Contribution analysis

Contribution, not attribution

Precise wording matters here more than anywhere else in this cluster. The attribution question asks how much of an observed change the programme alone produced — the question counterfactual designs answer by comparing outcomes with and against a world without the programme. The contribution question asks something different: given that the result occurred, did the programme make a difference to it, in what way, and alongside what else [1, 2]?

Contribution analysis exists because the attribution question is often unanswerable and sometimes the wrong question. Policy influence, capacity building, institutional reform and advocacy work through long, tangled chains involving many actors; there is no comparison group of parallel ministries, and the programme never claimed to be the sole cause. Mayne’s insight was that a rigorous, evidence-disciplined answer to the contribution question is still possible — and far more useful than either a spurious number or an unexamined success claim [1].

The method treats the intervention as part of a causal package: a set of causes — the programme plus other actors’ efforts, enabling conditions, external events — that together were sufficient to produce the result. The analytical claim being tested is that the programme was a necessary part of that package: without it, the package as it operated would not have delivered the result [2]. Naming the other package members is not a concession that weakens the story; it is what makes the story checkable.

The six steps

Mayne’s process, as set out in ILAC Brief 16, is a loop, not a pipeline — the explicit expectation is that early passes expose weaknesses that later passes repair [1]:

1. Set out the cause–effect question to be addressed. Which result, which contribution claim, over what period, at what level of the results chain — and how strong does the answer need to be for the decisions it will inform?
2. Develop the theory of change and the risks to it. Make the programme’s logic explicit link by link, including the assumptions each link rests on and the external factors that could produce — or block — the result. Contribution analysis presupposes a theory of change; it does not teach you to build one, and neither does this page — the Monival explainer covers the artefact itself.
3. Gather the existing evidence on the theory of change. Monitoring data, prior evaluations, research and administrative records, mapped link by link: which parts of the chain are already evidenced, and how well?
4. Assemble and assess the contribution story. Draft the argument: the result occurred; the theory of change’s links are evidenced to such-and-such a degree; these other factors also operated; here is why the programme’s contribution is (or is not yet) plausible. Assess it as a critic would — where is it weakest?
5. Seek out additional evidence. Target new data collection at the weak links and the live rival explanations identified in step 4 — not at the links already well evidenced.
6. Revise and strengthen the contribution story. Rewrite, then judge whether another iteration is warranted or the story is robust enough for its intended use.

Weighing evidence and rival explanations

The step most often skipped — and the one that decides whether the exercise is analysis or advocacy — is the explicit treatment of rival explanations [1, 2]. For each observed result, list what else could account for it: other programmes operating on the same population, secular trends, policy changes, selection of favourable sites, measurement artefacts. Then do the work: for each rival, state what evidence would be expected if it were the main driver, and check. A contribution story is strengthened as much by documented reasons to discount rivals as by evidence for the programme’s own links.

Evidence weighing should be visible on the page. For every link in the theory of change, a defensible analysis records: the evidence for it, the evidence against it, the strength of both, and the resulting confidence in the link. Where the method is combined with process tracing — an increasingly common pairing — the evidence tests on that page give this weighing a sharper edge: some single pieces of evidence are far more probative than volumes of weak consistency, and the tests say which.

What a defensible contribution story looks like

The end product is a narrative with structure and stated confidence, roughly: the result occurred and is well measured; the programme’s theory of change is explicit and its links are evidenced to stated degrees; the other significant influencing factors were X and Y, and their roles are described; the main rival explanations were examined and discounted for stated reasons; therefore the programme very likely made a material contribution of this kind — with these remaining uncertainties [1, 2].

Contrast the common abuse: a results narrative that recites activities, asserts outcomes, and labels itself a contribution analysis without a theory of change, without rivals examined, and without any statement of confidence. The label does not confer the rigour; the six steps do — and steps 4 through 6 are where the rigour lives. Reports should also resist converting the story into an unearned number: “a 40 % contribution” has no meaning the method can support. The evaluation reports page in the reporting cluster covers how to present structured causal narratives credibly.

Where the field disagrees

Contribution analysis sits on one side of this cluster’s central divide. The counterfactual school holds that causal claims require a comparison — observed or constructed — and treats theory-based narratives as, at best, hypothesis generation; the randomised controlled trials page presents that position and its own internal critics. The theory-based school, articulated for evaluation by White among others, replies that a causal chain tested link by link against evidence is a legitimate mode of causal inference — and the only mode available for many of the questions that matter most, from policy influence to systems change [4]. Mayne’s own position is deliberately modest: contribution analysis does not claim to out-perform an experiment where an experiment is feasible; it claims that where one is not, “we did our best with a weak design” is not the only alternative to silence [1, 2].

The working synthesis this site recommends mirrors the RCT page’s: choose the design the question and setting can support. When a counterfactual design is feasible and answers the actual question, use it — and note that even then, a contribution-style analysis of the causal chain makes the result interpretable [4]. When it is not feasible, run contribution analysis properly — all six steps, rivals included — rather than running nothing and asserting success.

Checklist before you commit to contribution analysis

• The question is genuinely a contribution question, and counterfactual designs have been considered and ruled out for stated reasons.
• An explicit theory of change exists (or will be built first), with assumptions and external factors identified link by link.
• Existing evidence has been mapped against the chain before new collection begins.
• Rival explanations are listed, investigated and addressed in writing.
• Other members of the causal package are named in the story, not airbrushed out.
• The final story states its confidence level and remaining weaknesses, and the report never converts the narrative into a fabricated attribution percentage.