vaccination access. Even more than the analysis with Primary and Community Health Centers, this is subject to bias. However, it does provide the only opportunity to consider this relationship outside of rural villages.
I take advantage of the cluster design of the NFHS (a cluster, in this case, is a collection of on average 750 households in the same area). I aggregate the data to the cluster level and calculate the average number of vaccinations and the difference between this average for boys and girls. This regression will be run using the 1992 and 1998 data together; since I do not rely on the village-level information (either about camps or distance) I can use the 1992 survey as well.
The primary regression will consider the shape of the relationship between the level of vaccinations and the gender difference in vaccination. The results are shown in Table 8, where the dependent variable is the gender difference in average number of vaccines received; Column 1 considers a monotonic relationship between the level and the difference, and Column 2 considers a non-monotonic relationship. The results seem consistent with a nonlinear relationship: in Column 2 both the average and the average squared are significant and have the expected sign. The number of vaccinations ranges from zero to eight. The magnitude of the coefficients suggests that the gender imbalance is increasing up to an average of six vaccinations and decreasing thereafter.13
Non-Monotonicities in Mortality over Time
The evidence above suggests that there are non-monotonicities in child health investments and gender inequality. Although this maybe interesting on its own, this is likely to be more important if these non-monotonicities map into non-monotonicity in mortality. The ultimate outcome of interest here is not vaccinations per se but mortality, for which vaccinations may be an important input.
Even without looking directly at data on mortality, existing evidence on the connection between vaccination and mortality rates among children does suggest that changes in vaccination will map into changes in mortality. If getting vaccinated decreases mortality in the developing world (as we know that it does – see, among a huge additional literature – Aaby et al, 2002; Clemens et al, 1988; Koenig et al, 1990) then any changes in vaccination are likely to map fairly directly into changes in mortality.14 However, this link is not necessarily sufficient. In particular, if the
13This result is in contrast to the results in Pande and Yazbeck (2003), who argue that gender di erences in immu- nization across states do not seem to be related to state immunization levels. This may underscore the importance of considering the regional relationship at a less aggregated level.
14Note that this evidence means that even if there are, for example, compensating di erences in behavior (I vaccinate the boys, but give the girls more food) the link will still hold. Data from the field on the e ectiveness of vaccinations means that we can be confident that, even in the real world situations in which these vaccinations are given, we still see