Any Change to ODA Eligibility Should Lower the Graduation Criteria and Account for Purchasing Power

The OECD Development Assistance Committee (DAC) is currently undergoing a review of the role and definition of official development assistance (ODA) alongside the way the DAC functions. One of the elements of this review is re-examining the ODA-graduation criteria, the rules that dictate which countries are eligible to receive ODA. Currently, a country graduates if its GNI per capita remains above the World Bank high-income threshold (currently just under $14,000) for three consecutive years. The review is considering adjustments, such as adding new non-income-eligibility criteria around vulnerability, market access, or fragility.

We argue that if reform is needed, it is to use a GNI per capita measure adjusted for purchasing power parity (PPP), rather than the current “Atlas” method, and to use a lower threshold, which would focus ODA more on its core purpose of improving the economic development and welfare of people in developing countries. While there is no straightforward, uncontroversial way to define which countries are “developing,” GNI per capita PPP provides the best benchmark, and adding other plausible indicators provides little new information. We suggest a new upper-end ODA-eligibility threshold of PPP $25,000 per capita, which would cover around 60 percent of the world’s population (compared to 83 percent currently). However, assistance to countries above this threshold should be eligible if responding to UN humanitarian response plans.

At a time when aid budgets are shrinking, prioritisation becomes even more important. The DAC should question why some ODA-eligible countries are richer than its own members. The eligibility rules should direct ODA to where it is needed most; anything else would shift ODA further away from its core purpose.

ODA eligibility under review

Since the mid-2000s, the DAC has used the World Bank high-income threshold as the threshold for ODA-eligibility: when countries exceed that level for more than three years, they are no longer ODA-eligible unless the DAC unanimously agree to postpone graduation. To define the high-income threshold, the World Bank uses GNI per capita measured by the “Atlas” method, which essentially converts GNI per capita in local currency to dollars, using an average exchange rate over the prior few years to smooth out fluctuations.

There are increasing calls to change this criterion to include measures of vulnerability, especially from (or on behalf of) small-island developing states (SIDS) on the basis that GNI per capita does not capture “structural vulnerability” and exposure to shocks, and that some countries are therefore unfairly excluded from needed concessional finance. This note examines those arguments.

Not everything has to be ODA

ODA has a specific purpose given by its definition. This purpose certainly does not, nor should it, cover every case where countries or international organizations provide finance or donate resources to help another country. It is possible to respond to a disaster even if that response isn’t ODA. There are plenty of examples: China provided millions of masks to Italy during COVID. The UK assisted the British Virgin Islands after Hurricane Irma. DAC countries often use military vessels in disaster relief efforts, for which the full cost cannot be counted as ODA.

Again, several countries that are not ODA-eligible currently receive finance from the multilateral system—officially non-concessional but nevertheless cheaper than what is available on markets. For example, in 2025, the Inter-American Development Bank and the World Bank made commitments to Barbados worth $587 million. This is equal to around 7 percent of its GDP, well above the median official financial flows to ODA-eligible countries in 2024 of 5 percent (from all sources). It is more than was committed to the DRC (5.7 percent of GDP) or Nigeria (3.6 percent of GDP), which together account for over a fifth of the world’s poor. And several multilateral development banks offer support for developing other forms of protection against disaster, such as the World Bank “Capital at Risk” catastrophe bonds, used by several non-ODA eligible countries.

These are all examples of support for countries that fall outside the ODA boundary. We do not need to try to redefine ODA to include every good and moral expenditure we can imagine.

Furthermore, the DAC eligibility criteria primarily apply to bilateral ODA, as multilaterals generally have their own methods for allocating resources (for example, the threshold for accessing concessional resources from the World Bank is already much stricter). Changing the graduation criteria would therefore only affect a limited subset of development finance.

Does adding vulnerability change much?

It might appear appealing to expand the criteria beyond just GNI per capita on the correct and reasonable grounds that GNI per capita is very far from a perfect measure of welfare. But that has been accepted since the measure was created, well before it was used by the DAC. The relevant question is: Are there reasons today to believe that adding more dimensions would better capture welfare or need? And here the evidence is weak.

For it to be worth adding another dimension to the eligibility criteria, there would need to be some other determinant of welfare that is measurable, somewhat predictable, and not already well correlated with GNI per capita. If it is not easily measurable, it can’t be included in an eligibility formula. If it impacts countries unpredictably, then there is no way of determining which additional countries should be ODA-eligible ahead of time. And if it is closely correlated with GNI per capita, then it adds little “information” to the eligibility criteria.

Vulnerability indices, perhaps the most popular potential additional criteria, do not perform well against these tests. As Beynon (2025) has demonstrated, the most popular indices do not agree with each other. Very few countries consistently appear in the top quartile (most vulnerable) across all indices, and the ranks of some countries differ by more than 100 places across indices. Measuring vulnerability is clearly difficult and highly subjective.

Furthermore, two of the most popular vulnerability indices are highly correlated with GNI per capita (Pearson’s correlation coefficient of 0.85 for NDGAIN and 0.69 for the Inform Climate Change index), so including them would have little impact on eligibility. The other two indices covered by Beynon (2025)—the World Risk Index and the Multidimensional Vulnerability Index—are poorly correlated with GNI per capita, but negatively correlated with each other, and so do not present a clear picture of which additional countries should be included. The one thing the indices do all agree on is that least developed countries and low-income countries are the most vulnerable, but of course they are already ODA-eligible.

As a result, including vulnerability measures may lead to some arguable decisions. Barbados often comes up in conversations about ODA-eligibility. But as well as having higher GNI per capita than multiple DAC members (by the Atlas method at least – see below), it is also less vulnerable than both Greece and Latvia according to every measure with data available for all countries. There is no reason to make Barbados eligible to receive ODA and not Greece and Latvia, according to these indices.

Discretion is built into the system already

If worst comes to worst, and there is a clear disconnect between GNI per capita figures and welfare within a country about to graduate, the DAC already has discretion over whether to approve that graduation. A recent example is Montserrat, which has been high-income for more than three years and should graduate in 2026, but for which the UK has successfully (if controversially) lobbied a pause. Similarly, if a country is on the verge of graduating but gets hit by a ferocious hurricane that will obviously affect future income, the DAC already have the option to delay graduation. It would only be worth adding new dimensions if there were systematic long-term deviations between GNI per capita and welfare, and it is not clear what these would be.

What changes would make sense?

Use a measure that accounts for in-country price levels

There are obvious reasons to think that GNI per capita using purchasing power parity (PPP: how much a dollar actually buys in a country) is a better measure of welfare: it is designed to reflect the fact that although incomes might be lower in some countries, prices also tend to be lower, and what really matters for welfare is how much that income can buy. And if you compare correlations between obvious indicators of welfare such as child or maternal mortality, or how many children complete primary school (or indeed vulnerability indices), the PPP measure is better correlated in every case.

Table 1. Pearson’s correlation between GNI per capita measures and welfare indicators, 2010-2024

Sources: World Bank, UN
Notes: Table shows Pearson’s correlation coefficient between the variables for each country-year observation with data available for both variables between 2019 and 2024. Being as inclusive as possible for each variable means that observations differ between rows (for example, if maternal mortality is unavailable for a country in 2019, but child mortality is available, it will be included in one row but not the other). All countries (including developed) are included for which there is data. *Data is downloaded from HDI and just refers to 2023.

This could affect small island states in particular—they cannot benefit from internal economies of scale and are more reliant on imports from afar. In these countries, income adjusted for purchasing power parity is likely to be lower, reflecting those additional costs.

What difference would shifting to a PPP income measure make? This would depend on where it is set, and there are no income group thresholds for GNI per capita in PPP terms. But to explore the above question, we first define PPP income thresholds by calculating the percentiles of the country distribution to which the thresholds correspond to the Atlas market exchange rate measure. For example, the low-income threshold is $1,135, which is roughly the 10^th percentile of 188^[1] countries based on the distribution of Atlas-method GNI per capita in 2024. So we define the low-income threshold in PPP terms as the 10^th percentile of the distribution of countries under the PPP measure ($3,340). The high-income threshold would be $32,300 using the same approach.

Most countries remain in the same income group: there is no debate as to whether Qatar is high income or Somalia low income. But there are 12 countries that fall to a lower income group under PPP (and therefore 12 moved up), and nine of those who fall are SIDS. Five countries move from high income Atlas to upper middle income PPP status, i.e. from outside to within the ODA boundary, and four of these are SIDS (Antigua and Barbuda, Barbados, Nauru, and Seychelles). The remaining country to fall back inside eligibility – Costa Rica – shares many features with SIDS. The countries that graduate to high income status under PPPs include Türkiye (which by the PPP measure is richer than DAC members Latvia and Greece), Kazakhstan and Malaysia (both of which are now wealthier than the UK was in the 1990s).

Is using the PPP measure plausible?

One justification for sticking with the Atlas measure may be that it is simpler. It is incredibly difficult to accurately compare the cost of living across countries in which typical consumption bundles differ substantially. The International Comparison Programme (ICP) that produces the PPP adjustments needs to make heroic assumptions in order to produce its datasets. And the PPP adjustments are only fully updated every few years, relying on extrapolations in intermediate years.

Furthermore, the World Bank defines a convenient (if arbitrary) threshold for ODA-eligibility: there is a logic to arguing that a country is no longer “developing” if it has reached the highest possible income group. The World Bank does not define such a threshold for the PPP measure, meaning the DAC would have to set one themselves, which could be contentious.

Neither argument is strong enough to justify sticking with the Atlas measure. It might be true that measuring PPP is hard, but the correlations in Table 1 suggest that it captures something real, those challenges notwithstanding. And even if there is some discomfort with relying on extrapolations between ICP rounds, the cadence of those rounds is roughly every four years. Given that countries currently need to exceed the threshold for three years before graduating, this will not be a significant delay in most cases. As for the threshold, it is already contentious, as evidenced by the decision to review eligibility criteria, the controversy around some of the countries that are still ODA-eligible (such as China), and past papers from the OECD themselves exploring arguments for changing the threshold.

Reduce the threshold

Shifting to a PPP measure would mean the DAC needing to set a new threshold. Rather than setting it at the same percentile as we discuss above, it should be set lower. As the Türkiye example shows, the current rules allow ODA to flow to places that really don’t need it. As ODA budgets are declining, the opportunity cost of giving to countries at a similar level of income to DAC members has increased. If the DAC is going to change eligibility criteria, it should start by revisiting the threshold.

Above, we calculated that the high-income threshold would be $32,300 if set at the same percentile as the threshold under the Atlas method. This would still include many relatively wealthy countries, notably China and Mexico. One proposal would be to reduce the ODA eligibility level to $25,000 PPP, covering around 60 percent of both the world’s countries and the global population, as opposed to 67 percent of the world’s countries and 83 percent of the population covered by the current definition. This was roughly the GNI per capita (PPP) of the US when the DAC was formed (adjusted for inflation).^[2] Several relatively wealthy countries would be excluded, but despite the lower threshold, Barbados and Nauru would re-enter eligibility. There was only $5.5 billion of ODA spent in countries that would be excluded under this new threshold, suggesting that this tighter definition would create little risk of excessively concentrating resources. But given a declining marginal return of a dollar of aid in terms of its impact on quality of life as recipient countries get richer, this concentration would increase aid effectiveness.

Table 2. Changes to ODA eligibility from moving to $25,000 PPP threshold

We think that this threshold still includes some countries where bilateral aid will achieve little, such as Brazil and Colombia (both of which even have their own aid programmes). It is still roughly equivalent to the UK’s income in the 1970s, or that of Spain and Portugal in the 1980s, and therefore a high threshold by historical standards. We would support reducing the threshold even further, to $20,000. This is (very) roughly the UK’s GNI per capita in PPP terms when the DAC was formed.

Table 3. Changes to ODA eligibility from moving to $20,000 PPP threshold

Make an exception for humanitarian response

Reducing the threshold would risk making some countries that are genuinely in dire situations ineligible. Syria, Venezuela, and Iraq were all upper-middle-income countries once, but to the extent that they are still in that category, it is because data hasn’t been updated (understandably not the priority mid-conflict). Where there is a humanitarian disaster unfolding that a country cannot raise the resources to deal with themselves, and will inevitably have huge economic impacts, the DAC should not necessarily wait until that country is officially back in the ODA-eligible bracket, especially as quick action is often required to save lives.

This suggests there should be an exception for humanitarian situations. This would be administratively easy to implement: any country with a UNOCHA humanitarian plan should be eligible to receive humanitarian aid, counted as ODA. This might preclude anticipatory action in countries not otherwise eligible for ODA, which is funded ahead of crises unfolding. But the vast majority of anticipatory action is spent in the poorest countries and provided by multilaterals not beholden to the eligibility criteria. And where crises are predictable (such as hurricanes in the Caribbean), there are other means of finance available, as discussed above (contingent credit, sovereign insurance facilities), many of which are indirectly ODA funded already.

ODA should retain focus on core purpose

ODA is a valuable and increasingly scarce resource that has its greatest impact in the world’s poorest countries. Any effort to reform the ODA eligibility requirements should be designed to increase that impact, not maximize the flows that can count as ODA and have the opposite effect. It is understandable that some countries not currently eligible would want to make the case for access to concessional funding. But bilateral ODA is far from the only source of funding, and other sources have their own funding rules. As bilateral ODA falls, it needs to become more targeted on the countries most in need, and GNI per capita (adjusted for purchasing power) still does the best job of capturing that.

[1] Several countries have missing GNI per capita data (for both atlas and PPP measures)

[2] This is based on data on GDP per capita (PPP, constant 2024 prices) from the Maddison project, but for the US GNI and GDP are very similar.