In the wake of last year’s global movement to “make poverty history,” governments of the richest countries have committed to doubling foreign aid to Africa and recommitted to halving extreme poverty by 2015. Ensuring the effectiveness of foreign aid is more relevant than ever. In this context, Abhijit Banerjee’s proposal to subject aid to randomized experiments is very welcome.

Randomized experiments are indeed a useful tool to test the effectiveness of interventions that aim to provide public goods or introduce new technologies. Such interventions might involve the introduction of new seed or health technologies, the expansion of a social-service program in health or education, or a new approach to motivating parents to enroll their children in school. These are the kinds of interventions—along with short-term projects—that donors typically like to fund. And Banerjee is right: donor agencies disbursing such aid ought to use randomized evaluative assessments before committing large amounts of money for the rapid scale-up of specific interventions. Not doing so is indefensible, although it must be recognized that such randomized experiments are prohibitively costly for private aid agencies working on a smaller scale.

But a more important question is whether randomized experiments represent the Holy Grail in evaluative research that will fix foreign aid. We believe not. While a large share of current aid budgets could be reviewed using randomized testing, there are other interventions that are not amenable to randomized experimentation. These are less popular among donors but perhaps no less important.

For example, at Oxfam we define poverty as social injustice rather than the absence of public goods or services. Our ultimate goal is to redress the power imbalances that limit the poor from accessing such goods and services while empowering them to defend their economic and social rights. We are very conscious of the fact that, despite tremendous technological and economic progress, millions of people remain trapped in poverty.

The global poverty challenge is therefore social and political as well as technological. While investments in agricultural technologies, health care, and education are essential, it is also essential to help poor people defend their rights. Indigenous peoples should be able to protect their land and water resources against the encroachments of extractive industries; small farmers should be able to organize and negotiate favorable prices with middlemen and transnational corporations; and women should be able to inherit wealth from their spouses and found their own businesses as promised by international law.

Advocacy projects such as these are context-sensitive and do not lend themselves to measurement by randomized experiments. For instance, there are many poor communities affected by large-scale mining projects, but their local circumstances are unique. Oxfam cannot standardize its interventions because the interest groups that favor or oppose the mining project may differ widely from one location to the other. Aid agencies funding advocacy projects therefore must rely on qualitative evaluation techniques and satisfy themselves with less tangible—though still meaningful—evidence of impact.

Furthermore, in many countries what is needed most is not new services but the basics. In a failing state even a new service that proved successful in a randomized experiment would falter at the project’s completion, when the donor moves to the next innovation. In such cases, aid must cover the recurrent costs of the social sectors over the long term: more classrooms with teachers who are well trained and retained with adequate pay. Long-term success depends on building the capacity of states to consistently deliver high-quality basic services.

Some official donors have risen to the challenge by offering sector-specific budget support to states that are willing but financially unable to deliver basic health and education services. That means committing large sums of money over several years for the whole health or education sector of a country, leaving the recipient government in charge of its allocation so that the aid can be used for both recurrent costs and innovations.

Banerjee criticizes this practice because the results cannot be measured. But why cannot such aid be made conditional on the achievement of certain defined service-delivery targets, such as school enrollment rates or vaccination rates? Donors could then adjust aid levels to penalize states that perform badly and reward those that perform well against these targets. It is thus possible to hold recipients of budget support accountable.

Banerjee’s call for randomized experiments has the potential to add real value to the assessment of programs in specific fields. But there are other, no less important, areas—such as governance, institutional capacity-building, and community empowerment—that are neither well-suited for nor easily adaptable to randomization. Solving part of the impact-assessment puzzle will certainly advance our work. We will need similar levels of imagination and creativity in other areas if we are to improve the outcomes of foreign aid.