Why Good Impact Work Fails Without a Theory of Causation

May 02, 2026

10 min read

3 views

84%

The hidden question behind every serious intervention

What if the biggest risk in social impact, policy, or program design is not that we fail to act, but that we act with a story about causation that is too simple to survive contact with reality?

Most organizations want a clean answer to a clean question: Did it work? But the moment you look closely, that question fractures. Work for whom? Under what conditions? Compared to what alternative? Through which mechanism? And perhaps most important, how much of the observed change can we reasonably claim, versus how much merely happened around us?

This is where the deeper tension lives. A theory of change gives you a map of intended causation. Contribution thinking asks whether your map actually resembles the terrain. One is a design tool, the other is an evidence discipline. Together, they expose a hard truth: impact is rarely the result of a single cause, and certainty is rarely available in complex systems.

The real challenge is not producing a beautiful logic model. It is learning how to think causally without pretending the world is linear.

The trap of elegant certainty

People often build theories of change as if they were blueprints. If we do A, then B should happen, then C will follow. The diagram looks crisp, the arrows look decisive, and the logic feels reassuring. But real-world change is not a relay race where one handoff leads neatly to the next. It is more like weather: multiple forces interact, some visible, some hidden, some reinforcing, some canceling each other out.

That is why assumptions matter so much. Every arrow in a causal chain contains a bet. A bet that participants will show up. A bet that incentives are strong enough. A bet that institutions will not block implementation. A bet that the context will remain stable enough for the intervention to matter. Most failures are not failures of action alone. They are failures of unspoken assumptions.

Consider a job training program. On paper, the pathway is obvious: train people, improve skills, place them in jobs, increase income. But in practice, that pathway depends on dozens of conditions. Are employers hiring? Do participants have transport or childcare? Are the skills actually aligned with available work? Are there discrimination barriers? If the outcome does not appear, the problem is not necessarily that the program was ineffective. It may be that one or more assumptions were false.

This is why a good causal model should not just describe hope. It should reveal fragility.

A theory of change is not most valuable when it sounds convincing. It is most valuable when it tells you where the chain can break.

That is an uncomfortable standard, because many organizations prefer theories that inspire confidence rather than invite scrutiny. Yet scrutiny is exactly what makes the model useful. A pathway that cannot be challenged is not a theory. It is a slogan.

Contribution is not a weaker claim. It is a smarter one.

Once you move from design to evaluation, another illusion appears: the illusion that you must prove absolute causation or admit defeat. In complex settings, that binary is misleading. You rarely get to isolate one intervention like a chemistry experiment. Other actors are active, external trends are moving, and outcomes often emerge from overlapping influences.

This is why contribution analysis matters. It asks a different question: given the evidence, how plausible is it that this intervention helped produce this outcome, and through what chain of influence? The goal is not to overclaim. The goal is to build a reasoned, evidence-based contribution story.

That shift is profound. It reframes evaluation from courtroom verdict to detective work. Instead of asking, “Can we prove we caused this?” you ask, “What evidence would make our contribution story credible, and where is the story weak?” That difference matters because many valuable interventions operate in environments where perfect attribution is impossible, unethical, or expensive.

Take a city homelessness initiative. Suppose housing placements rise over a year. Did the initiative cause the increase? Maybe partly. But what if a new landlord incentive program expanded supply, a colder winter pushed emergency reforms, and a nonprofit coalition improved outreach? A contribution approach does not wash its hands and say nothing can be known. It assembles the chain of evidence, tests assumptions, checks alternative explanations, and asks whether the observed changes match the expected pathway.

The best contribution analysis is not modest in insight. It is rigorous about uncertainty.

Attribution asks for monopoly on outcomes. Contribution asks for responsibility within a causal ecology.

That phrase, causal ecology, is useful because it changes the frame. In ecology, nothing acts alone. Every organism lives within relationships, constraints, and feedback loops. Social change works similarly. Your program is one force among many, and your job is to understand how it interacts with the rest of the field.

The synthesis: theories of change should be built for testing, not just telling

The deepest connection between these ideas is this: a theory of change is only as good as its assumptions, and contribution analysis is how those assumptions meet reality.

That means a mature organization does not treat the theory of change as a static planning document. It treats it as a living hypothesis. Every major link in the chain should be paired with an explicit assumption and a test. If the pathway says training leads to employment, ask what would have to be true for that to happen. If the pathway says employer partnerships increase placement, ask what proof would show the partnership was actually changing employer behavior.

A useful way to think about this is to separate three layers:

Intentional causation: what you believe your intervention is designed to influence.
Contextual causation: what else in the environment is shaping the outcome.
Evidentiary causation: what observable signs make your contribution claim credible.

Most weak evaluations fail because they collapse these layers. They confuse intention with effect, or effect with contribution. A strong causal approach keeps them distinct.

Imagine planting a tree. The tree itself is your intervention. Sun, soil, water, and weather are contextual factors. Growth rings, root spread, and leaf health are evidence. If the tree thrives, you cannot claim the seed alone did everything. If it struggles, you cannot assume the seed was bad. You need to inspect the conditions and compare them against what the planting theory predicted.

This is what makes assumptions so central. They are not footnotes. They are the load-bearing beams of the theory. If you never identify them, you cannot test them. If you cannot test them, your theory remains aspirational rather than operational.

The practical implication is powerful: the best theories of change are designed backward from evaluation. Not in a bureaucratic sense, but in a causal sense. Before launching, ask how you would later know whether the causal pathway held. What evidence would confirm each step? What evidence would disconfirm it? Which assumptions are most vulnerable? Which are so critical that the entire intervention depends on them?

This turns planning into disciplined curiosity.

A better mental model: change as a chain of contested bets

One of the most useful frameworks for combining these ideas is to see any intervention as a chain of contested bets.

Each bet has three parts:

Claim: if we do this, something specific should happen.
Assumption: a condition that must hold for the claim to be true.
Test: an observation that would strengthen or weaken confidence in the claim.

For example, a literacy initiative may contain the following bets:

If teachers receive coaching, classroom practice will improve.
If classroom practice improves, students will read more effectively.
If students read more effectively, test scores will rise.

Each link has assumptions. Teachers must have time to apply the coaching. The coaching must be relevant. Students must attend regularly. The test must measure the right skill. A contribution analysis looks at whether those assumptions held and whether the evidence fits the expected sequence.

This model is valuable because it avoids two common mistakes.

First, it avoids the miracle mindset, where impact is assumed to appear just because the intervention was well intended. Second, it avoids the cynic mindset, where complex causation is treated as unknowable and therefore not worth analyzing. A chain of contested bets accepts uncertainty without surrendering to it.

You can even use this approach to improve strategy, not just evaluation. If the weakest assumption is outside your control, maybe the intervention should be redesigned. If the evidence needed to support a claim is too costly to gather, maybe the claim is too ambitious. If multiple actors are required for the pathway to work, maybe the strategy should coordinate them explicitly rather than hoping alignment emerges on its own.

In that sense, contribution thinking is not only about proving worth. It is about stress testing strategy.

When assumptions are made visible, strategy becomes more honest. When strategy becomes more honest, it becomes more adaptable.

Why this matters beyond nonprofits and evaluation

Although these ideas are often associated with program evaluation, their importance is broader. Any domain that deals with complex change faces the same problem: management, public policy, philanthropy, education, health, and even product strategy all live in a world where causes are distributed, effects are delayed, and outcomes are noisy.

Think about a company launching a new onboarding system. Revenue may rise afterward. But was onboarding the cause? Maybe. Or maybe market demand improved, pricing changed, or a competitor stumbled. A contribution lens forces leaders to ask better questions than simple before and after comparisons. Did the new system improve activation rates? Which customer segments responded? What assumptions did the design rely on? What else changed at the same time?

Or consider public health. A vaccination campaign may coincide with declining infection rates. But attributing the decline requires more than temporal overlap. You need to know whether vaccine uptake was high enough, whether the target population changed behavior, whether other interventions were active, and whether the pattern matches the expected causal sequence. Again, the point is not to make certainty impossible. It is to make claims proportionate to evidence.

The broader lesson is that good decision-making depends on causal humility. Causal humility does not mean passive skepticism. It means knowing that every claim sits inside a web of assumptions, competing explanations, and partial evidence. That humility improves design, resource allocation, and learning.

It also protects organizations from a subtle but costly failure: confusing narrative coherence with causal validity. A story can sound beautiful and still be wrong. A spreadsheet can show movement and still mislead. What matters is whether the story survives testing.

Key Takeaways

Treat every causal link as an assumption to be tested. If a theory of change has no visible assumptions, it is not ready for real-world use.
Ask for contribution, not monopoly. In complex systems, the right question is often whether an intervention plausibly contributed to change, not whether it can claim full ownership of the outcome.
Design evaluation at the same time as strategy. Build your theory backward from the evidence you would need to trust it later.
Separate intention, context, and evidence. This prevents confusion between what you planned, what the environment did, and what actually happened.
Use weak points to improve the intervention. The most valuable thing a causal analysis can reveal is not success, but fragility.

The real payoff: replacing certainty theater with causal intelligence

The best organizations do not merely ask whether their work made a difference. They build systems that can explain how, where, and under what conditions difference was made. That requires a shift away from certainty theater, the performance of confidence, and toward causal intelligence, the disciplined practice of linking action, assumptions, context, and evidence.

That shift changes everything. It makes planning more rigorous, because assumptions are explicit. It makes evaluation more credible, because claims are proportionate. It makes learning faster, because failures point to specific broken links rather than vague disappointment. And it makes organizations more useful to the people they serve, because interventions are less likely to be built on comforting fantasies.

In the end, the deepest lesson is not that impact is hard to measure. It is that impact is hard to produce. And the only serious response to that difficulty is to think better about causation before, during, and after action.

The question is not whether you can guarantee results. You cannot. The question is whether your theory of change is honest enough to be tested, and whether your contribution story is strong enough to survive reality.

That is where real learning begins.