Recent World Bank research has sparked a major debate about aid effectiveness and its implications for aid allocations. The main focus of attention has been the importance of good policy as a determinant of aid effectiveness and criterion for aid allocations. This paper briefly recaps the main arguments and evidence generated by Burnside/Collier/Dollar and their critics. It then focuses on the Collier/Dollar aid allocation models, subjecting them to a wider range of sensitivity tests to assess the robustness of their results. Finally, it analyses the relative efficiency of aid allocations over time and between donors. Aid and growth regressions
The challenges levelled at the underlying Burnside/Dollar aid-growth model are now well known, the main areas of dispute being the appropriate functional form, the identification and treatment of outliers, the choice of instruments and the method of estimation. Other analysts have found a significantly positive impact of aid on growth without recourse to an aid*policy interaction term.
The Collier/Dollar analysis, which uses the CPIA as a broader measure of the policy environment, largely confirms the Burnside/Dollar results, but finds that the overall of impact of aid is higher and its sensitivity to policy more muted: an extra one percentage point's worth of aid (as % of PPP$ GDP) would on average increase the rate of economic growth by about 0.6 percentage points in countries with good policies, 0.4% in countries with average policies, and 0.2% in countries with poor policies. Other, microeconomic evidence confirms the view that policies matter for aid effectiveness, and the simple policy message of the World Bank research - allocate more aid to poor countries pursuing good policies - has proved highly attractive and influential.
More recent research has found a range of other variables to interact significantly with aid: for example, economic vulnerability (Chauvet and Guillaumont), the actual occurrence of external shocks (Collier and Dehn), recovery from conflict (Collier and Hoeffler), and geographical factors (Dalgaard et al.). These issues, together with the extent to which aid can in fact influence the policy environment, would all affect the simple policy conclusions of the original WB research and demand further analysis. However, many donors are already responding to the earlier message by making aid allocations more performance-related.
Robustness of the Collier/Dollar aid allocation models
Earlier analysis of the Collier/Dollar aid allocation models demonstrated that poverty and per capita income criteria are actually more important than policy criteria as determinants of poverty-efficient aid allocations. But the latest Collier/Dollar paper suggests that the appropriate direction of change in aid levels for most countries is remarkably robust to a range of sensitivity tests concerning both the parameter values of the underlying aid-growth regressions and the choice of poverty measure and associated elasticity.
These are very attractive findings. However, a number of caveats can be made, concerning the limited nature of the sensitivity testing, the interpretation of the high correlation coefficients between scenario results, the variability in regional and individual country allocations under different scenarios, the validity of some of the underlying poverty and policy data, the approach to constraining allocations to India and small country bias, the comprehensiveness and currency of the dataset, and the pattern of regional progress towards the Millennium Development Goals.
This paper addresses a number of these concerns by applying a more extensive of set of sensitivity tests (25 different scenarios) to the basic Collier/Dollar model. It finds that results continue to be highly correlated with each other and with the Collier/Dollar benchmark allocation. This is encouraging. But the variation in individual and regional allocations has increased significantly: sub-Saharan Africa's poverty-efficient share ranges from 25% to 83%, the number of potential recipients varies from 15 to 29 countries, and allocations to individual countries vary by a (median) factor of 6 overall. Moreover, the desired direction of change is unambiguous for only 32 (cf. 52 in the original Collier/Dollar analysis) of the 59 countries over the full set of scenarios. The practical value of this model for aid policy-makers in redirecting allocations to specific countries is therefore somewhat diminished. The critical factors to which the Collier/Dollar model results are particularly sensitive are the extent of diminishing marginal returns, and the treatment of India and small country bias.
Changing aid efficiency - a comparison over time and across donors
The paper then explores WB claims that the marginal efficiency (ME, the number of people lifted out of poverty with an extra $1m,) of aid improved over the period 1990-97/98, with IDA aid being more efficient than ODA. This finding is being widely used to help make the case for increased aid. We first find that changes in MEs are almost entirely due to changes in aid levels and policy performance, rather than the pattern of aid allocations. Disaggregating these two effects suggests that all of the improvement in efficiency of IDA allocations and much of the apparent improvement in ODA allocations can be accounted for by falling aid volumes in the 1990s. Apparent policy improvements may also be inflated by a re-scaling of the CPIA policy score scale in 1998; adjusting for this in full would eliminate virtually all the improvement in IDA aid and half the improvement in ODA (even before adjusting for aid levels).
However, these results do need to be interpreted carefully. Maximising MEs is not the objective, and the 'loss' in improvement in MEs when adjusting for changing aid volumes is not itself a cause for concern. Moreover, there are good reasons for not adjusting the policy scale, at least not in full. This analysis simply tells us that while there has been an improvement in aid efficiency in the period 1990-97, this is primarily due to changing aid levels and policy scores rather than a change in the pattern of aid allocations. There is plenty of other evidence from both project evaluations and aid-growth regressions to show that aid effectiveness has been improving, and the case for increasing aid remains strong.
The paper then extends the analysis to assess changes over the period 1990/91-99/00 and to compare results across all donors. This confirms the findings of the earlier analysis, with MEs more than doubling overall, but, again, entirely due to falling aid volumes and rising policy scores (though the effect of rescaling the policy score is less pronounced). Multilaterals donors were and remain marginally more efficient than bilaterals, with the AfDF showing a particularly large improvement to become the most efficient, and the EC being the least efficient. Of the major bilaterals, the UK (which improved substantially over the decade), the Scandinavians, the Netherlands and Italy perform particularly well. Spain and the US are among the least efficient.
Subjecting this analysis to some sensitivity testing using different variants of the aid-growth regressions suggests that the larger the value of the aid*policy interaction term, the bigger the improvement in the 1990s. But much more substantial differences arise when the term capturing diminishing marginal returns is adjusted: the lower the degree of diminishing returns, the higher the absolute ME of aid in both 1990/91 and 1999/00, and the more modest the improvement over the decade. The relative performance of different donors is little changed by this analysis, however, with bilateral rankings generally highly correlated with the base scenario. Some further analysis is provided by comparing ratios of aid/hd going to high as against low poverty countries, and going to good as against poor policy countries.
Conclusions and policy implications
Aid clearly needs to be better focused on poor countries with large numbers of poor people. And other things being equal, more aid should go to countries with better policy and institutional environments (though this should not be overstated), to countries recovering from conflict, and to countries facing external shocks. But there remains much that we do not know, and much more research to be done, in particular on issues of functional form, diminishing marginal returns and absorptive capacity, the extent to which aid can be better used to promote policy and institutional reform, and the effectiveness of different aid instruments, and on the implications of each for aid allocations.
The Collier/Dollar aid allocation models are helpful in highlighting major anomalies in the existing pattern of aid allocations and the potential for efficiency gains, but are not as robust as hoped, they need further development to incorporate the effect of other variables with which aid appears to interact, concerns about uneven regional trends in reducing poverty, and the achievement of other MDGs.
Finally, the analysis of improvements in aid efficiency in the 1990s should be taken seriously but interpreted carefully. There is plenty of other evidence to suggest that aid effectiveness has been improving, and the case for increasing aid remains strong. But the large differentials in performance of individual donors suggest that there is substantial room for improvement in aid allocations, though our measures of efficiency need to be better informed by the further analysis suggested here.