related morbidity and in general more noisy measures, we are less able to distinguish between group-specific estimates. There is, however, some evidence of an immediate reduction in rates of abortion in high and medium intensity areas, and a longer term reduction which is significant for high-intensity areas only. In both cases, we observe evidence to suggest that the arrival of the EC pill reduced rates of morbidity 14In particular, Goodman-Bacon (2018) shows that this is because units which have been previously treated but do not change their treatment status in a given period act as part of the control group, and can substantially bias estimates. 15 owing to reproductive health causes, in a way consistent with less need to access unsafe abortion. 4.4 The Impact of Rejection of EC Pill Requests on Outcomes Finally, we consider the role that rejected requests for the EC pill plays on maternal health outcomes. In Table 3 we present DD models where the independent variable is the rate of rejected pill requests per 1,000 women. We present two specifications; those in columns 1 and 3 which use the full set of years, including the pre-pill period where the rate of rejection is 0 by construction (given that there is no request or rejection), and those in columns 2 and 4 (our preferred specifications) where we consider only the post-EC pill period. We observe suggestive evidence of a mirror impact of the results described earlier in this paper. Here, a higher rate of rejected pill requests results in higher rates of haemorrhage (and this is statistically significant at least in the post-EC pill period), and higher rates of abortion morbidity, regardless of the specification examined. These results are consistent with failure to provide EC pills upon request shifting women into seeking more risky contraceptive options, such as clandestine abortion, with worse associated health outcomes. In general these estimates are relatively substantive in magnitude. For example, column 2 which is based on variation in the intensity of pill requests after the arrival of the EC pill suggests that – even conditional on municipal-specific and time-specific factors – an additional 1 pill rejection per 1,000 women in a municipality is associated with an increase in 0.15 cases of haemorrhage per 100,000 women. To put this in perspective, we can ask how a 1 standard deviation increase in rates of rejected requests would correlate with rates of haemorrhage. From Table 1, a 1 standard deviation increase in pill rejections is 12.55 pill rejections per 1,000 women. Thus, a 1 standard deviation increase is associated with 12.55×0.152 = 1.91 fewer cases of haemorrhage per 100,000 women, or around 7% of the mean of the dependent variable in this period. A similar calculation 16 for rates of abortion related morbidity is 12.55 × 0.077 = 0.97, or around 42% of the mean of the dependent variable listed at the foot of the table. We now turn to a more complete discussion of estimated effect sizes. 4.5 Effect Sizes in Context Given estimates presented up to this point, it is important to ask about how these effect sizes map into real reductions in morbidities. We place estimated effect sizes in context here, also providing a back of the envelope calculation of their implications for healthcare costs. This also acts as a sanity check of our estimates as it allows us to ask roughly what they imply for the efficacy of each pill disbursed. It is important to note as a caveat here that this discussion is all based on extrapolating marginal estimates presented from previous analyses. In this analysis we will document a calculation based on continuous DD models for illustration, and then also extend to ask what the point estimatesfrom all event studies imply in a similar strategy. To begin, consider the estimate presented in Table 2, Panel A, column 1. This estimate suggests that a marginal increase in pill availability by 1 per 1,000 women in all municipalities in Chile reduces haemorrhage by 0.439 per 100,000 women. The question in calculating the implications for the total effect is how many cases of haemorrhage nation-wide are 0.439 per 100,000 fertile-aged women? Based on Table 1, the average municipality has 13,016.5 fertile-aged female residents, and there are 346 municipalities in the country. We can thus estimate the total number of estimated avoided haemorrhages per marginal increase in EC pill as: 13, 016.5 residents municip. × 346 municip. × 0.439 cases of haemorrhage 100, 000 residents = 19.77 cases of haemorrhage. This implies that for each additional pill disbursed per 1,000 women in a given year, our estimate 17 suggests a total of 19.8 fewer cases of haemorrhage in the country. Note however that on average, in the post EC period in Chile there were a total of 3.05 EC pills disbursed per 1,000 women (figures provided in Appendix Table A3), and so the estimated total reduction in cases of haemorrhages owing to the EC pill is 19.77 × 3.05 = 60.3/year. Finally, on average, each year after the arrival of the EC pill, 14, 120.3 pills were disbursed, implying that for every 234 pills, a single case of haemorrhage was avoided.15 We can undertake a similar activity using estimated impacts from event studies. Consider Figure 3. If we examine the coefficient on the second lag of this plot, this coefficient is −3.92 in panel (a), and −1.59 in panel (b). These are expressed as morbidities per 100,000 women given