Sensitivity of the overall effect of conflict exposure on female years of schooling to modeling assumptions.

Column (1) uses a 50 km radius for conflict exposure, instead of 25 km. Column (2) removes household characteristics. Since these characteristics, taken from survey data, are measured at the time of the survey, they may have changed since the treatment (school-age conflict exposure) or may have been affected by the treatment. Column (3) removes sample weights. Sample weights are necessary to ensure that estimates are representative of all the countries in the sample, but may be sensitive to the population estimates used. Column (4) only includes observations that did not have conflict in the survey year and the year before the survey, to exclude the possibility that effects are due to more recent conflict. All models include cluster, country-birth year, and country-birth month fixed effects and use the same female sample as the main results (<a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0314106#pone.0314106.s002" target="_blank">S2 Table</a>, column (3)). Standard errors are clustered at a DHS cluster level. *p<0.1; **p<0.05; ***p<0.01. (PDF)