∆Priceit =β0+β1Returneesˆ it+Controlsi jtβ2+ηi jt (1.4)
∆Violenceit =γ0+γ1Returneesˆ it+γ2∆Priceit+Controlsi jtγ3+ζi jt (1.5) In the systems of equations approach, for the results to be consistent with the theory, ˆγ2βˆ1will be positive. Furthermore, when ˆγ1 is positive, this means there is an increase in violence as a result of refugee return shocks that is not explained by the food price mechanism, that is, the wage mechanism or other channel.
For the empirical approach to accurately identify the extent to which refugee return causes violence through food price increases, a number of assumptions need to be made. First, the exclusion restriction must hold, not only with respect to food prices, but also with respect to violence. In addition to what was previously discussed, there must also be no way that a natural disaster in a country of asylum can affect violence in a country of origin except through refugee return. To ensure this, I still control for natural disasters in the country of origin. I also control for wars in the country of asylum (which could otherwise have contagion effects) and for flows of refugees other than the returnees coming into the country.10 The exclusion restriction in the setup with violence as the dependent variable is robust to a number of tests (Camarena 2011).
The other necessary assumption is sequential ignorability. I have argued throughout that I am con- ditioning on the theoretically relevant set of controls to ensure ignorability (conditional independence). Sequential ignorability also requires that the effect of food prices on violence is identified. I make two modeling decisions that address this concern. First, I focus on changes in food prices and changes in
9The analysis presented uses a combination of programming from Baum, Schaffer, and Stillman (2002), Hicks and Tingley
(2011), and modifications of Tingley et al. (2014).
10These flows of refugees could affect violence through the same mechanism discussed. However, they may also affect
violence. This should limit problems with serial correlation. The final concern is reverse causality; food price increases are causing violence, and violence is causing food price increases.
Over a long time period, this could be a significant concern. However, it is less likely to be a concern given the time unit of analysis. The concern would be that the increased violence from the war, not refugee return, drove food prices up even more, and more violence ensued. Essentially, this would create a cycle of violence and increase food prices, which then increase violence again. Thus, the effect of food prices on violence was biased upward. This is less likely in the scope of a year, the time unit of analysis in the data used here. Typically, the concern is that violence may undermine the growing of food, decreasing food supply. Since a growing season is often a substantial part of the year, violence raising food prices may occur over the course of many years. It is less likely that this will be captured differences in food prices from one year to the next.11
Results
The analysis of the effect of refugee return and food prices on violence provides some evidence for the theory. Shock-like refugee return causes food prices to go up, and when food prices go up, violence also increases. However, caution should be used when interpreting these results because they have less precision than the evidence for the first hypothesis, and the results are also less robust with the inclusion of time fixed effects.
Table 1.4 displays the results of the analysis from the final specification above. On the left, the dependent variable is change in deaths, and the first column presents results with origin country fixed effects; the second column includes the linear time trend. Substantively, the estimates suggest that one battle death occurs for every 33 refugees who return to a given province in their country of origin, net of the effect of food price changes on battle deaths. Similarly, the coefficient on change in food prices suggest that a 10% increase in food prices leads to one additional battle death. These point estimates are both statistically significant at conventional levels. The inclusion of the time trend reduces the estimates a bit. Substantively, one battle death occurs for every 50 returnees, net of food price changes, and a 15% increase in staple food prices results in one additional battle death. These results are all
statistically significant at the 95% confidence level.
In the third and fourth columns in Table 1.4, the dependent variable is change in violent events. The estimates suggest that there is one additional violent event for every approximately 60 returnees to a given province, net of food price changes. With respect to food prices, there is one additional violent event for every 10.5% increase in food prices. With the inclusion of the linear time trend, the estimates are again reduced. Now there is one additional violent event for every 110 returnees, net of food price changes, and a 23% increase in food prices is required to result in one additional violent event. The estimates of the effect of refugee return on violence are statistically significant at the 95% confidence level, and the estimates of the effect of food price changes on violence are statistically significant at the 90% confidence level.
Combined, these results give added weight to the theory being tested. They demonstrate that shock- like refugee return does result in violence, even absent how food price changes may be affecting vio- lence. Furthermore, food price increases also impact violence.
Given the results in Table 1.4, the data can also be used to answer a few questions about the