What is the difference between attribution and contribution?
Attribution asks whether—and often how much—an intervention caused an observed change. Contribution examines the part an intervention played alongside other influences. Attribution does not mean the program was the only cause. Contribution is not a substitute for evidence. Both require reasoning that fits the question, context and available data.
Suppose employment increases after a training program. The increase is an observed outcome. It does not, by itself, tell you how much resulted from the program, a stronger labor market, participants’ own efforts or other support. The evaluation question is how to investigate those explanations.
This guide is for teams preparing an impact claim, commissioning an evaluation or deciding what evidence to collect. It focuses on program and policy evaluation, rather than marketing attribution or assigning credit to individual staff.
Attribution vs contribution, side by side
| Question | Attribution | Contribution |
|---|---|---|
| What are we trying to understand? | Whether a change can be causally linked to an intervention; sometimes the size of that effect. | How an intervention helped produce change within a wider set of causes. |
| Must the intervention be the only cause? | No. An effect can exist alongside many other influences. | No. Other influences are part of the explanation being examined. |
| What evidence might be used? | An appropriate causal design, which may use experimental, quasi-experimental or theory-based methods. | Evidence testing the proposed pathway, implementation, outcomes and competing explanations. |
| Does a before-and-after result settle the question? | No. Other changes can account for the difference. | No. The mechanism and alternatives still need examination. |
| Is a percentage of credit always the output? | No. The answer depends on the design and question. | No. A supported explanation need not assign numerical shares. |
| What must the report explain? | The causal reasoning, assumptions, uncertainty and limits of the estimate or conclusion. | Which parts of the contribution story are supported, challenged or still uncertain. |
The approaches can complement one another. A study might estimate an average effect and also investigate why it varied across settings. The Magenta Book’s analytical-methods guidance describes several evaluation approaches; choosing one starts with the question, not a preference for stronger-sounding wording.
A worked example: employment after training
In this fictional example, a team enrolls 100 people. Six months later, it has employment information for 70, of whom 42 report paid work. The other 30 have not provided a follow-up response.
What the records establish
The team can report that 42 of the 70 respondents, or 60%, reported paid work at follow-up. It should also state that follow-up coverage was 70 of 100 enrolled participants. It cannot treat the missing 30 as employed or unemployed without further evidence, or present 60% as the employment rate for the whole enrolled group.
What a contribution investigation adds
The team examines whether relevant training was delivered, whether participants used the skills, how employer introductions worked and what other support they received. It interviews participants with different outcomes, including those who did not find work, and checks records that can corroborate their accounts. It also reviews labor-market changes and access barriers.
If this evidence supports a plausible pathway, the report can explain the role the program appears to have played and where the explanation remains uncertain. Selecting a few positive quotations would not be enough.
What an attribution study would need to address
To estimate the program’s effect on employment, the team needs a credible account of what would have happened without it. Depending on feasibility and design, that might involve random allocation, a suitable comparison group, a policy threshold or another defensible strategy. Simply comparing participants with all nonparticipants would risk differences in motivation, eligibility and circumstances driving the result.
The illustrative records above contain no such design. They support a descriptive result and further investigation, not a numerical causal-effect claim.
Choose the question before the method
- “Did outcomes change?” Begin with reliable outcome definitions, timing, coverage and comparable observations.
- “How did the program help?” Examine the proposed mechanism, what was delivered and other explanations.
- “How much difference did the program make?” Plan a causal evaluation capable of answering that question.
- “Why did it work differently across sites?” Investigate implementation, participant differences and context, potentially alongside effect estimates.
A small team does not need to promise an expensive study for every decision. It does need to match the strength of its claim to the work it can do. Set that expectation with funders or leadership before collection begins. The UK guidance on quality in impact evaluation emphasizes appropriate, proportionate design and the credibility of the counterfactual.
Build a contribution claim that can be challenged
Contribution analysis is a structured approach, not simply adding the word “contributed” to a success story. It tests a proposed explanation of change against evidence and considers other causes. See BetterEvaluation’s contribution-analysis guidance. The following is a practical planning checklist for a delivery team.
- State the claim precisely. Name the outcome, people or organizations, setting and period. “We improve lives” is too broad to investigate.
- Write the proposed pathway. Explain what the intervention is expected to change first and how that could affect the later outcome. Use a theory of change to make assumptions visible.
- Check what actually happened. Intended delivery is not necessarily actual delivery. Document access, participation, timing and variations between sites.
- Examine the outcome evidence. Check measurement quality, missing data, contradictory findings and whether the timing makes sense.
- Investigate alternatives. Identify other services, personal circumstances, policy changes and environmental influences that could explain the result.
- Review and qualify the explanation. Seek evidence that could weaken it. Document unresolved questions and decide what additional investigation would be useful.
The strongest next step may be to collect an overlooked source rather than another satisfaction survey. For example, if a claim depends on people using a new skill, evidence about access to real practice may matter more than an additional rating of the workshop.
Keep a claim-to-evidence record
| Record element | What to retain | Why it matters |
|---|---|---|
| Claim and version | The exact proposed statement, its scope and revision date. | A reviewer should know which claim was examined. |
| Outcome definition | Measure, unit, population, period and denominator. | Similar labels can conceal different calculations. |
| Source evidence | Relevant observations, documents and accounts with dates and provenance. | A summary should be checkable against its sources. |
| Alternative explanation | Other plausible influences and evidence for or against them. | The preferred explanation should not be the only one investigated. |
| Coverage and uncertainty | Missing observations, measurement limits and unresolved questions. | A precise number can still rest on incomplete evidence. |
| Review decision | Reviewer, interpretation, approval and follow-up action. | Source retrieval and judgment are different parts of the work. |
Preserve privacy when connecting evidence. Reviewers may need access to source records, but public reports generally do not need names or identifiable narratives. Set access and consent rules for the purpose of the evaluation. Traceability means authorized people can inspect the working; it does not mean publishing personal information.
Compare sites without forcing one survey on everyone
A program operating across schools, chapters or delivery partners may have locally different instruments. Agree the small set of common outcomes and registration fields needed for comparison, then document them in a data dictionary. Keep local questions where they serve a real purpose.
Specify which definitions and time windows are shared, what can be mapped and what cannot. A three-month job-retention measure cannot simply be pooled with “ever had a job after joining.” Differences in context may also explain why a pathway works in one location and not another. Aggregation should preserve those distinctions instead of hiding them.
What Sopact can help with—and what still needs judgment
Sopact’s practical role is to connect collection, context, analysis and governance so the evaluation team can work from an organized body of evidence. Quantitative observations, open-ended responses and supporting records can be considered together rather than repeatedly assembled from disconnected exports.
For qualitative evidence, people still define and review the coding framework. Applying those definitions consistently across a larger dataset can make it easier to investigate patterns, exceptions and alternative explanations. An AI-generated theme or a retrieved quotation is not, on its own, a causal finding. Reviewers must assess relevance, quality and competing explanations.
The total effort includes organizing instruments, matching records, maintaining definitions, recoding after revisions and rebuilding reports. Test the workflow with one real claim: change a definition, inspect affected records, reproduce the calculation and review the cited evidence. Measure the repeated work your team actually avoids instead of assuming a universal saving.
No platform can repair an unsuitable evaluation design merely by making a dashboard easier to inspect. A causal estimate also needs a defensible analytical approach. Use outcome evaluation and impact measurement for the wider planning context.
Write a report that makes the distinction clear
Use separate sentences for the observed result, interpretation and limitation. In the fictional training example, an initial report could say: “At six months, 42 of 70 respondents reported paid work; 100 people enrolled. We are investigating the role of training, employer connections and other support. The follow-up data alone does not establish the program’s effect on employment.”
If further analysis supports a contribution claim, explain which evidence supports the mechanism and which alternatives remain. If a causal study estimates an effect, describe the method and uncertainty. Do not assign the program an arbitrary percentage of credit because several organizations were involved.
For practical reporting help, use the How to Write an Impact Report ebook, explore report examples, or start with the impact report template.
Frequently asked questions
Does attribution mean claiming all the credit?
No. A causal effect can be estimated in a setting with multiple influences. “The program was the sole cause” is a much stronger claim and is not the definition of attribution.
Is contribution always easier to demonstrate?
No. A credible contribution explanation requires careful investigation of the pathway and alternatives. It can be demanding, especially in a complex setting with incomplete evidence.
Does attribution always require a randomized control group?
No. Several approaches can support causal inference, depending on the question and assumptions. Randomization is one option; quasi-experimental and theory-based approaches may also be relevant. Select the design with suitable evaluation expertise.
Can participant comments establish causation?
They can provide valuable evidence about experience and mechanisms, but they may be incomplete or affected by recall and reporting biases. Assess them alongside other evidence and plausible explanations.
Can we report outcomes before completing a causal evaluation?
Yes. Report what was observed, how it was measured and who is represented. Clearly separate those observations from any claim about the program’s causal role.
Should every impact report include a percentage attributed to our work?
No. Only report a quantitative causal estimate or allocation of credit when the method supports it. A well-supported explanation with clear limits is more useful than an invented percentage.
Watch: the reasoning behind a theory of change
This introduction explains outputs, outcomes and the proposed pathway to change. Use it as background to the framework; a diagram or video does not establish a program’s causal effect.
Choose your next step
King Center
See how participant feedback supports program review and learning. This story is separate from the fictional teaching examples above.
Read the customer story →