Version classiqueVersion mobile

The Effect of Foreign Aid on Sub-national Development

Dumebi Ochem

2. Literature Review

Texte intégral

1The literature on the effectiveness of foreign aid has predominantly revolved around cross-country analyses of the impact on national-level outcomes, particularly economic growth. Despite this being a long-standing debate, no consensus has been reached on whether foreign aid actually has any impact on aggregate development outcomes, and if so, to what extent. Some studies have found that foreign aid has a robust and positive effect on economic growth in recipient countries, while others obtained a significant effect only in countries with certain characteristics, such as good governance and institutions.

  • 1 Examples of these include: binary variables indicating whether a recipient country is a “friend” of (...)

2Clemens et al. (2012) argue that the main reasons behind the large divergence in the literature are contingent on the timing of the aid effects and the methodology used. They suggest that many studies erroneously measured the contemporaneous effect of aid on growth, when in fact most aid-funded projects would only have a discernible impact after a time lag. Such results would therefore suffer from simultaneity bias. With regards to the methodology, the majority of research papers have relied on instrumental variables as an identification strategy for foreign aid. These instruments have consisted of some measure of political ties to donors, lagged aid flows, or population size of recipient countries, all of which could arguably pose concerns about both their relevance and validity.1

3In order to address these endogeneity issues, Clemens et al. (2012) re-analyse the results from three of the most influential studies in the aid-growth literature, namely Boone (1996), Burnside and Dollar (2000; 2004), and Rajan and Subramanian (2008). The authors adapt the methodologies in each of the papers by including a time lag in the specification and by first-differencing to remove time-invariant omitted variables. In addition, they only consider those portions of aid that would be most expected to influence growth in receiving countries (e.g. budget support and project aid for real sector investments in infrastructure).

4Their findings demonstrate aid flows to be systematically associated with modest but positive growth, with evidence of a non-linear relationship indicating potential limits to the aggregate aid effects. However, a replication of this study by Roodman (2015) raises further questions, as he finds that the two strategies relating to lagging and first-differencing fail to remove contemporaneous endogeneity. In fact, once aid is lagged by two periods, the results suggest a zero or negative Granger causation flowing from aid to growth.

5On a different note, researchers are now paying more attention to the effects of foreign aid at the sub-national level, thanks to the increased availability of georeferenced data on aid and relevant outcome variables. Many argue that the non-robust results in the macro literature are dependent on the spatial unit of investigation, as the effects of aid would be too small and localized to have an impact on aggregate outcomes (Kotsadam et al., 2018; Bitzer and Gören, 2018). In addition, the lack of suitable country-level controls, sample selection bias and measurement error in both the outcome and aid variables all aggravate concerns of endogeneity.

6The work by Dreher & Lohmann (2015) introduced a new identification strategy to estimate the causal effects of aid on growth at the sub-national level, which has grown more popular in subsequent research. Using a sample of 478 first-order and circa 8,400 second-order administrative regions (ADM1 and ADM2, respectively) from 21 countries between 2000 and 2011, the authors instrument aid flows with a binary variable indicating whether a country has crossed the income threshold for eligibility to the World Bank’s concessional aid, interacted with a region’s probability of receiving aid. When controlling for the levels of the interacted variables, the authors argue that this instrument satisfies the exclusion restriction, and is akin to a difference-in-differences estimator.

7Overall, their baseline OLS results show that growth increases with aid in ADM2 regions, but once endogeneity is accounted for, the effects become completely insignificant. However, this strategy poses some concerns towards external validity, as the analysis is only limited to countries that have crossed the IDA income threshold during the sample period.

8Bitzer & Gören (2018) also study the effects of aid on growth of night-time lights, but instead use equally sized and arbitrarily constructed grid cells as the spatial unit of observation. One of the main advantages of using such an approach is the ability to create potential counterfactual observations, allowing one to identify a causal relationship between aid and growth.

9Using a full-set of grid and country-year fixed effects, the authors estimate a night-time lights logarithmic growth model using four different estimators: pooled OLS, Fixed Effects, difference GMM and system GMM. Their baseline estimates suggest a positive and significant relationship between foreign aid and economic activity, which remains robust across different specifications. They also find that projects targeted towards water and sanitation, health and infrastructure have the strongest effect on nightlight growth, and that short-term projects are more effective than long-term projects. However, all estimation strategies raise doubts concerning the reliability of these results, due to the obvious inconsistency issues in estimating a dynamic model using OLS, and the poor performance of the GMM instruments revealed by various diagnostic tests.

10Moving away from economic growth, Kotsadam et al. (2018) is arguably the first systematic attempt to study the effects of official development assistance on infant mortality at the sub-national level. Using a difference-in-differences analysis, the authors compare infant mortality rates in regions in close proximity to aid projects before and after implementation, to regions located further away. The results support their hypothesis that children born in areas close to aid projects have a higher probability of survival, with an even stronger relationship within less privileged groups.

11On the other hand, several studies remain critical of foreign aid, particularly from multilateral donors, with regards to whether or not it actually achieves its principal objective of helping the poorest. Briggs (2018, p. 134) reasons that “aid cannot help the poor unless it both works and reaches where the poor live”, a point all the more relevant when one considers the high degree of sub-national geographical inequality in Africa. Using a spatially gridded dataset, the author examines whether aid from the World Bank and African Development Bank is targeted towards relatively poorer areas within African countries, for the period between 2009 and 2010.

  • 2 The author used a linear probability and a logistic model for the regressions involving a binary ai (...)

12Briggs (2018) stresses the fundamentally descriptive nature of the paper’s research question, and thus only includes a poverty measure and grid cell-level population as covariates in all estimations, to avoid obscuring the spatial relationship between aid and poverty. He operationalises the dependent variable using three indicators for grid cell-level aid, and uses five separate proxies to describe the poverty level. These are: night-time luminosity, travel time to nearest city with at least 50,000 inhabitants, distance to the national capital, prevalence of child malnutrition and the infant mortality rate. The results suggest an anti-poverty bias in the within-country allocation of aid, as grid cells with more nightlight intensity, shorter distances to the capital and shorter travel times receive a greater amount of aid. These findings are consistent with previous work in the literature (Dreher et al., 2015; Nunnenkamp et al., 2017), and remain robust across different specifications and estimation strategies.2

13Nevertheless, there are several limitations to the paper’s methodology that may challenge the strength of the results. For instance, the author does not attempt to identify any causal mechanism in the model and only estimates correlations between aid and poverty at the sub-national level, therefore making his conclusions of anti-poor targeting somewhat reductive. In addition, the time period of observation is very small and does not account for already existing projects, thus potentially biasing the results. In fact, the dataset would suggest that Cape Verde, Somalia and Zimbabwe receive absolutely no aid, which is highly unlikely given that all three are eligible for World Bank borrowing.

14Furthermore, the investigation fails to acknowledge potential spill-overs from neighbouring regions within the same country. For instance, this mechanism would apply in the case of aid projects financing the construction of “connective infrastructure”, such as roads, bridges and railways, which are believed to disperse economic activity (Bluhm et al., 2018). Therefore, it is worth considering that a given grid cell may indirectly benefit from a project implemented in a neighbouring grid.

15This paper aims to expand on the sub-national strand of the aid effectiveness literature and address the aforementioned challenges through two empirical contributions. First, it examines the effects of aid at a high degree of spatial resolution, across 10,600 grid cells over a sample period of almost 20 years. This approach, which has not been used in the literature, should tackle endogeneity concerns more effectively. Second, it highlights the heterogeneous effects of foreign aid at different levels of development within countries. This concept has often been touched upon in previous studies, but to the best of my knowledge, it is yet to be investigated in this context.


1 Examples of these include: binary variables indicating whether a recipient country is a “friend” of the US or OPEC (i.e. receiving more than 1% of the donor’s total aid budget); indicators for common language and/or former colonial relationship; lagged arms imports as a share of total imports; interaction terms between an initial level of income or population and a policy measure (e.g. indices of inflation, budget balance, openness to trade etc.).

2 The author used a linear probability and a logistic model for the regressions involving a binary aid indicator; for the count analysis (number of projects), the negative binomial and Poisson models were used; finally, OLS was used for the analysis using the total value of aid.


Rechercher dans OpenEdition Search

Vous allez être redirigé vers OpenEdition Search