An Assessment of the Effects of Norwegian Development Assistance on Poverty Reduction and Conflict Prevention 1This report was prepared by Paul Collier and David Dollar of the Development Research Group of the World Bank, with useful inputs from Shanta Devarajan, Giuseppe Iarossi, and Dennis Tao.. The views expressed are those of the authors and do not necessarily reflect official views of the World Bank or the Government of Norway.

A Report Commissioned by the Norwegian Ministry of Foreign Affairs
and prepared by Paul Collier and David Dollar,
Development Research Group of the World Bank

The document in pdf-format.

February 2001

1. Overview

Recent research into aid effectiveness has established that development assistance can be a powerful force for poverty reduction in the developing world. It is most effective in this objective when it is targeted to countries that are very poor, and – among these poor countries – is focused on ones that have made substantial progress reforming economic institutions and policies. As for spurring policy change, aid normally does not play a leading role, though it can provide critical support to governments and societies in which there is a real movement for change ("ownership" of reform).

Using this framework, we are able to estimate the allocation of total development assistance that is "poverty efficient" – that is, gets the maximum poverty reduction impact from a given amount of assistance. We view this allocation as a benchmark that is a starting point for analysis, not as a rule to be followed mechanically. In using the framework to assess the impact of Norwegian assistance, our main findings are:

• Overall, the poverty efficiency of Norwegian assistance is high compared to typical ODA. (We can actually provide a specific numerical estimate: a dollar of aid from Norway has about 50% more poverty impact than a dollar of aid from the average DAC donor.) Norwegian assistance is relatively effective because it is sharply focused on very poor countries, and many of its priority countries have reasonably good policies and high aid effectiveness.
• The impact of Norwegian assistance could be enhanced further by some reallocation of resources among recipient countries. Of its twelve priority countries, Sri Lanka and Zimbabwe stand out as ones with relatively low aid effectiveness: Sri Lanka because it is not that poor a country, and Zimbabwe because it has very poor policies. Also, Norway tends to "over-finance" some moderate policy countries such as Tanzania and Zambia. Overall poverty impact would be increased if aid were reallocated from any of these countries to the priority countries with higher effectiveness: Uganda, Ethiopia, and Bangladesh. (One important caveat, however, is that our assessment of Ethiopia’s policies comes from 1999, before the war with Eritrea escalated, and it is reasonable to mark down the policy ranking because of this issue that obviously worsens the environment for development and poverty reduction.) How exactly Norway might want to reallocate resources is more a political question than a scientific one: but as an illustration we analyze a modest reallocation among priority countries that would increase the total poverty impact by 20%.
• Beyond the priority countries and aid given for humanitarian purposes into crisis environments, Norway gives small amounts of assistance to 63 other developing countries. A priori it seems unlikely that spreading assistance so thinly is likely to have the maximum impact. In our analysis we find that most of the recipients on this list are not high aid effectiveness countries. There are a few notable exceptions (Ghana, India, Vietnam). We estimate that Norway’s assistance to this list of 63 countries has the same mediocre poverty impact as typical ODA. For the resources in this category, the poverty impact could be doubled by focusing on a smaller number of countries (either the ones on this non-priority list that have high effectiveness, or ones from the priority list that are under-funded).
• We introduce political economy considerations about aid and policy reform, and ask whether these considerations would alter any of the judgments. In most cases the political economy considerations reinforce the judgments: in countries such as Uganda, Ethiopia, Eritrea, and Bangladesh, assistance should have a direct poverty impact and increase the likelihood that reform programs will continue. In Zimbabwe, on the other hand, not only is aid likely to have little immediate impact, but the long-standing autocratic regime there is a very poor candidate for reform. The cases that require more consideration are Sri Lanka and South Africa. Neither country is especially poor, but each has a relatively new government that is trying to make changes.
• In the final section of the report we consider the objective of conflict prevention instead of poverty reduction. We apply a new model which estimates the risk of large-scale civil conflict. Aid can be effective in reducing conflict risk, but as with poverty reduction, its effectiveness depends upon the policy environment. Hence, as with poverty reduction, effective Norwegian intervention can only occur in environments where policy is already reasonable. While with the objective of poverty reduction a donor should target aid on those good-policy countries which have high poverty, when the objective is conflict prevention, aid should be targeted on those good-policy countries with a high risk of conflict. Two of Norway’s current high-priority countries fall into this category and this potentially reinforces the case for their priority status.

2. General Analysis of the Allocation of Norwegian Assistance

Our analysis of the poverty-efficiency of Norwegian assistance is based on the research in Assessing Aid and the background papers for the report. This and other research established four important points about aid, growth, and poverty reduction:

• The impact of aid on growth depends on the quality of economic institutions and policy in the recipient country;
• Donors in general do not have a large effect on these institutions and policies (though there are important exceptions that we will treat below);
• There are diminishing returns to aid, so that in any one year there is a limit to how much aid even a "good performer" can absorb; and
• There is a close link between growth and poverty reduction in developing countries.

In this section we are going to assume that donors have no influence on recipient country policies at all, and examine how aid should be allocated in this context. This should be viewed as an initial benchmark. In later sections we will discuss how one might want to deviate from this benchmark in specific country cases where there are good indicators of impending policy reform.

If we take recipient country policy as given, how should we allocate foreign aid in order to have the maximum effect on poverty reduction? The research results noted above provide the answer: aid should be allocated on the basis of how poor countries are and on the quality of their economic institutions and policies.

As an illustration, Figure 1 provides a scatter plot of 130 developing countries with information from the late 1990s on the extent of poverty (share of the population living on less than $2 per day) and on the quality of institutions and policy. The specific measure used here is the World Bank’s Country Policy and Institutional Assessment, which is described in detail in Box 1.

In general, aid is going to be more effective in the countries in quadrant I. The relatively good policy here means that assistance will be used effectively. The high poverty in these countries means that growth spurred by aid will have a large effect on poverty reduction. In quadrant IV, aid will also be effective at promoting growth, but it is not efficient to give a lot of aid to these countries because poverty is relatively low. Chile or Thailand would be examples of countries with good policies but relatively low poverty.

In quadrant II there are countries with severe poverty. However, the weak policies mean that aid is not that effective in generating growth and poverty reduction. The countries in quadrant III have poor policies and relatively low poverty (these are mostly transition economies such as Russia or Ukraine).

Figure 1 is useful as a heuristic device to emphasize that among poor countries there are large differences in economic policy and that aid will be more effective at reducing poverty in the countries that are in the upper right (high poverty, good policy). However, our model of efficient aid does not say that the moment you cross the line between quadrants II and I that aid suddenly becomes effective. Rather, the model says that aid becomes more effective as you move to the right in the figure. We have devised a specific algorithm for allocating aid to have the maximum effect on poverty, an algorithm in which the amount of aid that a country receives increases with the quality of policy and also increases with the extent of poverty (Collier and Dollar, 2000).

For this exercise we calculated a "poverty-efficient" allocation of the world’s aid for 1998 (Table 1). (The year 1998 is the most recent one for which we have the data on the allocation of total world aid.) This provides a useful benchmark for looking at the efficiency of Norwegian assistance. Figure 2 shows how our "poverty-efficient" aid is allocated across the countries in the four quadrants. (Because China and India are so large and not very relevant for the Norwegian aid program, we left them out of this calculation.) In particular, 68% of assistance goes to the "good policy, high poverty countries." Note that the countries in the "high poverty, poor policy" group get 28% of the allocation. This drives home the point that our recommendation is not to give zero assistance to the poor policy countries. The main thrust of our analysis is that donors should be giving more assistance to the good policy countries than to the poor policy ones. Up through the mid-1990s, donors were not doing this at all. In the past

two years, however, there has been a notable change in donor behavior, in the direction that we have advocated. Some poor policy countries such as Kenya have seen clear cuts in their aid receipts. Nevertheless, it can be seen in Figure 2 that the allocation of total ODA in 1998 was far from the "poverty-efficient" allocation. Countries in quadrant I received more assistance that those in quadrant II. However, donors in the aggregate continue to give a lot of assistance to middle-income countries in which aid is not likely to have much impact on poverty.

We make the same calculation in Figure 2 for Norway’s 1999 aid. Norwegian assistance is more efficient that average ODA, by our criteria. Whereas 38% of ODA is allocated to the quadrant I countries, 48% of Norwegian aid goes to this group. About 21% of Norwegian assistance goes to the less poor countries, compared to 33% of total ODA. As a final reference point, the figure also includes the allocation of the World Bank’s IDA concessional resources. There is slightly more IDA going to the quadrant I countries (50% of the total), but in general the allocation of IDA and the allocation of Norway’s aid are similar. Both are more efficient that average ODA, but still somewhat far from the "poverty-efficient" allocation.

A second approach to analyzing the general efficiency of Norwegian aid is through regression analysis. The poverty efficient model says that aid should be allocated on the basis of a country’s poverty and its policies. In appendix table 1 we regress Norwegian aid on per capita GDP, population, and the CPIA index. (Here we use all the data from the 1990s and create a panel.) After controlling for per capita income and population, there is a clear relationship between how much aid a country gets from Norway and its policy as measured by the CPIA. This relationship is summarized in Figure 3, which shows how much additional Norwegian assistance is associated with a one point increase in the CPIA. (One point is a fairly large change in the index; the standard deviation of the index is 0.7. One point is roughly the difference between Kenya’s policies and Uganda’s policies.) One point better on the CPIA scale is associated with 78% more Norwegian assistance. The t-statistic on that coefficient is 4.92, so that the relationship is quite a significant one. It is interesting to carry out the same exercise for some of the different components of Norwegian assistance. The relationship with the CPIA is particularly strong for program aid (165% more for a one-point change in the index) and also for technical assistance (162% more). For project aid, one point on the CPIA is associated with 37% more aid, though that number is not statistically different from zero. These findings suggest that Norway uses its program aid and technical assistance to support poor countries that have reformed, while continuing to pursue projects in all types of policy environments.

We investigated whether this relationship between Norway’s aid and the CPIA changed during the 1990s, and found that it has been a stable feature of Norwegian assistance throughout the decade.

The evidence so far is that Norwegian assistance is more effective that ODA in general in promoting poverty reduction. A natural question to ask, then, is how much more effective? In our framework, we can estimate the impact of an additional dollar (or Kroner) of aid on poverty in each developing country. Poverty efficiency requires that aid be allocated so that these marginal impacts are equalized. (Otherwise, there would be efficiency gain from shifting aid from where its effect is low to where its effect is high.) In the real world, total ODA is far from efficient, so that the marginal effect of aid varies to a considerable extent across countries.

One way to estimate the efficiency of different donors’ allocations of aid is through the following thought experiment. Suppose that there were an additional million dollars of aid, estimate its impact on poverty if it were allocated

• Proportional to Norway’s 1999 aid
• Proportional to 1998 IDA
• Proportional to 1998 ODA
• Proportional to Norway’s 1990 aid.

Basically, we are asking what would happen if each of these aid programs had a small proportional increase. The answers tell us how well different aid programs are targeted to countries in which aid is effective.

We estimate that an additional million dollars of Norwegian aid would lift 298 people permanently out of poverty. One has to recognize that a specific point estimate like this has quite a bit of uncertainty around it. Nevertheless, it is a useful, broad estimate that is quite plausible. It says that it takes an investment of around $3,000 to permanently lift someone out of poverty. Figure 4 shows how this marginal efficiency of Norway’s 1999 assistance compares to other donor allocations. Norway’s aid is about 50% more productive – in terms of poverty reduction – than average ODA. It is not quite as efficient as 1998 IDA, which has an estimated marginal productivity of 336 people per million dollars. Furthermore, the efficiency of Norway’s assistance improved over the decade, from 211 people per million dollars in 1990. (In other work we estimated that the marginal productivity of ODA overall in 1990 was quite low, about 100 people per million dollars.) Thus, Norway’s assistance is more poverty efficient than ODA in general and has improved over the decade. As noted, the specific point estimates should be treated with some caution. However, the basic finding that Norwegian aid is more productive than average ODA is quite robust.

3. Potential Poverty Gains from Reallocating Aid among Norwegian Aid Recipients

The basic message that comes through in the previous section of this report is that Norwegian aid is relatively efficient at its objective of poverty reduction. However, the word "relatively" is important. One of the main findings of research into aid is that overall ODA is not very efficient. It is often given for political, strategic, or commercial reasons that have nothing to do with poverty reduction. So, while Norwegian aid is considerably better than much other ODA, it is reasonable to inquire whether it could be made more efficient through plausible reallocations. In this section we are going to focus first on Norway’s twelve priority cooperation partners and examine what gains could be achieved by reallocating aid volumes among these countries. Then we will broaden the analysis to include all of the countries to which Norway gives development-oriented assistance.

Table 2 lists Norway’s twelve priority development partners in descending order of aid receipts in 1999. The table also includes an estimate of poverty (population share living on less than $2 per day) and an assessment of policy, in one of four categories – very good, good, moderate, and poor. The total amount of assistance to these countries in 1999 was Kroner 2.1 billion (about $300 million). The table also shows a number of counterfactual reallocations of Norwegian aid, among these countries, holding the total amount constant. Reallocation (1) is proportional to the "poverty-efficient" allocation of aid that we have calculated.

In general, very populous countries such as Bangladesh and Ethiopia are under-funded in terms of aid. Thus, in the poverty-efficient allocation, Bangladesh would get about half the total, where in reality it gets about 13%. There are political reasons why donors may not want to concentrate too much assistance in any one country. In the cases of Bangladesh and Ethiopia, there are other factors as well. Reallocating a large amount of aid to Bangladesh would essentially mean taking it from African countries. And, in the case of Ethiopia, the recent war with Eritrea is another consideration. The policy score used here is for 1999. Realistically, the war is a factor that increases military expenditure as a proportion of the budget, thereby probably reducing the poverty-efficiency of aid. Further, the recent war raises political considerations beyond the scope of our analysis but of legitimate concern to donors.

For these reasons we consider two other reallocations. In Reallocation (2) we constrain Bangladesh to its actual Norwegian aid receipts, and otherwise reallocate proportional to poverty-efficient aid. In Reallocation (3) we also constrain Ethiopia. Reallocation (3) is probably the most politically realistic. This reallocation would halt the aid to Zimbabwe and Sri Lanka, for somewhat different reasons. Sri Lanka has pretty good policy, but is simply not a very poor country. Hence it gets no aid in our poverty-efficient allocation. Zimbabwe is poorer than Sri Lanka, though not as poor as many other African countries. It has poor policy, and that is the primary reason that it gets no aid in the optimal allocation.

Reallocation (3) would also take modest amounts of assistance away from Tanzania, Mozambique, Nicaragua, and Zambia – all countries with moderate quality policies and large amounts of assistance from Norway. Uganda is the main country that gets additional aid in Reallocation (3), because it is a poor country with very good policy. Nepal also gets more aid, despite its poor policy; this emphasizes the fact our model does allocate some aid to truly poor countries, even with poor policies. Norway in 1999 gave a very small amount of aid to Nepal, so aid would have to be increased to be proportional to poverty-efficient aid.

What would be gained from reallocation? Recall that our estimate of the marginal impact of Norwegian aid was that it lifts 298 people out of poverty per million dollars. The figure for this subset of countries is somewhat higher – 340 people per million dollars – because in general this is a "good" list of countries: it includes several countries where the effectiveness of aid is quite high. We estimate that reallocation (3) would increase poverty reduction to 410 people per million dollars. In other words, the roughly $300 million in Norwegian assistance to the priority countries lifts about 100,000 people per year out of poverty; the same volume of aid could lift about 20% more people out of poverty if it were allocated more efficiently. This gain arises from shifting funds from countries in which there is little impact of aid (Zimbabwe) to countries where there is a large impact (Uganda).

While it may be politically difficult to reallocate a large amount of aid to Ethiopia or Bangladesh, we should nevertheless report our estimates of the large gains that would arise from such reallocation. Reallocations (1) and (2) each yield estimated additional poverty reduction of about 57,000 people – raising the aid productivity by about 50% from its actual 1999 level. Keep in mind the caveat that the assessment of the environment in Ethiopia comes from 1999 and may not fully reflect the impact of war. Still, these estimates are indicative of what could be achieved if donors were more willing to give large amounts of assistance to highly populous countries.

Besides the twelve priority countries, Norway gives aid to a large number of other countries. Some of this is humanitarian assistance to countries involved in conflict, and we have been directed in the Terms of Reference for this project to separate that from our analysis, as our model of poverty efficient aid is not really relevant to humanitarian assistance given into a crisis environment. Appendix Table 2 lists these countries and their assistance from Norway in 1999. Leaving them aside, Norway gave development assistance to 63 other developing countries in 1999. Together with the twelve priority countries, this makes a total of 75 aid recipients – quite a large number given that Norway itself is not a very large country. Most of these 75 aid recipients received less than Kroner 30 million in 1999. Table 3 lists the 63 non-priority countries receiving aid in 1999 and the amounts that they received. We have calculated for each the estimated marginal efficiency of a million dollars of aid in reducing poverty, and have listed the countries in descending order of aid efficiency. Some of the countries near the top are poor-policy countries where donors have cut total aid back quite substantially (for example, Kenya or Pakistan). Three countries are good policy countries that receive modest amounts of support – Ghana, India, and Vietnam.

For the majority of countries on this list, the marginal efficiency of aid is below the Norwegian average of about 300 people per million dollars, either because the country has poor policies and/or is simply not that poor. Any money reallocated from countries low on this list to ones near the top (or to the priority countries other than Sri Lanka or Zimbabwe) would increase the net poverty reduction effect of Norwegian assistance. For this group as a whole, the assistance in 1999 was Kroner 1.2 billion (about $170 million); because it is spread among high aid effectiveness and low aid effectiveness countries, the average productivity of this assistance was an estimated 208 people lifted out of poverty per million dollars. That productivity would essentially be doubled by concentrating the assistance in countries such as Ghana, India, or Vietnam.

In summary, both for the priority countries and for the non-priority countries, Norwegian aid is spread among countries that have quite different aid effectiveness because of their differing levels of poverty and the fact that some have better policies than others. The overall impact of Norwegian aid could be increased significantly if the aid were concentrated on a smaller number of countries, focusing especially on the ones in which aid effectiveness is high. Reallocation (3) above concentrates the aid for priority countries on ten countries. If from the non-priority list, aid were concentrated in ten to twenty countries that are in the top half of this ranking in terms of aid effectiveness, the overall productivity of Norwegian aid could be lifted to the range of about 400 people per million dollars, from the current level of about 300.

4. Political Economy of Aid and Reform

We emphasized that our model of poverty efficient aid takes as given the quality of institutions and policies in recipient countries. This is clearly an extreme assumption, as much assistance is aimed at improving institutions and policies. Nevertheless, much research inside and outside of the World Bank has established that donors in general have greatly exaggerated their influence over policy. Worse, the wrong type of aid in the wrong environment can actually make policy worse. Meaningful, sustained policy reform requires deep commitment and ownership from the societies and governments in question. So, it is useful for donors to begin by looking at aid effectiveness on the assumption that they have no influence on policy at all – which is what we have done in the previous sections. This provides an initial benchmark allocation of aid from which a rational donor might want to deviate for a variety of reasons. One of those reasons is that in some cases assistance can help build and sustain successful reform programs. Hence, in this section we will look at some political economy considerations that might alter the judgments made above.

A useful starting point is findings from recent research into aid and the political economy of reform – both cross-country econometric analysis and the case studies from the Aid and Reform in Africa project. Taking the cross-country work first: Dollar and Svensson (2000) look at 220 economic reform programs supported by the IMF and World Bank, mostly carried out in the 1980s and the very early 1990s, and ask: are there common features of successful programs and unsuccessful programs? The measure of success here comes from the World Bank’s Operations Evaluation Department (OED), an outcome assessment of whether the targeted policy measures were carried out. (The paper shows that a successful outcome rating is highly correlated with better economic management – lower inflation, more sustainable fiscal situation -- several years after reform, which suggests that the OED measure is a good one.) In their sample, about one-third of the reform programs had failed.

What Dollar and Svensson find is that the outcome of reform programs can be predicted quite well by information on the recipient country's characteristics that is available before the reform starts. For example, the success rate for new governments is far higher than the rate for governments in power for a long time. On top of that, the success rate was higher for democratically elected governments. These two findings are put in terms of the probability of success of a reform program in Figure 5: a new, democratically elected government has a 95%

probability of success, compared to 67% for an authoritarian government in power for 12 years. This result makes intuitive sense. Countries that have poor policies over significant periods of time develop vested interests who benefit from the policies (distorted exchange and trade regimes, inefficient state enterprises, corruption more generally), and it is unlikely that an entrenched government is going to take on those vested interests.

One of the positive findings in the Dollar and Svensson paper is that, after controlling for these characteristics, the success rate for low-income countries and middle-income ones is the same, as is the success rate in different regions. In other words, the low success rate of reform programs in poor countries or in certain regions (Africa) can be explained to a large extent by characteristics that can change.

In another study, Alesina and Dollar (2000) look at the relationship in general between official finance and policy reform. One aspect of their paper is quite relevant to aid and reform: they ask whether or not there is any tendency for increases in finance or decreases in finance to lead policy change. This is important because it gets at the timing of assistance and policy change. It is possible that even a failed adjustment program sets the stage for further policy reform, and success at a later date. If that were true, then donors should not be too concerned about providing program aid in low probability environments. What Alesina and Dollar find, however, is that there is no tendency for surges in finance to lead policy reform. Specifically, they find more than 100 episodes in which there are "surges" in finance (defined as a change of at least one standard deviation relative to the country's own history of financing). Many of these are associated with Bank-Fund supported adjustment programs. In only a handful of cases does policy significantly improve in the following three to five years, and in just as many cases policy significantly worsens. The most striking fact here is that in general policy is quite persistent. Large changes in policy are the exception, not the rule.

Alesina and Dollar also look at the converse question: are large changes in policy typically followed by surges in financing? They find that donors have responded quickly to democratization episodes (political reform), but that they have not responded consistently to large economic policy changes with significant increases in finance. That may seem surprising since in many of the well-known reform cases there are adjustment loans bringing finance. But when we look at the overall pattern of donor behavior, it has not been the case that changes in policy have been met by major changes in financing.

Finally, we should also mention the Burnside-Dollar paper, "Aid, Policies, and Growth," (2000). That paper is primarily about the effect of aid on growth (and it finds that the effect of aid on growth increases with the quality of policy). But it also considers the question of whether the amount of aid that countries received affected their policy. They found no evidence that the amount of aid systematically affected policy. In some cases, however, that finding has been misunderstood. First, the fact that there is no systematic relationship does not mean that aid could not have influenced policy in specific cases. If in some cases aid supported policy reform and in other cases it retarded reform, then what one would find in a large sample is no systematic relationship. Second, research is always about the past, and in this case the research covers a period in which donors overall were not putting much weight on economic policy. That is true in a cross-section of countries: ones with better policy, after controlling for factors such as poverty level and population, did not receive more aid. And it is true in a time series: when a typical country reformed, it did not receive a significant increase in finance. Given that pattern of donor behavior, it cannot come as a surprise that there is no systematic evidence that more aid has led to better policy. And if donors change their behavior, then the past results are not an accurate predictor of what will happen.

While the econometric studies are useful for summarizing regularities in the data, they cannot have the richness of institutional and historical detail that one gets in a good case study. The World Bank research department followed up the econometric work described above with the project, "Aid and Reform in Africa," which carried out case studies of DR Congo, Cote d'Ivoire, Ethiopia, Ghana, Kenya, Mali, Nigeria, Tanzania, Uganda, and Zambia. The group is diverse in terms of policy reform, with Ghana and Uganda well known as relatively successful cases; Congo and Nigeria with very poor policies up through the mid-1990s; and the other countries in between. This project received financial support from a range of donor countries (France, Germany, the Netherlands, Norway, Sweden, and Switzerland). It was innovative in that all of the case studies involved the participation of African researchers and policy-makers. The risk with case studies is that there may not emerge any clear generalizations, but in this case there was consensus on a range of issues concerning aid and policy reform.

First, the studies were clear that aid cannot bring about sustained policy changes to which the government is not committed. Zambia under the Kaunda regime is probably the best example of the impotence of policy-based assistance in the face of a non-reforming government. By objective measures, policy got continually worse in Zambia throughout most of Kaunda's tenure (Figure 6). During the period covered here, there were 18 adjustment loans from the IMF and the World Bank. In the case of the Bank, all of the loans fully disbursed, and yet there was no policy improvement. Partly as result of this adjustment lending, the total volume of assistance to Zambia continued to rise. Worse still, the Zambia, Tanzania, and Kenya case studies all argue that the large amount of finance to poor policy governments actually sustained bad policy:

Does aid sometimes help defer reforms? It is probable that the heavy infusion of budgetary support which Kenya received during the 1980s assisted the government in financing the cost of civil service overmanning and public enterprise inefficiencies, thus permitting the government to defer reforms in these areas until the 1990s. (Kenya case study, p. 27)

Initially aid probably delayed reforms by helping to finance schemes that would have been wholly unviable without aid backing… (Tanzania case study, p. 44)

Much of this assistance came in the form of adjustment loans. Without government commitment, the conditionality did not successfully lead to policy change:

The reform experience in Zambia reiterates the importance of local ownership of the reform process: Conditionality is a relatively impotent tool in terms of bringing about policy change unless the reform measures are supported by the political leadership. (Zambia case study, p. 17)

We would argue that at times of severe economic crisis, as in 1980-82 and 1993, the government’s need for financial support was desperate and the promise of support did induce the government to come to agreement on far-reaching reform programs. However, these agreements were not always implemented. Sometimes the probability of successful implementation was low from the outset. Other times the lenders or donors may have aligned themselves with well-intentioned technocrats who wished to achieve the results contracted for but lacked the political support to do so. (Kenya case study, p. 27)

On a more positive note, the case studies of both Ghana and Uganda argue that foreign assistance helped with policy reform, and that specifically adjustment lending from IDA was helpful. There are several things different about Ghana and Uganda, compared to the other countries in the study. Both countries received very small amounts of aid during a period of poor policy in which their regimes were estranged from western governments. Regime changes led to new governments that were committed to making things better, but which were not initially committed to market-oriented reforms. (Incidentally, Vietnam -- another relatively successful reformer -- fits this pattern exactly: estranged from the west, new leaders came to power in the late-1980s searching for a new approach.)

Figure 7 shows the pattern of aid and policy for the three successful low-income reformers, Ghana, Uganda, and Vietnam: small amounts of assistance (and no adjustment loans) when they had bad policy; and then sharply increasing finance associated with adjustment loans as they put in place major reforms. The Ghana and Uganda case studies argue that this pattern of assistance had several advantages. First, during the poor policy period assistance focused on policy dialogue and technical assistance. In this period, the governments were searching for their basic policy orientation, and they did some experimentation. It was useful in this case not to have adjustment loans tying the government to plans -- the government was still searching for its plans. Low level assistance without conditionality can help that learning process (which involves studying other countries and trial and error). Once Ghana and Uganda moved decisively to put policy reforms in place, it was important to have the increasing finance, which helped bring forth a strong response from the reform program: it is important that citizens see the benefits of reform quickly, and aid increasing in lock-step with policy improvements helps in this way:

[B]alance of Payments support "provided the government with the breathing space it required to contain domestic opposition to market-based reforms… [It] allowed imports that helped fill the shelves of supermarkets and other traders. The filled shelves provided a psychologically-induced breather for the government because …people saw this as a sign of better things to come." (Ghana case study, p. 35)

When countries actually reform, finance increases the benefits of those reforms. That is, the growth impact of a particular improvement in policy is enhanced by the flow of aid. There are two reasons for this. Aid increases confidence in the reform program and calls forth greater private investment. Also, it enables the government to provide public services that are complimentary to private investment. By increasing the benefits of reform, aid enhances the likelihood that they will be sustained. As the Ghana study notes,

Coming back to politics, ultimately economic reform was only politically sustainable because some results emerged quickly. (Ghana case study, p 19)

It is important to link this point back with the Burnside-Dollar finding that aid did not systematically affect policy. The two findings are not inconsistent. The Ghana and Uganda case studies argue that, because finance increased as policy improved, this financing help sustain the reforms (aid contributed to good policy). But, up through the mid-1990s, this pattern of aid giving was not typical donor behavior. So, when Burnside and Dollar ask, has aid typically supported good policies, the answer is "no." Another way to look at this issue, is that the positive role of aid in supporting reform in Ghana and Uganda was offset by its negative role in Kenya, Tanzania, and Zambia, so that on balance aid did not systematically lead to good policy.

5. Political Economy Considerations in Norway’s Aid Recipients

The different findings from the political economy research can be summarized in Figure 8. If we go back twenty years, most of the countries that are low-income today had weak policies at that time. Poor policies produced poor results, and eventually economic and political crisis that has spurred attempts at reform. Virtually every developing country has initiated structural reforms one or more times in the past decade. Some countries have achieved sustained success, while others have attained limited success or backsliding. We can identify four phases in the figure. Phase I is pre-reform, of which there are only a few examples left in the developing world (Cuba would be an example). Phase II – which in reality is often spurred by crisis – is an intense period of attempted reform. In some cases this leads to sustained good policy (Phase III). But in other cases serious reform fails (Phase IV).

To support reform, donors need to concentrate their financial assistance on countries in Phase II or Phase III. In Phase I, history has shown that large amounts of finance will typically stave off reform (dialogue and technical assistance are different matters). In Phase II, dialogue, technical assistance, and increasing finance as policies improve are all useful. In Phase III, finance is highly effective at promoting growth and reducing poverty – so that it directly meets poverty reduction objectives and helps sustain good policy by strengthening the impact of reform. The countries that fail to sustain reform and slide backwards (Phase IV) are essentially returning to a state like Phase I.

Now, placing actual countries into this framework requires some judgment. But the guides that we have are the actual state of policy and political economy factors such as how long the executive has been in power and how democratic the country is. Table 4 lists Norway’s twelve priority countries and also eight others that are major recipients of Norway’s assistance. For these twenty countries, the table shows how Norway’s assistance has evolved over the 1990s, an assessment of current policy, and several political economy indicators (indices of voice or democracy, rule of law, and graft, and the length of time that the executive has been in power). Based on this information, we have tentatively placed the countries in the different phases as follows:

Phase III: Bangladesh, India, Ethiopia, Eritrea, Ghana, South Africa, Uganda, Vietnam

Phase II: Guatemala, Malawi, Mali, Sri Lanka, Tanzania

Phase IV: Angola, Kenya, Mozambique, Nepal, Nicaragua, Zambia, Zimbabwe

How do we combine this information with the results from the earlier analysis in which we took policies as given? First, the countries listed in Phase III are all ones previously identified as candidates for substantial amounts of aid on the basis of their being low-income countries with pretty good policies. This is the environment in which aid has a large effect on poverty reduction, and accelerating poverty reduction will help these countries sustain good policy.

Some of the countries in the Phase II list would not receive a large amount of aid based on the current level of their policies and the extent of poverty in these countries, but they have relatively new governments – and all except Guatemala are relatively democratic. These are cases where support may help the achievement of good policies and hence have a larger effect than we estimate if we hold policies constant.

The countries in the Phase IV list have weak policies and governments that have been in power a long time. We emphasize that there is an element of subjectivity in creating these lists. The executive in Zimbabwe has been in power for 20 years and the country has really poor policy; so too for Kenya (22 years, actually). We feel quite comfortable saying that these are low probability reforms and low aid effectiveness countries. In the cases of Zambia or Mozambique, on the other hand, the governments have been in power less long (9 years for Zambia and 14 for Mozambique) and policies are in the moderate range – not truly good, but better than those of Kenya or Zimbabwe. For both Mozambique and Zambia, Norway has reduced its aid as the decade has proceeded, and that seems a reasonable reaction to the slow progress with reform.

In most cases, considering the political economy impact of Norway’s aid leads to the same allocation conclusions as the application of the "poverty efficient" aid model that takes policy as given. Among the priority countries, aid is going to be relatively effective in Uganda, Ethiopia, Eritrea, and Bangladesh and help sustain policy there. Countries such as Mozambique or Zambia are lagging reformers, and it makes sense to modestly reduce the support there – though not to zero. Zimbabwe is not a high aid effectiveness country nor a high probability reformer. The political economy consideration would not change the conclusion that aid would be better utilized in other countries. For Sri Lanka, on the other hand, there is a relatively new government that has made progress with reform, so a different judgment is required.

6. Aid and Conflict Prevention

So far we have focused on the objective of poverty reduction. However, a second legitimate objective of aid is to reduce the risk of conflict. Recent research has begun to quantify the risk of civil war, and to analyze the effect of aid on this risk (Collier and Hoeffler, 2000, 2000a). We now briefly describe this research and apply it to Norway’s current priority countries.

The Collier-Hoeffler model of civil war is based upon global data for the period 1960-99. The model predicts the risk of conflict during a five-year period, on the basis of characteristics prior to the period. Appendix Table 3 presents the results of the core logit regression. Collier and Hoeffler find that economic factors are highly significant in determining the risk of conflict, so that potentially both policy and aid can be effective in reducing risks. Three aspects of economic performance are directly important. A faster rate of economic growth directly reduces the risk of conflict. A higher level of per capita income directly reduces the risk of conflict. Reduced dependence upon primary commodity exports directly reduces the risk of conflict. Surprisingly, the obvious indicators of political grievance, notably poor political rights and high economic inequality, have no effect on the risk of conflict. Collier and Hoeffler suggest that perhaps most societies have groups who are willing to resort to violence for some cause, so that the determining factors in whether civil war occurs are not grievances but rather the financial and military ability of such groups to engage in large-scale combat. Rapid growth and high per capita income make it more difficult for rebel organizations to escalate combat, whereas a high degree of dependence upon primary commodity exports offers rebels opportunities for financing their organization (as with alluvial diamonds in West Africa, drugs in South America, and timber in East Asia).

Aid potentially affects the risk of conflict through several different routes. First, it might directly affect risk through augmenting the government budget. This might enable the government to increase its military expenditure, or it might act as a lure to rebels seeking to capture the state. Collier and Hoeffler add aid into their regression of conflict risk. Aid is lagged by one five-year period to reduce problems of endogeneity: donors will evidently reduce funding in countries with a very high risk of conflict. They find that there is no significant direct effect. All the effects of aid work through the three economic variables noted above: the level, growth and structure of income.

The effect on the level of income evidently works by means of the cumulative effect of growth. The effect of aid on growth was central to the analysis of Assessing Aid, and so the same analysis now applies in the context of the reduction of conflict risk instead of poverty. The effect of aid on the structure of income works through two distinct routes. A priori, we would expect aid to cause `Dutch disease’: the provision of foreign exchange through aid tends to appreciate the real exchange rate and so reduces the incentive to export. Collier and Hoeffler find that aid indeed directly reduces dependence upon primary commodity exports and Dutch disease is the most likely explanation. Additionally, to the extent that it raises income, aid further reduces primary commodity dependence. As economies grow they typically change their structure away from primary commodities.

To summarize, although there is no direct effect of aid on conflict risk, there are four indirect effects, all favorable. Three of these depend upon policy: with reasonable policy in place, aid raises growth. This directly reduces conflict risk, cumulatively raises income, which further reduces conflict risk, and gradually changes the structure of the economy away from primary commodity dependence, which also reduces conflict risk. Only the fourth effect, Dutch disease, does not depend upon policy.

In Appendix Table 4 we simulate the effect of a donor-government package of additional aid of $1 per capita per year, sustained for five years, and a one point improvement in economic policy as measured by the World Bank’s Country Policy and Institutional Assessment, also sustained for five years. We take a hypothetical country which has the characteristics of the average aid recipient country. Initially, the conflict risk for such a country is 11.3%: that is, during a five year period there is approximately one chance in nine that a major civil conflict will be initiated, causing more than one thousand combat-related deaths. Sustained over five years, the aid and policy improvement package reduces this risk to 7.9%. Thus, the risk is reduced by around 30% in a relatively short period. With policy reform alone, that is, without the increase in aid, the risk would have been reduced to 8.6%. Hence, the relatively small increase in aid reduces the risk by around 10% and the (relatively large) improvement in policy reduces risk by around 20%. While these are simulations for a hypothetical country, they illustrate the orders of magnitude which donors might expect from aid used to prevent civil conflict.

In the above example we considered a representative aid recipient country. We now consider differences between countries so that aid for conflict prevention might be targeted where it is most effective. Where the objective of aid is poverty reduction, our previous argument has been that Norwegian aid should target countries with the combination of good policies and high poverty. Just focusing on countries with good policy is wasteful, because many of these countries have little poverty, and just focusing on countries with high poverty is wasteful because many of these countries have policies which are too poor for aid to be effective. When the objective is conflict prevention there is an analogous need to target countries which have the combination of good policies and a high risk of conflict. Most countries with good policies do not have a high risk of conflict. Conversely, some of the countries with high conflict risk have policies which are too poor for aid to be effective in risk reduction.

We now turn to the Norwegian high priority countries. In Table 2 we have already noted the approximate policy rating. The remaining required information is on the rating of conflict risk. While our model of conflict can be used to assess conflict risk, such information must necessarily remain confidential. Of the priority countries, Sri Lanka is currently conflict-ridden. Sri Lanka has good policies and so aid should be effective in reducing conflict risk. Our model suggests that there are substantial differences in the risks among the other priority countries, with four standing out as having markedly higher risk than the rest. These are (in alphabetical order), Ethiopia, Mozambique, Nicaragua, and Zimbabwe. In generating this list, the model makes no use of either current political information or detailed country knowledge. It is based upon a few economic, social, geographic and historical characteristics which globally have some predictive power. There are many countries which the model mis-predicts: countries which have survived peacefully for long periods, despite unfavorable characteristics, and countries which have suffered conflict despite favorable characteristics. The countries identified by the model may therefore not, in fact, face relatively high risks of conflict and this must be a matter of judgement for country specialists. However, if for present purposes we accept the list, then consider the scope for policy to mitigate the risk. Of the four, Zimbabwe has poor policies and so aid is likely to be ineffective in reducing conflict risk. Mozambique and Nicaragua have moderate policies so we might expect aid to be modestly effective. However, they are both very large recipients of aid, and given diminishing returns to aid, this reduces the effectiveness of additional aid. Hence, in these two countries there also seems no basis for expecting additional aid to reduce conflict risk significantly. Only Ethiopia has both good policies and a high risk of conflict. This constitutes a good case for aid to Ethiopia, in addition to the case based upon the high effectiveness of aid in reducing poverty. We should make it clear that the role of aid in conflict reduction refers only to internal conflicts. This analysis has no bearing upon the international war between Ethiopia and Eritrea. Although Eritrea was indeed formerly part of Ethiopia, the recent war was not a civil war. This is not merely a legal matter. The occurrence of civil wars depends upon very different processes from international wars, since civil wars require an informal rebel organization to become viable, whereas international wars are fought by governments which already have a secure tax base and the power of conscription. Hence, our analysis of conflict risk in Ethiopia concerns internal conflict, rather than renewed international hostilities with Eritrea. The case for aid to reduce the risk of internal conflict in Ethiopia faces precisely the same political difficulty as does the case for aid to reduce poverty, namely that the Ethiopian government has recently conducted an international war.

To summarize, among Norway’s priority countries, Sri Lanka and Ethiopia stand out as facing either current conflict or a significant risk of conflict, and having policies which would make aid effective in reducing this risk. Sri Lanka would not be a priority country on the criterion of poverty-reduction, since it has relatively little poverty. Hence, whether it should remain a Norwegian priority might depend upon the relative weight of poverty reduction and conflict prevention as Norwegian objectives. Ethiopia should be a high priority country on both the criteria of poverty reduction and conflict prevention.

References

Alesina, Alberto, and David Dollar, 2000, "Who Gives Aid to Whom and Why?" Journal of Economic Growth, March.

Burnside, Craig, and David Dollar, 2000, "Aid, Policies, and Growth," American Economic Review, September .

Collier, Paul, and David Dollar, 2000, "Aid Allocation and Poverty Reduction," World Bank, revised June 2000.

Collier, Paul and Anke Hoeffler, 2000, "Greed and Grievance in Civil War," Policy Working Paper 2135, World Bank.

- , 2000a, "Aid, Policy and Peace," mimeo, Development Research Group, World Bank.

Devarajan, Shanta, David Dollar, and Torgny Holmgren, 2000, "Aid and Reform in Africa," World Bank mimeo.

Dollar, David, and Jakob Svensson, 2000, "What Explains the Success or Failure of Structural Adjustment Programs?" Economic Journal, October.

World Bank, 1998, Assessing Aid: What Works, What Doesn’t, and Why. Oxford U. Press for the World Bank.

Appendix Table 1.

Explaining the Allocation of Norwegian Aid

1991-1998

t-statistics in parentheses

Appendix Table 3

The Collier-Hoeffler Logit Model of Conflict Risk

(dependent variable is ln(p/(1-p)); where p is the probability of civil conflict during a five year period.)

Notes: All regressions include a constant. Standard errors in parentheses. ***, **, * indicate significance at the 1,

5 and 10 percent level, respectively.

Appendix Table 4: Aid, Policy and the Risk of Conflict: a Simulation of the Effects of Increased Aid and Improved Policy for the Mean and Recipient Country