population, and ideal sex ratio) from the 1998 survey.5
In the first column, I simply show the regression on village population, village area, distance to another source, and state fixed effects (the coefficients on state are not shown, but the test that they are all equal is reported). As suspected, all four of these parameters matter – there are more camps in larger villages and in those villages closer to a PHC or CHC, and we can reject the equality of the coefficients for each state with high confidence. Column 2 of Table 2 includes a set of additional demographic controls. None of these additional controls are significant; placement does not seem to be related to education, income, age, religion, or number of children. This is also true if we do not condition on state, village size, and distance (results available from the author).
A more specific concern is that camp placement might be correlated with initial vaccination conditions. It is not possible to examine this possibility using the 1998 data alone since we expect the number of camps to affect vaccination levels. To test for a relationship between vaccination camp placement and pre-existing gender differentials in vaccinations, I take advantage of the 1992 wave of the survey. Figure 2 shows the relationship between the average number of camps (at the district level) and vaccinations rates for the district in 1992. Although the sample sizes are small as we increase the number of camps (so the data is somewhat noisy), the general picture does not suggest either a strong linear or non-monotonic relationship between number of camps and gender difference in vaccination. Table 3 estimates this relationship between the average number of camps (at the district level) and gender difference in vaccination rates for the district in 1992. Column 1 includes the gender difference in vaccination and average vaccinations linearly; column 2 includes a quadratic in each. This regression shows no relationship between the gender differences in vaccination and camp placement. This suggests that although there are important district-specific drivers of the number of camps (as in Table 2), these do not seem to be related to gender imbalances in vaccination.
This discussion should provide some confidence that, although placement of these camps is by no means completely exogenous, the primary drivers of the placement can be observed and controlled for. It is also the case that, even if these camps were placed endogenously (for example, targeting areas with low vaccination levels), the endogenaity would have to be of a particular form in order to induce the non-linear results seen here. Targeted camp placement could drive the results only if areas with high male preference were targeted to receive a few camps, while areas with low male preference were targeted to receive either no camps or many camps. It is obviously not possible
5In this case, and throughout this section, the number of camps is top-coded at ten. Ninety-eight percent of villages have ten or fewer camps, and the top-coding avoids allowing outliers to drive the results.