of the EC pill to different municipalities by estimating the following difference-in-differences (DD) event-study specification: Healthct = α0 + ∑ 8 j=−9 δ−j∆EC Pillc,t+j + X ′ ctΓ + ϕc + µt + εct (1) These event-study specifications are increasingly common in DD settings, and here we adopt the notation of Freyaldenhoven et al. (2018), normalising δ by setting δ−1 = 0 so that our reference period is one year prior to pill adoption in each municipality. In this specification, we are interested in the leads and the lags of the policy change, where leads capture any prevailing trends prior to the reform in earlier versus later-adopting municipalities, and lags show the change in health outcomes 9 following the availability of the EC pill. If the EC pill has any impact on health outcomes, we should see a divergence in health outcomes emerge only after the arrival of the EC pill to each municipality. In specification 1, we include a set of 9 yearly leads and 8 yearly lags to capture dynamic impacts of the reform. Given variation in reform timing, initial lags and leads capture differences in treatment status (treated vs. untreated), while later periods capture pure variation in timing. Recent work by Goodman-Bacon (2018) suggests event-study specifications as a key strategy in such a setting if treatment effects are not constant over time. Year and municipal fixed effects µt and ϕc absorb time- and municipal-invariant factors, and standard errors are clustered by Chile’s 346 municipalities. As well as capturing any dynamic impacts of the reform, for example any technology diffusion over time, specification 1 provides evidence in favour of parallel (pre-)trends if we can reject that each δj = 0 ∀ j < 0, given that prior to the reform outcomes in both treated and untreated municipalities were following similar tendencies. The observation of parallel pre-trends provides some support for identifying assumptions of parallel trends in post-treatment counterfactuals. In equation 1 Healthct refers to average rates of morbidity in municipality c at time t, and EC Pill refers to the availability of the emergency contraceptive pill in the same municipality and time period. We examine stability to the inclusion of the time-varying controls Xct capturing sociopolitical characteristics of each municipality, as