vaccination and aims to demonstrate, among other things, that the results are not driven by the use of health camps as the shifter of access.
I first proxy for access using the reported distance from the nearest Primary Health Center, Community Health Center, or Government Hospital, as reported in the village survey in the 1998 NFHS. Approximately 50 percent of women report this as their source of immunization, so it seems to be a good proxy for access. Of course, access to these centers has other implications and may be correlated with unobservables. However, there is no obviously apparent bias that would produce a non-monotonic relationship in gender imbalance.
Figure 4 mimics Figure 3, this time considering the relationship between distance to the nearest source (5 groups) and vaccination gender differences. Again, as with vaccination camps, we see a non-monotonic relationship. As the distance increases, vaccination levels decrease for both boys and girls, but the difference increases and then decreases.
Table 7 shows the relationship between total vaccinations and gender, interacted with both distance and distance squared. In this case, we measure vaccinations for all children over the age of six months. Unlike in the measure of health camps, we do not need to focus only on children who were vaccinated in the last year, since the distance to the nearest source will be constant across years. In this case, the theory would predict the interaction with distance to be negative and with distance squared to be positive.12 A sizable fraction of people report having a health facility in their village. These are coded as zero distance, although this may not be strictly correct.
In Column 1 of Table 7, I show the regression with all observations; in Column 2, I restrict to people who report a non-zero distance to avoid any issues with the in-village measure. The coefficients have the expected sign, although they are much more significant in the second column, when I leave out people who have a health facility in their village. This may reflect the fact that there is variation in distance even within the village which is not measured. Columns 3 and 4 of Table 7 split the entire sample based on gender preferences (as was done in Table 6). Column 3 shows the regression for women who report wanting more girls or wanting equality; Column 4 shows the regression for women who report wanting more boys. Again, as expected, the results are much stronger for those with a male-biased preference.
As a second test, I consider the cross-regional relationship between the level of vaccination and the gender difference in vaccination. In this case, the level of vaccination is the proxy for
12These predicted interactions have the opposite sign from the interactions on number of camps because increases in the number of camps imply increases in access and increases in distance imply decreases in access.