Attribution

“Attribution is the difference between change after a programme and change because of a programme. ”

Attribution is the claim behind many impact statements. Yields rose after training, so the training caused the increase. Forest loss fell after a sourcing policy, so the policy prevented deforestation. Household income improved after certification, so certification created the gain. Sequence is suggestive. It is not sufficient.

The OECD evaluation glossary describes attribution as establishing a causal link between an observed change and a specific intervention. The challenge is that the world does not pause while a programme operates. Weather, prices, policy, migration, other projects and people's own decisions can affect the same outcome. A familiar example is a yield increase.

Farms participating in a programme rise from 500 to 650 kilograms per hectare. The programme reports a 30 per cent effect.

Yet non-participating farms in the district rose similarly because rainfall improved and disease pressure fell. The before-and-after change is real; the attributed effect may be small. The central problem is the missing counterfactual: the same units cannot be observed at the same time with and without the intervention.

Evaluation designs estimate that alternative through random assignment, matched comparison, difference-in-differences, regression discontinuity, synthetic control or other methods. Each depends on assumptions. Randomised controlled trials can strengthen attribution where assignment is ethical and feasible.

They are not automatically appropriate for system change, regulation, landscape programmes or severe human rights risks. Randomisation can estimate an average effect while saying little about mechanism, implementation quality or why effects differ across groups.

Quasi-experimental methods use naturally occurring or designed comparison. Their credibility depends on whether the comparison represents what would have happened to participants. Unobserved differences, spillovers and selective participation can bias estimates. Statistical sophistication cannot substitute for a plausible design. Qualitative evidence also supports causal attribution.

Process tracing examines whether expected causal steps occurred and whether alternative explanations fit the evidence. Interviews, documents, timing, mechanism tests and contradictory cases can strengthen or weaken the claim. The question is not quantitative versus qualitative, but whether evidence discriminates among explanations.

Attribution is often unnecessary at the strongest level. Complex sustainability outcomes rarely have one cause. A company may contribute to reduced deforestation alongside enforcement, market change and community action. Claiming exclusive attribution can be less credible than demonstrating a material, evidence-based contribution. The level of claim should match the design.

Activity completion may require administrative evidence. Outcome contribution may be supported through a strong theory of change and triangulation. A quantified causal effect requires a design capable of estimating it. Public language should not outrun method. Distribution matters. An average attributed effect can conceal harm or failure among subgroups.

Training may improve yields for farmers with access to credit and reduce income for those who take on debt without achieving the expected response.

Causal analysis should examine heterogeneity, not only the mean. Attribution also carries accountability. When positive outcomes are claimed, organisations may take credit. When negative outcomes occur, they may point to context. The same causal discipline should apply in both directions, including unintended effects and cost transfers. Attribution may vary across components and groups.

A programme can be responsible for making credit available, partly responsible for adoption and only one influence on income. It may have a strong effect in one region and none where infrastructure or land constraints bind. A single headline percentage can flatten this variation and invite a broader claim than the evidence supports. Causal language should therefore be graduated.

Randomised or strong quasi-experimental evidence may justify saying an intervention caused an average effect under specified conditions. Observational evidence may support association or a credible contribution. Mechanistic and qualitative evidence can explain how change occurred. Precision in wording is not timidity; it enables readers to understand the strength, population and limits of the inference.

The discipline is to ask what else could explain the change and what evidence would distinguish the intervention from those alternatives. Attribution is not a reward for being present before improvement. It is a claim that must survive comparison.

Practical application

Define the causal question and counterfactual before data collection. Select experimental, quasi-experimental or theory-based methods suited to the intervention and ethics. Record assumptions, spillovers, selection and external changes. Use triangulation and examine subgroup effects.

Phrase conclusions according to strength of evidence, distinguishing association, contribution and attribution rather than using impact as a generic label. Pre-specify the main causal questions, comparison strategy, outcomes and subgroup analysis where feasible. Record departures and test sensitivity to alternative specifications.

Match public language to the design: caused, increased, was associated with, or plausibly contributed to are not interchangeable. Review positive and negative effects under the same standard of causal evidence.

Why it matters

Attribution influences funding, reputation and replication. Weak causal claims can scale ineffective programmes, reward the wrong actors and hide the forces that actually produced change.

Common misconception

A change observed after an intervention is often treated as attributable to it. Temporal sequence is necessary for causation but does not rule out other explanations or show what would have happened anyway.

Connections

Counterfactual provides the comparison underlying attribution. Contribution offers a more appropriate causal claim in complex systems. Evaluation tests the evidence, while Additionality asks whether the change goes beyond the baseline scenario.

A question worth asking

What evidence would make you conclude that the observed improvement would have happened without your programme?

Selected references

OECD. 2023. Glossary of Key Terms in Evaluation and Results-Based Management for Sustainable Development, Second Edition. Gertler, P. J. et al. 2016. Impact Evaluation in Practice, Second Edition. World Bank. Shadish, W. R. , Cook, T. D. and Campbell, D. T. 2002. Experimental and Quasi-Experimental Designs for Generalized Causal Inference. Rubin, D. B. 1974.

Estimating Causal Effects of Treatments in Randomized and Nonrandomized Studies. Journal of Educational Psychology 66(5): 688-701. White, H. and Phillips, D. 2012. Addressing Attribution of Cause and Effect in Small n Impact Evaluations.