Introduction
According to the UNHCR, an unprecedented 65.3 million people were displaced from their homes by war, internal conflicts, natural disasters or poverty in 2018. As developed countries like those in the European Union increasingly feel the impact of this instability and conflict through increased migratory flows, a frequent reaction is a call for larger amounts of development aid, to reduce poverty and other root causes of displacement.1 Still, while the literature on growth effects of aid converges towards a small, mostly positive, effect on development outcomes (Clemens et al., 2011; Dreher et al., 2018a; Galiani et al., 2017; Kilby, 2015), other studies have raised the question whether aid might fuel conflict, instead of reduce conflict (e.g., Nunn and Qian, 2014; Child, 2018; Crost et al., 2014, 2016).
Our paper provides a comprehensive evaluation of the aid-conflict nexus. We combine the strengths of existing approaches on the country level (e.g., Bluhm et al., 2016; Nunn and Qian, 2014), with the advantages of studies focusing on sub-national aid data in specific sectors in individual countries (e.g., Berman et al., 2011; Child, 2018; Sexton, 2016; Van Weezel, 2015).
By considering a large set of all aid-eligible countries in Africa, the continent most consistently plagued by reoccurring conflicts, our results can be meaningfully interpreted beyond the context of an individual country. We use sub-national georeferenced project-level data concerning the World Bank (WB) and China in order to link aid projects, as well as conflict events, more directly than previous country-level studies. Our identification strategies use specifications with comprehensive sets of fixed effects (FE) and time trends, as well as instrumental variable (IV) estimates. The IV strategy adapts those in Nunn and Qian (2014) and Dreher et al. (2017), and interacts the predetermined probability of a region receiving aid with exogenous temporal variation in the WB’s IDA liquidity and Chinese steel (over-)production.
Our results provide several important insights. Most importantly, a wide range of fixed effects specifications, as well as IV specifications, provide no indication that on average aid fuels lethal conflict. Using a fixed effects specification, a standard deviation change of one in the WB aid decreases the conflict likelihood by about 1.6 percentage points. The effect remains negative but becomes insignificant when using an IV specification. Nor is there any evidence for a conflictfueling effect of Chinese aid. The IV estimates are negative, but close to zero and statistically insignificant.
To explore the potential heterogeneity in the results, we then consider aid projects in different sectors individually. We find a significant negative effect on the likelihood of conflict from projects in the finance sector (concerning the WB only), as well as for those in the transportation sector (the WB and China). There is no specific sector in which aid significantly increases conflict. Moreover, when considering the actors involved in, and responsible for, a conflict, both WB and Chinese engagement consistently lead to a reduction in lethal violence by governments against civilians.
For both donors, we find no positive effect on lower-level, non-lethal types of conflict like demonstrations, riots, and strikes. We do, however, find that increased Chinese engagement leads to an increase in government repression against their citizens. Survey evidence, using georeferenced Afrobarometer data, supports these results on the effects of Chinese aid.
We make four main contributions. First, we provide causal evidence on two important donors: the WB and China. The WB is a multilateral donor that emphasizes scientific expertise, frequently imposes human-right, as well as, sustainability conditions, and specifically engages in ”conflictsensitive programming” (e.g. Bannon, 2010). China, in contrast, has become one of the most important bilateral donors, but is often portrayed as a "rogue" donor (Naím, 2007), with economic targets, such as securing resource supply, as a central part of its aid strategy. Although, in some regards, China does not seem drastically different than other bilateral donors (see Dreher and Fuchs, 2015), but there are some objective differences. China propagates a policy of "noninterference" in the internal affairs of recipient countries, emphasizes "mutual economic benefits" over political freedom as well as democracy, and grants high levels of discretion for the host governments to distribute and use aid resources as they see fit. Analyzing the WB and China, thus, covers donors that represent the two ends within the broad spectrum of potential approaches to development and their impact on conflict.
Our second contribution is that we cover aid projects in a broad set of developing countries, while simultaneously assigning project locations to specific sub-national administrative units (based on Strandow et al., 2016; Strange et al., 2017). Our geographical focus on Africa is determined by a trade-off between external and internal validity. On the one hand, we augment the literature by studying comparable aid projects in more than one country, so as to increase the relevance of our results beyond the narrow single-country context. On the other hand, we want to make sure that the results can be meaningfully interpreted within our sample. We restrict ourselves to Africa because conflicts here differ in important dimensions from Latin America or the Middle East; for instance, with regard to the role of ethnic groups, religious tensions, as well as the production and trafficking of illegal drugs. Moreover, Africa is comprised of major recipient regions of WB as well as Chinese aid during our sample period; and offers the best quality of data regarding aid and violent and non-violent conflict.
Third, the degree of precision in our dataset allows us to link conflict events to aid in the same region. We can thus rule out the possibility that, aid in one region and conflict in another region, are coincidentally correlated with each other. Moreover, the data structure enables us to flexibly control for a wide range of potentially distorting factors through time trends, country-year, and region-specific fixed effects which hence, eliminates bias arising from unobserved conflict trends, region-specific time-invariant factors and country-level time-variant factors. We find that across a wide range of specifications that eliminate more or less potentially biasing variation, the average effect of aid on conflict is either significantly negative or indistinguishable from zero. Our IV strategy essentially emulates a difference-in-difference approach during the first stage, in that we compare the effect of donors’ budget expansion on regions with differing pre-determined probabilities. As the assumptions for this type of instrument are comparable to a simple version of a Bartik or shiftshare instruments, we carefully examine and rule out potential problems highlighted by Christian and Barrett (2017) and Goldsmith-Pinkham et al. (2018). The average null effect of aid on conflict also holds at both a smaller and a larger administrative level of aggregation, as well as when using different definitions of conflict and aid, or various different variations of the IV strategy.
Fourth, our distinctions between aid sectors, conflict actors, and types of conflict expand upon results from the existing literature. The sectoral distinction augments, for instance, the findings in Child (2018) on intersectoral differences in Afghanistan, as well as previous results on aid projects in specific sectors within specific countries (e.g., Crost et al., 2016; Berman et al., 2011). Our finding of a significant reduction in conflicts enacted by states against civilians, resulting from aid given by both the WB and China, suggests that the fear of losing aid money seems to notably affect recipient governments. This adds to the scarce evidence on differing incentives related to development projects like Crost et al. (2014), who focus on changing incentives for rebel groups.
While we observe no significant increase in citizen protest related to aid projects, the observed increases in government repression associated with Chinese aid supports the hypotheses in Kishi and Raleigh (2016), that China puts less weight on conditions regarding human rights. The results also match both within-China policies, which rank stability over political freedom, as well as previous results that Chinese aid correlates with more corruption (Isaksson and Kotsadam, 2018a) and lower unionization (Isaksson and Kotsadam, 2018b). Our own estimations using geolocalized Afrobarometer responses suggest that WB aid is linked to supporting democratic values, while the result of Chinese aid tends to be less opposition to autocratic regimes within the regions. Survey answers also suggest more political intimidation and the strong belief that one must always abide by the rule of law in those regions. Despite the correlational nature of this final piece of evidence, it matches the previous results suggesting that Chinese aid increases stability, however this stability to some extent comes at the cost of democratic development.
The paper proceeds as follows. Section 2 summarizes the existing literature and outlines proposed theories linking development finance to conflict. Section 3 explains the data and the corresponding sources and provides descriptive statistics. Section 4 presents the specification and empirical strategy. Section 5 shows and discusses the results, and Section 6 concludes.