There is an increasingly familiar response in the international development sector whenever an intervention fails. We are told that the organisation and its leaders should be praised for their honesty, applauded for their willingness to learn, and commended for changing course when the evidence no longer supports their programme. If a randomised controlled trial shows weak results, if the model does not deliver what was promised, if the evidence points against continuation, then the ethical thing to do is to stop.
There is truth in that. Too many organisations persist with broken ideas for too long. Too many donors reward scale, confidence and good storytelling over reflection, correction and restraint. So yes: when evidence shows that an intervention is not working, stopping is better than pretending otherwise.
But stopping is not the same as being accountable.
I have been thinking about this again after reading Evidence Action’s recent reflection on its chlorine dispenser programme in Kenya. The organisation explains that new evidence led it to conclude that the programme was not performing as expected and that it would therefore significantly reduce it. The message is clear: we tested, we learned, we adapted, we closed.
On one level, this deserves recognition. It is better than wasting hard-to-get development dollars. It is better than denial. It is better than institutional stubbornness. It is better than dressing up failure as success.
But there is another truth we need to hold in our heads at the same time. It is not enough to say that an organisation has learned.
We must also ask: who bore the cost of that learning? And: should discontinuing a public service be treated in the same way as discontinuing a commercial one?
That second question matters because development interventions are too often presented as neutral technical experiments, pilot products, or delivery mechanisms. But many of them do not function like ordinary market goods and services. When an organisation provides access to treated water, migration support, health-related services or any other intervention that people begin to depend on, it enters a moral and political space, not just a technical one. It is no longer merely testing an idea. It is shaping behaviour, expectations, routines, coping strategies and, sometimes, survival itself.
To withdraw such a service is not the same as a company discontinuing an unsuccessful product line. It is closer to a public authority scaling back a service on which people have come to rely. Yet the organisations making these decisions are often not public authorities. They are not elected. They are not locally accountable. They are not embedded in the communities whose lives they are changing. They have no personal stake on the outcomes.
And yet they assume the power to introduce services, structure access to them, test them, redesign them and withdraw them.
This is where the language of learn and close becomes ethically too neat.
It suggests that the central moral act is to stop once the evidence turns against one. But stopping does not erase what came before. If a programme was introduced without sufficient legitimacy, without meaningful local ownership, or despite foreseeable harms, then closing it does not resolve the deeper ethical problem. It may be the least bad option at that point. But it does not cancel the responsibility incurred by acting in the first place.
I think researchers struggle with this because this is not what they were supposed to do. Research, even the most applied kind, is always at arm’s length from delivery and beneficiaries. The few times I have seen think tanks step into the space of implementation, I’ve also witnessed nuanced and difficult ethical debates about the organisation’s functions and responsibilities.
This is because they recognise that stepping into the “doing” comes with additional responsibilities that most research organisations and researchers are not prepared for.
And it creates an inevitable conflict of interest. Are they more interested in learning or in succeeding?
But in international development, this nuance is sometimes lost amid a sense of urgency and, let’s be honest, a feeling of technical and moral superiority.
This is why I keep returning to No Lean Season, Evidence Action’s earlier programme in Bangladesh. The project emerged from a hypothesis that poor rural households could be helped during the lean season if they were given a small grant or loan to support temporary migration to urban areas for work. The reasoning was that a family member could migrate, earn income, send money home, and help the household survive the months when there were no crops and little local work.
On paper, it was plausible. In practice, it involved experimenting on some of the most vulnerable people in the world.
The ethical issue was never only whether the hypothesis would prove correct. It was also whether the people designing and testing the intervention had the legitimacy to expose those families to the risks involved. Those risks were not abstract. They were foreseeable. If very poor rural households were given money to send someone away for work, it was entirely predictable that some would send their youngest and most vulnerable members. Anybody with knowledge of these contexts could see the dangers: unsafe travel, exploitation, abuse, trafficking, and a broader transfer of risk from institutions to families with almost no protection.
A researcher in Peru told me that even before No Lean Season had been fully tested, the intervention was being presented to the Peruvian government as a policy alternative. Luckily, researchers in Peru had the contextual knowledge and political clout to stop it on its tracks. The risk was too high to consider.
When a tragedy later occurred and No Lean Season was ultimately shut down, the response from many quarters was to praise the organisation for its openness. It had learned. It had reflected. It had acted responsibly by stopping.
Again, yes, well done for stopping, but that response asks too little.
The fact that researchers eventually recognise risk does not answer the question of why they were entitled to impose that risk in the first place. The fact that an organisation is transparent after the event does not settle whether it was legitimate for it to structure the experiment as it did. The fact that the evidence later turned against the programme does not change the fact that real people had already borne the cost of generating that evidence.
This is the core of the problem: methodological rigour is not the same as ethical legitimacy.
An RCT may help answer a causal question. It may tell us whether an intervention had a measurable effect under certain conditions. But it does not answer the prior political and moral question: who gets to decide that this intervention should be tried, on these people, in this place, with these risks?
This is at the core of IDRC’s approach to scaling impact. The first question it seeks to address is: What is the moral case for scaling?
When I asked a former MineduLab staff member about the role of ethics in the multiple experiments carried out by researchers from the J-PAL network with data from the Peruvian government, I got a very clear answer: ‘if this issue [of research ethics] had been addressed at the time, many of the experiments would have been impossible to test.’
Randomisation is not a moral waiver
In fact, the stronger the method, the easier it often becomes to obscure the politics behind it. The trial appears objective. The intervention appears technical. The decision appears scientific. But underneath all this sits an exercise of power. Someone decides whose lives will carry the uncertainty. Someone decides who gets the service and who does not. Someone decides what level of risk is acceptable. Someone decides when the experiment begins and when it ends.
And in contemporary international development research, those decisions are still very often made by actors far removed from the communities concerned: researchers affiliated to elite universities, international NGOs, foundations, and organisations based in Washington, New York, London, Geneva or California. They may be sincere. They may be careful. They may even be more reflective than most. But sincerity is not legitimacy.
This is why the learn and close narrative is so unsatisfactory. It turns accountability into a narrow technical matter. Were the researchers honest enough to revise their priors? Were they evidence-driven enough to stop? Were they transparent enough to publish the disappointing results?
These are not trivial questions. But they are insufficient.
A focus on the researcher instead of the researched
At the heart of the problem is that these questions – and the statements made around the cases of learning and closing – centre on the integrity of the researcher rather than the rights of the researched. They create a morally flattering story in which powerful actors appear responsible because they admit error, while the deeper asymmetry remains untouched: those who designed the intervention learn (and are praised); those who live with its consequences absorb the cost (and remain anonymous numbers in a database – at scale!).
And this matters even more when the intervention resembles a public service.
If you introduce a water treatment service (any service) into a community and people begin to use it, you are no longer simply testing a theory of behaviour change. You are, in effect, participating in the organisation of everyday life. If you then withdraw that service because the usage rates are lower than expected, the cost-effectiveness model no longer holds or even if you find that its use has a negligible effect on development indicators, you are not merely ending a study. You are altering access to something people may have integrated into their routines.
What do you know? Maybe the time spent around the water dispensers offered users the chance to catch up on the latest gossip and take a break from an otherwise impossibly busy day.
A commercial firm may be entitled to discontinue a product if demand falls. But even this is limited. When commercial firms offer a public service, they agree to be regulated and assume some responsibilities.
Development actors should not be able to borrow the language of public purpose while retaining the discretionary freedoms of private experimentation. They cannot claim moral seriousness because they serve the poor and then behave as if the withdrawal of service were merely an operational decision. If an intervention functions like a public service, then the ethical standard for introducing and withdrawing it must be much higher.
This brings us back to legitimacy and accountability.
Accountability; but seriously
Researchers and implementing organisations cannot continue to act as though robust methods alone confer authority. They do not. Nor is it enough to say that participants consented, especially when power asymmetries are extreme, alternatives are limited, and the people most affected by the decision may not be the same as those formally enrolled in the study. Nor is it enough to say that local ethics review boards approved the intervention if those boards themselves operate within systems shaped by unequal funding, external agendas and limited bargaining power.
The question is not only whether people agreed to participate. It is whether the communities involved had meaningful power over the design, purpose, risks, continuation and closure of the intervention.
It is whether local researchers, local institutions and local public authorities shaped the agenda, or whether they were asked to legitimise decisions already made elsewhere. It is whether those running the programme are accountable in a political and moral sense, not just in a technical or academic one.
Hence, for a Global North initiative working in poor and vulnerable communities halfway around the world, accountability must mean more than publishing a thoughtful retrospective once the programme is over.
It must mean accepting responsibility for foreseeable harms. It must mean accounting for the dependencies created by temporary interventions. It must mean recognising that withdrawal is itself an ethically significant act. It must mean being answerable not only to donors, academic peers and organisational boards, but to the people whose lives were affected by the decision to intervene at all.
If organisations and researchers are unwilling to accept that level of accountability, then they should be much more cautious about intervening in the first place.
A possible solution is to treat this as a regulated industry.
None of this means that hypotheses should never be tested, or that organisations should be condemned for changing course when evidence turns against them. Quite the opposite. We should want institutions that are capable of learning. We should want interventions to be examined critically. We should want organisations to stop doing things that do not work.
But we should also reject the comforting idea that learning is enough.
Because the deepest ethical question is not whether researchers learn from failure. It is whether they had the legitimacy to fail in this way, with respect on these populations, and then narrate that failure as proof of responsibility and as leverage for future funding!
Until that question is taken seriously, we will continue to mistake methodological seriousness for ethical legitimacy. And they are not the same thing.