ABSTRACT
Our fixed effects (OLS) and instrumental-variable results all point in the same direction with regard to four main findings. First, and as expected, resource dependence decreases ANS per capita. Second, both armed conflict and homicides have a negative impact on ANS per capita. Third, education expenditures are far more path dependent than ANS, are lower in resource dependent countries, and do not seem to be influenced by violence dynamics. The latter comes as a surprise but is consistent with previous empirical findings. Fourth, our instruments are highly relevant and explain the endogenous variables well. The results are presented in Tables IV and V. Table IV presents the impact of armed
conflict and resource dependence on ANS and education expenditure per capita, while Table V presents the impact of homicides and resources dependence on the same outcome variables. In each table, the first pair of columns shows the findings for ANS per capita as outcome variable, and the second for education expenditures per capita. The first column of each pair refers to the instrumental variables specification, with the coefficient associated with the instrument reported at the bottom of the table. The second column presents results stemming from the OLS specification. We include all covariates in the reduced form equation of the two-stage procedure but do not report them for the sake of readability. In Table IV, the reduced-form result shows that the more agricultural land available, the
less the propensity for conflict. According to our estimates, an increase of 10,000 square kilometers in agricultural land reduces a country’s propensity for conflict by roughly 6% (or 18.2% of the sample standard deviation). Conflict itself is shown to have a considerably development, as it decreases ANS per capita by almost
DEFENSE SPENDING, NATURAL RESOURCES, AND CONFLICT
$220. Column 2 of the same table shows that the OLS coefficient associated with armed conflict accounts for only 15.3% of the instrumental variables coefficient. Put differently, conflict reduces ANS per capita by $33. While the impact of conflict estimated in the OLS specification is substantially lower, it points in the same direction as the IV specification. The coefficient of the instrumental variables estimate is larger because of the loss in precision and its analogy to a local average treatment effect (LATE): We only identify the effect of conflict on ANS for the sub-sample of countries whose propensity for conflict depends on agricultural land availability. As a result, the point estimate associated with conflict in the instrumental variables specification has to be considered as an upper bound. Given the limitations of our data, we agree with Imbens (2009) that it is better to have a LATE result than no result. Moreover, when comparing the IV and the OLS results, the latter appear to be biased towards zero pointing to an inherent requirement for instrumenting armed conflict to get a more complete picture of the magnitude of the negative impact conflict can have on ANS per capita. Resource dependence also has a significant and negative impact on ANS per capita. In
the IV specification, the effect is significant and the coefficient estimate amounts to -1.41. In the OLS specification, the magnitude and sign of the coefficient remain similar but the estimate is no longer significant. Apparently, the endogeneity of armed conflict also biases the other coefficient estimates towards zero in the OLS framework. When instrumenting conflict, however, the negative relationship between resource dependence and ANS per capita amounts to more than 10% based on the sample standard deviations. The other control variables remain virtually unchanged across the two specifications.
Sustainable development as measured by ANS per capita shows a moderate level of persistence with a coefficient estimate of roughly 0.42. Richer countries in terms of GDP per capita tend to have higher levels of ANS per capita. However, faster growing economies invest less in genuine savings. Consistent with the argument presented earlier, resourcedependent, fast growing economies tend to deplete their natural capital base faster. The same logic applies to the agricultural sector, where the agricultural value-added per worker is with ANS while growth in the agricultural sector decreases ANS.
