The welfare effects of civil conflict in Africa

Abstract

I explore the question ‘How has civil conflict affected welfare in Africa?’. To answer this question I utilize descriptive statistics, naïve OLS estimates, and furthermore I construct a shift-share instrument for a 2SLS estimation. I construct a panel dataset that includes variables for constructing the main conflict measure, fatalities per 1000 population, and the outcome variables of interest, GDP per capita and under 5-year-old child mortality. The key results from the descriptive statistics are that the distribution across countries in terms of conflict is very unequal, and the temporal distribution of conflicts has a high variance and does not seem to indicate a clear intertemporal trend. Results from the naïve OLS regressions are mixed. The coefficients for non-state conflict fatalities receive non-significant results when the outcome variable is GDP per capita, and only barely significant results for child mortality. Neither the coefficients, nor the significances are particularly sensitive to adding a population density control. A shift-share instrument (Bartik-instrument) is constructed by interacting an indicator for the net-export status of a country with a global food price indicator to create an instrument which attains negative values for net importers and positive value for net exporters. This construction is inspired by a recent paper by McGuirk & Burke (2020). The main results from the IV-estimations reveal that the first stage of the constructed instrument is quite strong (𝑝 < 0.01) regardless of included controls. Also, the relevance of the instrument is high with similarly low p-values for the reduced form analysis. The final 2SLS-estimations reveal a statistically significant (𝑝 < 0.05) effect of fatalities per population on both the outcome variables, and comparing to relevant conflict measure percentiles, the economic impact for both of the outcome variables is large. The final estimation is however subject to a number of assumptions and threats to validity, which cannot be directly tested for, and must be qualitatively assessed.