Contribution Analysis¶
Evaluation method — instantiates Causal Mechanism Mapping
Builds a plausible contribution story — assembling evidence and weighing other influences — when a clean control or randomized comparison is unavailable.
Contribution Analysis is the method for the honest middle ground: you cannot run a controlled experiment, so you will never isolate your intervention's effect, but you still need a defensible answer to "did it help, and how much?" Its defining move is to reframe the question from attribution ("this and only this caused the outcome") to contribution ("this plausibly made a meaningful difference, alongside other factors"). It does that by assembling a contribution story and then actively hunting for everything else that could explain the result — other programs, favorable trends, selection — and showing, with evidence, that the outcome still moved in a way those rivals do not fully account for. It is affirmative reasoning under acknowledged messiness: not a control group, but a disciplined argument that survives its own strongest counter-explanations.
Example¶
A philanthropic foundation funds a rural sanitation program across dozens of villages in a developing region, aiming to reduce childhood diarrheal disease. Two years in, reported cases have dropped. There was never any prospect of a randomized trial — the program went where local partners could operate, not to a random sample — so the evaluator uses contribution analysis. She first writes the expected contribution story: latrines built → open defecation falls → water-source contamination drops → fewer infections. Evidence is gathered along it: latrine-use surveys, water-quality samples, clinic records.
Then comes the method's real work — the rival column. That same period saw a national measles vaccination drive (which cuts child mortality broadly), an unusually dry two years (fewer floods, less contamination), and the fact that the most organized villages volunteered first (selection). For each rival she asks what the data would look like if it were the true driver, and checks. Diarrheal cases fell but measles-preventable deaths tracked separately; the decline held in the wetter districts too; and late-joining, less-organized villages showed the same drop as early ones, denting the selection story. Her conclusion is scoped, not triumphant: within these districts and this two-year window, the sanitation program plausibly contributed a substantial share of the decline, with the dry spell a real but partial co-cause. That calibrated verdict — earned by surviving its rivals — is exactly what a funder needs to decide on renewal.
How it works¶
- Start from an explicit contribution claim and its story. State what contribution is plausible and the pathway by which the program would produce it; the story is the thing to be corroborated or dented.
- Gather evidence along the story. Assemble observations, monitoring data, and accounts that the intermediate steps actually occurred — not just that the endpoint moved.
- Enumerate and confront rival explanations. List every other plausible driver — co-occurring programs, secular trends, selection, measurement changes — and for each, derive what the evidence would show if it were dominant, then check.
- Assemble a scoped contribution verdict. Conclude how much the program plausibly contributed, explicitly bounded to the population, period, and conditions where the evidence holds — never a universal claim.
Tuning parameters¶
- Rival breadth — how many alternative explanations are seriously entertained. More rivals make the verdict more defensible but lengthen the work; a lazy rival list is the method's most common shortcut.
- Evidence triangulation — how many independent sources must agree before a link is treated as supported. Higher bars resist wishful reading but may leave the story unresolved.
- Contribution granularity — whether the conclusion is "contributed / did not" or a graded share ("a substantial but not sole part"). Graded verdicts are more honest and harder to defend crisply.
- Scope tightness — how narrowly the claim is bounded to population, period, and setting. Tighter scope is safer but less quotable; loose scope invites overgeneralization.
- Skeptic involvement — whether an outside challenger stress-tests the rival list. Involving one is the strongest guard against a self-serving story.
When it helps, and when it misleads¶
Its strength is giving a credible, calibrated causal answer exactly where the cleaner methods cannot run — messy social programs, half-implemented policies, anything where randomization was never on offer. Done well, it neither overclaims sole credit nor throws up its hands; it produces a contribution verdict a decision-maker can act on, with its rivals visible.
Its failure mode is the confirmatory contribution story — a narrative built to flatter the program, with a rival list that is present but toothless, each alternative named and then waved away rather than genuinely tested.[n1] The classic misuse is treating the existence of a plausible story as proof, when plausibility is cheap and the whole discipline lives in whether the story survived its rivals. The guarding discipline is to make the rival explanations do real work: for each, specify in advance what evidence would favor it, then report honestly when some of that evidence shows up — a contribution analysis with no surviving co-causes is usually one that did not look hard.
How it implements the components¶
Contribution Analysis fills the affirm-under-messiness components — the ones a no-control method must carry:
candidate_cause— the program or intervention whose contribution is argued; named as one driver among several from the outset.rival_explanation_set— its signature contribution: alternative drivers are enumerated and confronted with evidence, and the verdict is only as strong as the rivals it outlasts.causal_evidence_record— evidence is assembled along the contribution story and for each rival, forming the corroboration the claim rests on.causal_claim_scope— the verdict is explicitly bounded to the population, period, and conditions where the evidence holds.
It does not audit statistical identification — confounder_check and counterfactual_anchor (a control or comparison group) are the province of Causal Inference Review; contribution analysis is used precisely when that clean comparison is unavailable, so it argues contribution rather than adjudicating identification.
Related¶
- Instantiates: Causal Mechanism Mapping — supplies a credible causal verdict where controlled comparison is impossible, keeping rival explanations central.
- Consumes: Theory of Change Model — the program's stated theory supplies the contribution story that this method corroborates or dents.
- Sibling mechanisms: Causal Inference Review · Intervention Test · Mechanism Map · Causal Diagram · Causal-Loop Map · Failure Tree Analysis · Process Tracing · Root-Cause Analysis
Editorial Notes¶
Form Classification¶
Form family: Assessment, Review & Assurance
Rationale: The method evaluates an explicit contribution story against pathway evidence and rival explanations and issues a scoped verdict about the program's contribution, so its operative form is evaluation.
Nearest alternative: Analysis, Modeling & Optimization — Causal analysis organizes and weighs the evidence, but the defining output is an evidence-based evaluative finding rather than a prediction or formal model.
Review outcome: Adjudicated after independent review; high confidence.
Origin Attribution¶
Primary origin: Public Administration & Policy
Origin pattern: Single lineage
Present-day reach: Specialized
Rationale: John Mayne's public-program evaluation framework cohered contribution analysis as a tested causal story that weighs other influencing factors when experimental attribution is unavailable.
Review resolution: Both reviewers agree on public-administration provenance. Statistical causal methods may be used within contribution analysis, but they are comparators and evidence tools rather than an independent origin of the contribution-claim method.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
[n1] Contribution analysis is associated with John Mayne's framework for public-program evaluation, whose central discipline is the serious treatment of "other influencing factors." The method's credibility rests entirely on rivals being tested rather than listed — a decorative rival column is the standard way the method is hollowed out. ↩