hold only if the cost of vaccinations decreased over this period. It is difficult to measure costs of vaccination directly. However, we do observe that in nearly all states the average level of vaccination increases over this period (by an average of 0.30 vaccinations). There is no reason to think that the benefits of vaccination have been increasing, so the increase in level of vaccine is very likely to have been driven by a decrease in cost. It is worth noting that there are a few states for which the level of vaccination actually has not increased over this period, which potentially provide an additional test. When presenting the results I will show estimates for all of the states together, and then divide the sample based on whether the vaccination levels increased to test whether this relationship shows up only in areas where vaccination levels have gone up. The theory would suggest that we should only see this effect for states which saw an increase in vaccination rates over this period.15
Table 9 shows the primary results on mortality.16 In Column 1, I include all states. We see evidence of the non-monotonicity in mortality: the coefficient on the interaction between gender, time and initial vaccination levels is positive and significant, whereas the coefficient on the interaction between gender, time and initial vaccination level squared is negative and significant. Columns 2 and 3 divide the sample based on whether the state saw an increase in vaccination – our indicator for a decrease in cost – during this period. In Column 2, limited to the states in which we see vaccination increases, the coefficients have the expected sign and remain significant. However, in Column 3 the coefficients are not significant, and do not have the expected sign. This is what we would expect given the theory. If there is no decrease in cost, we do not expect to see any non-monotonic variation in mortality over time.
As in the case of vaccinations, here I also divide the sample based on gender preferences and re-estimate the regressions based on the reported ideal child gender ratio. Column 1 of Table 10 includes only people who report wanting more girls or equal numbers of boy and girl children, and Column 2 includes those who report wanting more boys. As in the regressions on vaccination, we see here that the non-monotonicity is much stronger and more significant for those individuals who express stronger male-biased gender preferences. Overall, the evidence in Tables 9 and 10 points to the conclusion that the non-monotonicities in vaccination are reflected in non-monotonicities in mortality.
15Related to this, I define vaccinations here based on vaccinations marked on a vaccination card only, not reported by the mother. This will combine changes in vaccination with changes in official sources of vaccination (i.e. those that will reliably require a card), and the hope is that it captures changes in costs more completely.
16The coefficients reported are marginal e ects from a probit model; since death is a relatively unusual event, an OLS specification is not appropriate.