Why Some Interventions Work Only After You Understand the Story Around Them

The Strange Problem with “What Works”

We are often told that the central question in policy, medicine, and social change is simple: what works? But that question hides a deeper problem. What if the same intervention works brilliantly in one place, fails in another, and does both at once depending on who is involved, when it is introduced, and what people believe it is for?

That is not a flaw in the real world. It is the real world.

The deeper question is not whether an intervention works in some abstract sense. It is how, for whom, and under what conditions it works, fails, or changes shape. The moment we ask that question, we stop treating programs like bolts that either fit or do not fit, and start treating them like stories unfolding inside social systems. The outcome depends not only on the tool, but on the context that receives it and the reasoning people bring to it.

This shift matters because many of our most expensive failures come from confusing an intervention’s design with its destiny. A policy can be elegant on paper and still collapse in practice if it ignores local routines, incentives, trust, timing, and interpretation. In that sense, the real puzzle is not implementation after the fact. It is understanding why implementation was always part of the intervention.

Context Is Not Background Noise, It Is Part of the Mechanism

A common mistake is to treat context as the scenery around a program, something to document briefly and move past. But context is not passive. It actively shapes what people notice, what they believe is possible, and which actions they take. A new clinic protocol, for example, does not simply enter a hospital. It enters a world of staffing shortages, professional hierarchies, patient anxieties, and habitual workarounds.

This is why two neighborhoods can receive the same public health campaign and produce opposite results. In one place, the message may trigger trust and collective action. In another, it may be filtered through skepticism, fatigue, or competing local narratives. The campaign has not changed, yet its meaning has. That meaning is part of the intervention’s effect.

A useful mental model is to think in terms of mechanism plus context equals outcome. The mechanism is not just the visible program activity. It includes the human reasoning that the activity activates: motivation, fear, legitimacy, obligation, pride, status, or convenience. Context determines whether those reactions are amplified, muted, or redirected.

The same intervention does not produce different results because reality is messy. It produces different results because people are not machines, and systems are not empty containers.

This is why “scaling up” is often misunderstood. We imagine scaling as copying a successful program at larger volume. But the thing being scaled is not just a set of steps. It is a relation between a program and the world that allows those steps to have meaning. Without that relation, replication becomes imitation.

Why Good Evaluations Need Better Stories

Traditional evaluation often tries to answer whether a program succeeded by comparing before and after, or treatment and control. That can be useful, but it can also be misleading if the goal is learning. A program might look ineffective on average while quietly working for a specific group under specific conditions. Or it may appear effective while hiding a fragile mechanism that will break the moment conditions shift.

This is where a more realist approach becomes powerful. Instead of asking only whether an intervention produced an effect, it asks what pattern of conditions produced that effect. In practice, that means tracing the sequence from context, to mechanism, to outcome. The evaluation becomes less like a verdict and more like a detective story.

Consider a youth mentoring program. If attendance rises, the wrong interpretation is “mentoring works.” The better question is: what happened? Did mentors create belonging? Did the setting reduce shame? Did the program become a rare stable adult relationship in a chaotic environment? Did the effect appear only for participants who already had some family support? Each answer changes what we think the program is.

The same logic applies to public health. A vaccination drive can succeed not only because of logistics, but because it signals institutional seriousness, local respect, or peer endorsement. If those mechanisms are not understood, future attempts may copy the visible surface, such as posters, clinics, or reminders, while missing the deeper reason the first effort worked.

This is why evaluation should be seen as a theory building exercise, not just an accountability exercise. The point is not merely to judge. The point is to learn what kind of world the intervention requires in order to work.

The Hidden Variable Is Human Reasoning

The deepest connection across these ideas is that interventions do not act directly on outcomes. They act on people, and people interpret. A policy is not a command traveling through empty space. It is a cue that enters a crowded field of beliefs, identities, norms, and fears. What matters is not only what is delivered, but what people think is being asked of them.

This is why the same program can have very different effects even when all the logistics are identical. If a community sees an initiative as imposed, it may resist. If it sees the same initiative as legitimized by trusted local actors, it may embrace it. If staff see a reform as surveillance, they may comply superficially while subverting it in practice. If they see it as support, they may adapt it creatively.

An intervention therefore has at least two layers:

1. The formal layer: the activities, resources, and procedures on the page.
2. The interpretive layer: the meanings people assign to those activities.

The formal layer is easy to document. The interpretive layer is where the real causal action often lives.

Think of a school attendance program. On paper, it might offer text reminders to parents. In one setting, those reminders signal care and prompt action. In another, they may be ignored because parents work multiple jobs and can’t respond during the day. In a third, they may produce embarrassment, because they imply blame. The technical intervention is identical, but the social meaning is not.

This is the lesson many linear reforms miss. They assume people receive a program and then behave accordingly. In reality, people ask, often unconsciously: What is this really saying about me, about others, and about the system? That question determines whether mechanisms activate.

A Better Mental Model: Programs as Hypotheses

One of the most useful ways to think about interventions is as hypotheses about social change. A program is not a solution in itself. It is a testable claim about what conditions will cause what kinds of behavior, through what kinds of responses.

That model changes how we design, evaluate, and scale initiatives. It encourages humility. Instead of saying, “We know this works,” we begin saying, “We think this works here, for these reasons, under these conditions.” That phrasing sounds less confident, but it is far more useful.

It also helps explain why the best programs are often adaptive. If a policy is a hypothesis, then implementation is not mere compliance. It is part of the experiment. Frontline workers, local leaders, and participants are all co-producers of the result because they help decide which mechanisms are actually triggered.

Imagine two anti-smoking campaigns. One relies on graphic warnings and medical facts. Another combines those warnings with peer-led discussions and local champions. The first may be informative, but the second is more likely to shift behavior if it changes social norms. The difference is not just message content. It is whether the campaign changes the interpretive environment in which the message is received.

This is why some of the most successful interventions seem modest on the surface. They do not brute force behavior. They make a desirable action feel normal, safe, credible, or socially rewarded. They work by modifying the system of meaning around the choice.

If you want to know why something worked, do not only ask what was delivered. Ask what became thinkable, sayable, and doable as a result.

What This Means for Public Health and Policy

Public health is full of examples where averages obscure reality. A campaign can lower infections overall while leaving some communities untouched. A service can improve access while increasing inequalities for people with low digital literacy. A new protocol can raise compliance while demoralizing workers. A policy can succeed numerically and still fail socially.

That is why the goal should not be to maximize a single score. The goal should be to understand the distribution of effects and the mechanism behind them. Who benefited? Who was left out? What had to be true for the intervention to succeed? Which pieces are essential, and which are just local adaptations that made the program viable?

This matters for decision makers because implementation is always a choice between different risks. If you demand rigid fidelity, you may preserve the original design but destroy local fit. If you allow too much adaptation, you may lose the mechanism that made the program work in the first place. The trick is to distinguish core mechanism from surface form.

Here is a practical way to do that:

• The core mechanism is the causal logic that must remain intact.
• The surface form is the local expression of that logic, which may need to change.

For example, a vaccination outreach effort may need trusted messengers as its core mechanism, but the messenger could be a nurse in one place, a faith leader in another, and a neighborhood organizer elsewhere. The surface changes. The causal logic remains.

This distinction is crucial because it helps avoid two common errors. First, overstandardization, where we preserve the same format at the cost of effectiveness. Second, overlocalization, where we adapt so much that the mechanism disappears. Realist thinking does not merely sit between these extremes. It asks us to identify what must stay and what should flex.

Key Takeaways

• Stop asking only whether an intervention works. Ask for whom it works, in what context, and through which mechanism.
• Treat context as causal, not decorative. Trust, norms, incentives, and local history are part of the intervention environment.
• Look for human reasoning. People do not just receive programs, they interpret them, and those interpretations drive outcomes.
• Separate core mechanism from surface form. Protect the logic of change while allowing local adaptation in how it appears.
• Use evaluation to build theories, not just scores. The goal is to understand why change happened so it can be repeated intelligently.

The Real Lesson: Causality Is Social Before It Is Statistical

The deepest insight here is that social interventions do not fail first in measurement. They fail first in imagination. We imagine that action is separate from context, that delivery is separate from interpretation, and that success can be copied without understanding the social conditions that produced it. That is a comforting fantasy because it makes complexity look manageable.

But the world is less mechanical and more relational than that. Programs work when they fit into local realities in ways that activate the right responses. They fail when they ignore those realities or misread them. This means that the true unit of analysis is not the intervention alone, and not the context alone, but the relationship between the two.

If you change how you think about that relationship, you change how you design everything else. You begin to build interventions that are not merely delivered, but understood. You evaluate not just outcomes, but explanations. You scale not by copying forms, but by preserving functions.

That is the larger shift: from asking whether a program is universally effective to asking whether it is contextually intelligent. Once you see that distinction, many policy debates look different. Success stops looking like a fixed thing we can transport from one place to another. It starts looking like a pattern of alignment between people, purpose, and place.

And that is a far more demanding standard. It is also the only one that consistently works.