9 Contribution Analysis and Theory of Change
Contribution analysis assesses whether an intervention made a credible contribution to an observed outcome in a setting where several influences may be operating (Mayne, 2001). It does not establish attribution by removing every other cause. It develops and tests a contribution story: an account of how the intervention was expected to matter, what evidence supports each link, and which alternatives remain plausible.
The method is most closely associated with John Mayne’s formulation of contribution analysis and its iterative contribution story (Mayne, 2001, 2012). Ton (2017) is included for a different reason: it shows how that framework can organise a real mixed-method evaluation of an innovation fund. The distinction matters. A foundational source defines the method’s inferential standard; an application shows how that standard was pursued in a particular programme and setting.
The bibliometric map (Chapter 5) retrieves 10,791 candidate records for the phrase “contribution analysis,” narrowing to 1,715 title-and-abstract pairs after document-type filtering. That count cannot be read as a comparative measure alongside the other method families in this book: the same phrase also names engineering source-apportionment and economic decomposition methods that have nothing to do with Mayne’s framework, and the record-level screening needed to separate these uses from evaluation applications is not yet complete. The figure is reported here as a discovery count, a reason to keep searching the literature, not as an estimate of how often contribution analysis is used in evaluation. No subfield, venue, or trend detail is presented for it in this chapter for the same reason.
9.1 Theory of change as a precursor
A theory of change is not a method in its own right. It is a structured account of the proposed causal process. It becomes useful for contribution analysis when it specifies the activities, intermediate changes, assumptions, contextual conditions, and rival processes that matter to the claim. A linear list of inputs, outputs, and outcomes does not do this work unless it states why one stage is expected to lead to the next and what observation would cast doubt on that expectation.
Each link should be stated in a form that can be investigated. The evaluator needs to establish whether the intended activity occurred, whether the proposed intermediate change followed in the relevant period, whether the outcome changed as expected, and what else could account for the change. The theory should also identify conditions under which the sequence may fail. These conditions are not a decorative risk register; they direct attention to the parts of the contribution story that need evidence.
9.2 Building and testing a contribution story
Mayne’s framework begins by defining the contribution question and the intervention’s claimed role (Mayne, 2001, 2012). The evaluator then develops or refines the theory of change, assembles existing evidence for its links, assesses the resulting contribution story, and seeks further evidence where it remains weak. The sequence is iterative. New evidence may narrow the claim, identify an untested assumption, or show that an alternative explanation accounts for part of the outcome.
Contribution analysis draws on mixed evidence: monitoring data, administrative records, implementation information, interviews, surveys, prior research, and case studies. The volume of material is not the standard. Evidence should bear on a specific link or rival. An implementation record may show that an activity occurred; it does not establish the behavioural or institutional change expected to follow. A trend in the outcome may be compatible with the intervention’s contribution but may also reflect another change. The contribution story should state what each source establishes and what it cannot establish.
The assessment of links should remain graduated. Some links may be well supported, while others are plausible but insufficiently tested. A contribution claim is weakened when a central link lacks evidence or when a strong rival remains unresolved. This is useful information for commissioners: it identifies whether an evaluation needs better monitoring, a focused within-case investigation, a comparison design, or a revision of the intervention theory. The method does not require every link to be confirmed before the findings can inform adaptation, but it does require the uncertainty to be visible.
9.3 Defining the contribution question
Contribution analysis begins with a claim that can be investigated. ``Did the programme work?’’ is usually too broad. The evaluator needs to specify the intervention, the outcome, the population or setting, the period, and the contribution being claimed. A programme may claim to have contributed to the adoption of a policy, a change in organisational practice, improved access to a service, or a population-level outcome. These are different questions. Evidence that an organisation adopted a new procedure cannot establish a contribution to a later health or learning outcome without evidence about the intervening links.
The contribution question should also identify the decision for which the evidence is needed. A commissioner deciding whether to continue a programme may need to know whether implementation reached the intended groups and whether the assumptions behind delivery are holding. A commissioner deciding whether to scale may need a stronger account of the conditions under which the programme contributed and evidence about outcomes across sites. Contribution analysis can serve both purposes, but the evidence plan and scope of the conclusion will differ.
The initial contribution story should distinguish activities, outputs, intermediate outcomes, and final outcomes. More importantly, it should explain the causal proposition between them. It needs to specify how programme resources are selected and used, what capability or practice is expected to change, how that change affects subsequent activity, and why this should influence the outcome. The proposition may rely on assumptions about authority, demand, complementary resources, political support, or the behaviour of other institutions. These assumptions are part of the causal account. They should be investigated where they are necessary for the claimed contribution.
9.4 Designing a contribution analysis
The design can start with a contribution matrix. For each link in the theory, the matrix records the claim, the evidence already available, its limitations, the alternatives that need attention, and the additional inquiry proposed. This creates a disciplined link between the theory of change and the data collection plan. It also helps prevent a familiar problem: a large evaluation that collects substantial information on activities and participant views but little evidence on the links that make the contribution claim credible.
| Contribution claim | Existing evidence | Main limitation or rival | Further inquiry |
|---|---|---|---|
| [Specify the contribution claim] | [Record available evidence] | [Identify the limitation or rival] | [Specify the targeted inquiry] |
The evidence plan should include both expected and adverse observations. It should ask what would count as evidence that the proposed practice did not change, changed only temporarily, or changed for another reason. It should also identify the sources most likely to be independent. Several monitoring reports prepared for the same donor may provide useful detail but may not constitute separate corroboration. Administrative records may be contemporaneous but incomplete. Interviews may explain decisions but be shaped by recall and respondents’ interest in the programme. The design should state how these limitations will affect the weight given to each source.
Contribution analysis does not prescribe one data-collection method. It can combine routine monitoring, document review, interviews, surveys, outcome data, process tracing, and comparison across cases. The choice follows from the contribution story. A weak implementation link may require direct observation or a focused process trace. An outcome claim may require a counterfactual comparison. Heterogeneity across sites may justify a configurational analysis. Using several sources is valuable when they address different aspects of the claim. It is less useful when they are different versions of the same programme account.
9.5 Testing assumptions and alternative explanations
The evaluator should treat assumptions as propositions that may hold, fail, or hold only under defined conditions. The evidence plan should identify the records, observations, and accounts that could establish whether each necessary condition was present. Stating assumptions in this way prevents them from becoming a background list that is acknowledged but never investigated.
Alternative explanations must be selected from knowledge of the setting and the outcome. A service improvement may have been influenced by a concurrent policy, a change in funding, a change in staff, an epidemic, altered measurement, or activity by another organisation. The point is not to list every conceivable influence. It is to identify alternatives that could plausibly account for the observed change or for a substantial part of it. The evaluation should explain why each alternative was considered and what evidence bears on it.
An alternative can be relevant at a particular link rather than to the whole story. A programme may have genuinely changed staff practice while a separate funding increase explains the final outcome. Conversely, an external policy may have triggered the practice change, leaving the programme’s training as a supporting rather than initiating factor. The final contribution story should allow for such mixed findings. It is stronger when it identifies the intervention’s role and the role of other processes than when it allocates all observed change to one programme.
9.6 An iterative workflow
The work can be organised in six stages. First, define the contribution question and the case or programme boundary. Second, develop the initial contribution story, including assumptions and credible alternatives. Third, assemble and assess existing evidence link by link. Fourth, identify the links and rivals for which the evidence is insufficient or dependent. Fifth, collect targeted additional evidence and revise the story where the evidence requires it. Sixth, state the resulting contribution claim, the alternatives that remain, and the conditions under which the claim applies.
Iteration is a substantive part of the method. A revised story is justified when the evidence shows that an activity was implemented differently from the original theory, that an expected mechanism did not operate, or that an overlooked condition affected the outcome. Revision should be documented. The original account, the evidence that led to change, and the implications for the conclusion should remain visible. Rewriting the theory only after the outcome is known can turn contribution analysis into a coherent narrative with no clear test of its causal propositions.
The final assessment should not collapse all links into a single score. It may find convincing evidence that a programme reached its intended organisations, mixed evidence that it strengthened their capacity, and insufficient evidence that changes in capacity affected members’ outcomes. It may also identify a programme contribution in some sites and no contribution in others. These distinctions give commissioners a basis for adaptation and for deciding whether a later evaluation needs a different design.
9.7 Working with an existing theory of change
Many evaluations inherit a theory of change from a funding proposal, programme document, or workshop. These documents can be a useful starting point, but they often mix several purposes: communicating a programme to funders, setting out operational activities, and making a causal claim. The evaluator should not treat the document as a settled explanation. It may omit decisions made after the programme began, assume that outputs will be used, or give little attention to organisations and policies outside the programme’s control.
The first task is to establish which version of the theory governed the period under evaluation. Programmes change. Activities are added or dropped, target groups are redefined, delivery moves between organisations, and the stated outcome may remain the same while the practical route towards it changes. The evaluation should record these changes and ask whether the contribution claim concerns the original design, the implemented intervention, or a later adapted form. Treating them as one intervention can obscure the reason for a result.
Interviews and workshops with programme staff can help reconstruct the theory, but they should not be the only source. Delivery plans, budgets, contracts, routine data, policy documents, and accounts from participants can reveal a different account of what the programme did and why. They may also show that a causal link assumed by managers was not available to front-line staff or participants. The final theory should reflect the evidence, including disagreement. It is not necessary to secure consensus on every arrow before inquiry begins.
An existing theory may be too extensive for a single evaluation. A national programme can have dozens of activities, several intended outcomes, and a long list of conditions. The study should identify the contribution claims that matter for the decision at hand and concentrate on the links that are uncertain, consequential for the claim, or vulnerable to alternative explanations. This is a matter of scope, not a reason to describe the unexamined parts of the theory as supported. The report should state which pathways were selected and what remains outside the study.
9.8 Data architecture and evidence use
Routine monitoring data are often the starting point for contribution analysis. They can show whether activities were delivered, who was reached, whether implementation differed across sites, and when a change was recorded. They can also reveal gaps in the programme’s evidence base. If monitoring records only training sessions delivered, the evaluation cannot use them to establish whether staff changed a routine. If outcome indicators are aggregated annually, they may be unable to show whether a change followed implementation in the relevant period.
The evaluator should examine how each dataset was generated. This includes the definition of the measure, who recorded it, incentives for accurate reporting, missing values, changes in coverage, and any revisions to the indicator. A rising programme-reported outcome can reflect a real improvement, expansion of reporting, changed definitions, or a change in which sites submit data. These possibilities do not make monitoring unusable. They determine which claim the data can support and what supplementary evidence is needed.
Linking sources is often more informative than adding more sources. A delivery record can be linked to a site-level implementation account, a change in organisational practice, and an outcome measure for a defined period. The linkage may show that a reported outcome improvement occurred where delivery was weak, or that apparently similar sites implemented different versions of the intervention. It can also identify cases for closer investigation. The records need not be perfect for this purpose, but the analysis should state where matching is incomplete or timing uncertain.
Data collection should include actors who can speak to the programme’s assumptions and alternatives. Programme managers can describe intended delivery and adaptations. Participants can describe access, use, and barriers. Front-line staff can describe operational constraints. Officials, other providers, or local organisations may identify concurrent processes. The sampling logic should follow the causal links. A contribution story about a decision made by local government is poorly tested if the evidence comes only from the implementing organisation and programme beneficiaries.
The team should decide early how it will handle confidential information and evidence that cannot be published. A confidential document may be reviewed and described in a way that protects the source while allowing a reader to understand its bearing on a claim. An anonymous assertion that cannot be corroborated should carry limited weight. The report should also identify important sources that were unavailable, such as financial records, records held by an implementing partner, or accounts from a group that could not safely be interviewed. Access constraints are part of the evidence assessment, not an administrative footnote.
9.9 Programme adaptation and causal continuity
Contribution analysis is often used while a programme is still adapting. Adaptation can be a legitimate response to evidence about delivery or context. It creates a problem for causal claims when the evaluation continues to describe several materially different versions as the same intervention. The study should identify the changes that affect the contribution story: changes in eligibility, intensity, delivery organisation, resources, timing, or the activities expected to produce the outcome.
Some adaptations preserve causal continuity, while others alter the mechanism, the population reached, the selection process, or the conditions under which an activity can contribute to change. The theory and evidence plan need to be revised accordingly. The final claim may apply only to a particular phase or implementation model.
Adaptation also creates opportunities for learning. A programme that documents why a delivery change was made, which sites adopted it, and what happened afterwards creates evidence for a revised contribution story. This does not turn an adaptation into a controlled experiment. It can nevertheless identify whether the expected process became more plausible, whether an implementation constraint was addressed, and what new alternatives need to be considered. The evaluation should avoid treating every adjustment as confirmation that the programme is learning effectively; the evidence should show what changed and what followed.
9.10 Roles, challenge, and governance
The programme team often holds essential information about its theory, delivery, and records. Its involvement can strengthen the evaluation when it exposes assumptions, identifies missing data, and enables access to the relevant process. It can weaken the study when staff select only favourable cases, frame every problem as external, or control the evidence presented to the evaluator. The protocol should state who will contribute to theory development, who controls data access, and how the evaluation will obtain accounts that do not depend on programme management.
An explicit challenge process is useful. The evaluator can ask an informed colleague, steering group member, or external reviewer to examine the main contribution claim and identify the strongest alternative explanation. The challenge should identify the alternative, the part of the contribution story it addresses, and the evidence that could distinguish it from the preferred account. Its purpose is to improve the evidence plan and final judgement, not to create procedural disagreement.
Governance arrangements should protect the distinction between a finding and a management preference. Programme managers may disagree with an interpretation; they can provide further records or identify an error in the chronology. They should not be able to remove an evidence-based finding because it is inconvenient. Conversely, evaluators should not interpret limited access as proof of a particular alternative explanation. The report should state where the evidence supports a conclusion and where access or source dependence leaves the claim open.
Ton (2017) used contribution analysis to examine the FONDOECAS innovation fund for farmer organisations in Bolivia. The study examined three parts of the intervention logic: the relevance of farmer organisations for local economic development, the fund’s effectiveness in strengthening them, and the efficiency of its grant-allocation mechanism. It combined time-series data, in-depth interviews used for process tracing, and qualitative comparative analysis of conditions associated with organisational success. The study found that grants to organisations that already had substantial resources were particularly unsuccessful. It is a worked example of a contribution story tested through several methods, not a simple before-and-after account of the fund.
Choi et al. (2023) applied Mayne’s six-step contribution-analysis process to six health-professions and health-sciences programmes at a large Australian university. The evaluation combined documents with focus groups involving academic staff, industry representatives, and graduates. It developed theories of change for the programmes and examined the teaching and learning factors that supported or hindered the development of graduate capabilities. The design did not estimate the effect of a curriculum against an alternative curriculum. It used contribution analysis to assess how programme activities, context, and other influences could plausibly relate to the graduate outcomes under study (Choi et al., 2023).
Downes et al. (2018) report a modified contribution-analysis approach used by the US National Institute for Occupational Safety and Health to evaluate two research programmes. The evaluators organised evidence around occupational-safety and health outcomes, programme-specific review panels, and a theory-driven account of how research could reach workplace practice. The application is relevant because it treats research impact as a long causal chain in which regulations, industry action, and other external factors can also matter. The approach generated programme feedback; it did not claim to isolate the programmes’ effects on morbidity or mortality (Downes et al., 2018).
Delahais & Toulemonde (2012) draw on five contribution-analysis evaluations of European Commission policies covering development aid, agriculture, employment, and governance to set out lessons from applying the method’s six steps in practice. Rather than presenting a single case, the article documents how the evaluators drafted contribution stories, made contribution claims, and refined concepts such as contribution claim, causal mechanism, and strength of evidence across the five assignments.
The example is included as a practice synthesis rather than a worked case: it shows what contribution analysis looks like when applied repeatedly by the same evaluation team across genuinely different policy sectors, and where the method’s six-step process needed the most adaptation in application. Readers looking for a single detailed contribution story should turn to the Bolivian and Australian examples above; this source is better used to see what recurs across several independent applications.
Morton (2015) develop the Research Contribution Framework, adapting contribution analysis to assess how social science research is taken up and used in policy and practice. The framework was built and tested using a case-study approach exploring the nature of research impact, and it focuses on the roles of research users and on both process and outcome, rather than only on a final outcome indicator.
The example is a further contrast with the NIOSH case above: both concern research-impact assessment, but Morton’s framework was developed from UK case studies of social science research reaching policy and practice, while the NIOSH application evaluated occupational-health research programmes using programme-specific review panels. Comparing the two shows that “research impact” is not evaluated the same way across research fields, funders, or national contexts, even when both draw on the same underlying contribution-analysis logic.
9.11 Weighing evidence and revising the contribution story
Evidence should be assessed against the job it is being asked to do. Delivery data can establish whether resources were disbursed or activities took place. They are weaker evidence of how recipients used the resources or whether those activities changed practice. Outcome data can show that an indicator moved; they may not distinguish the programme from a concurrent policy or change in measurement. Interviews can provide information about decisions, informal practice, and perceived influence; they require attention to who was interviewed, when, and whether their account can be compared with records or other positions in the process.
The contribution story should identify these differences explicitly. A claim may rest on a combination of evidence: records showing that an intervention was delivered; observations that an intermediate practice changed; accounts from decision-makers about why it changed; and data consistent with the expected outcome. The combination can be persuasive where the sources are independent, the timing fits, and strong alternatives have been examined. It remains limited where every source derives from programme staff, where the sequence is reconstructed years later, or where an alternative process explains the same observations equally well.
The evaluator needs a record of how evidence changed the story. A simple evidence register can identify each item, the link or assumption it addresses, its provenance, the interpretation initially considered, and the eventual judgement. The register need not be published in full, particularly where it contains confidential information. But the report should retain enough detail for a reader to see which evidence was decisive, which was derivative, and what material contradicted the preferred account. Without such a record, contribution analysis can become a narrative assembled from selected supporting material.
Negative evidence is often informative. A programme may have delivered the planned support but not altered the practice that was expected to mediate an outcome. A site may show no change despite receiving the intervention, or an outcome may improve in locations that were not exposed. These observations do not automatically disprove every contribution claim. They require the theory to be reconsidered: perhaps an enabling condition was absent, the mechanism was not as assumed, the intervention reached only part of the target group, or another process produced the outcome. The revised story should state which of these explanations is supported and which remains uncertain.
9.12 Reporting a credible contribution claim
The final report should begin its findings with the programme and outcome to which the claim applies. It should describe the contribution story, the evidence for each major link, the alternatives investigated, and the resulting judgement. The conclusion needs to say what the programme is judged to have contributed to, in which sites or period, and with what qualification. A statement that a programme ``made a credible contribution’’ is incomplete unless the reader can see the outcome, mechanism, evidence, and unresolved alternatives behind it.
| Component | What the report should show |
|---|---|
| Contribution question | Intervention, outcome, setting, period, and intended use of the finding |
| Contribution story | Causal links, assumptions, contextual conditions, and principal alternatives |
| Evidence assessment | Sources for each link, timing, dependence, limitations, and contradictory material |
| Final judgement | Supported links, weak or untested links, alternatives that remain, and claim boundary |
The language of the judgement should be proportionate. The evidence may support a conclusion that the programme contributed to a change in an intermediate outcome; that it probably contributed to a final outcome alongside other influences; that it enabled an outcome but its independent role cannot be separated; or that the available evidence is insufficient to assess contribution. These are substantively different findings. A report should not use the same favourable label for each.
The judgement also needs to distinguish contribution from attribution. An attribution claim says that a specified outcome difference is due to the intervention rather than an alternative treatment condition. It normally requires a credible comparison, whether experimental, quasi-experimental, or within-unit. A contribution claim says that the intervention formed part of a causal process consistent with the evidence after relevant alternatives have been considered. It can be strong without assigning a share of the outcome to the intervention. It should not imply that share where the design has not estimated it.
Commissioners sometimes ask contribution analysis to settle questions that require a different design. If the decision depends on an estimate of an average effect, the study needs a counterfactual component. If the decision depends on whether a particular mechanism operated in a disputed case, a focused process trace may be required. If it depends on patterns across a set of cases, configurational analysis may be appropriate. Contribution analysis can organise these elements around the programme theory, but it does not replace their distinct evidential tasks.
9.13 Scope, comparison, and learning across sites
Programme-level contribution stories often span sites with different delivery conditions. The evaluator should not assume that the same theory operated in every location. A grant may be used differently by new and established organisations; a service intervention may depend on staffing that exists in some districts but not others; a policy may be interpreted differently across administrative levels. The initial theory can specify such conditions. The evidence plan can then examine whether they explain variation in implementation or outcomes.
Cross-site comparison can serve several purposes. It can identify where the programme was implemented as intended and where a link failed. It can test whether an expected condition was present in cases with the outcome and absent in cases without it. It can select cases for deeper process tracing. It cannot, simply by listing several cases, establish that the programme was effective across a population. The comparison needs a clear selection rule, comparable definitions, and a statement of what each case contributes to the broader claim.
Negative and deviant cases deserve the same attention as cases that fit the theory. A case in which the intervention was delivered but the expected outcome did not occur may reveal a missing condition or a weak mechanism. A case in which the outcome occurred without the intervention may point to a rival explanation or show that the programme was not necessary. These cases can refine the contribution story and identify its scope. Excluding them because they complicate an affirmative programme narrative is a serious limitation.
The final claim should therefore name its boundary. It may concern a particular phase of a programme, a defined group of sites, a mechanism under specified conditions, or a period before a policy or funding change. Generalisation beyond those conditions requires further comparison, replication, or an evidence base that extends beyond the evaluation. This is not a formulaic caveat. It determines whether a finding about a pilot, a district programme, or a group of organisations can support a decision about wider implementation.
9.14 Focused case studies within a contribution analysis
Case studies can investigate links that aggregate programme evidence leaves uncertain. Programme-level data may establish that activities were delivered across sites, while selected cases can examine use, subsequent decisions, and the conditions under which the anticipated mechanism held or failed. The cases need a stated selection logic, including contrasting outcome and implementation patterns where these are relevant. Such selection does not make the cases statistically representative. It makes their role in testing the contribution story clear.
The case-study protocol should be aligned with the wider theory. It needs evidence on the activities received, their use, the proposed change in practice or capacity, the relevant outcome, and the other conditions that could explain the result. A general descriptive case study of programme activities will not test the proposition. The relationship between the case evidence and the programme-level evidence should be stated from the outset: the case may investigate a mechanism, test an assumption, assess an alternative explanation, or explain an outlying pattern.
Process tracing is often appropriate for this role. It can assess the sequence within a particular site and the diagnostic value of the records and accounts available. Contribution analysis then uses the trace as evidence about a part of the wider contribution story. The two methods should not be treated as interchangeable. The process trace may support a causal account of one decision or implementation episode; the contribution analysis must still consider how that account relates to the programme’s outcomes, other sites, and alternatives operating at programme level.
Case comparison can also be used to refine the theory. Cases with different outcomes can indicate whether an assumed condition was present where a mechanism operated and absent where it did not. The comparison needs common definitions and an account of selection. It should avoid a familiar error: choosing only successful cases, finding the expected mechanism in them, and then presenting the mechanism as the explanation for the whole programme. Unsuccessful and deviant cases provide evidence about the boundaries of the contribution claim.
9.15 Combining contribution analysis with other evidence
Contribution analysis provides an organising framework when an evaluation uses several methods, but the methods should retain their own standards. A counterfactual component can estimate an outcome difference under a specified treatment contrast. Contribution analysis can examine whether the programme theory provides a credible account of how that difference arose, whether implementation reached the relevant population, and whether the effect is likely to depend on conditions that differ across sites. The counterfactual estimate should not be relabelled as a contribution story, and the contribution story should not be reported as an effect estimate.
Configurational analysis may be useful where the programme has several cases and the question concerns combinations of conditions associated with an outcome. Its results can identify configurations for further inquiry or conditions that the contribution story should address. It does not establish the process within any one case without additional evidence. Likewise, realist evaluation can develop and test context–mechanism–outcome propositions that enrich a programme theory, but it makes a different type of explanatory claim. The evaluation design should state why each method is included, what evidence it will use, and which uncertainty it is expected to reduce.
Mixed methods do not remove the need for judgement. A survey, a monitoring dataset, and a set of interviews may all be affected by the same implementation boundary or the same measurement change. Their convergence may therefore offer less independent support than it appears. The evaluator should ask whether sources were generated separately, whether they cover the same period and population, and whether each provides evidence that the others cannot. The strongest designs use methods to address different weaknesses in the contribution story, not simply to multiply favourable findings.
9.16 When contribution analysis is not enough
Contribution analysis is a poor choice when the causal claim is not defined, the programme cannot describe the intervention that was actually delivered, or no plausible evidence can be obtained about the central links. In these circumstances, an elaborate theory diagram will not produce a credible contribution claim. A narrower descriptive study may establish what was delivered and to whom. A formative inquiry may identify operational problems. Neither should be presented as evidence that the programme contributed to its intended outcome.
The method is also limited where a single alternative explanation is decisive and can only be addressed through a credible comparison. Contribution analysis can examine implementation and mechanisms, but it cannot on its own estimate an outcome difference between intervention conditions. Where a policy outcome is shaped by several reforms, the study may be able to make a carefully bounded contribution claim, but it should not imply that the programme’s share can be quantified without a design capable of doing so.
Time and access can set similar limits. A short endline study may find accounts of a programme’s influence but have no contemporaneous records, no access to participants who disengaged, and no way to investigate a concurrent intervention. The resulting evidence can inform a revised theory or a later evaluation. It is not a basis for a confident causal conclusion. The evaluation should say where the main uncertainty remains and what additional evidence or design would be required to reduce it.
Contribution analysis can still be useful when its conclusion is qualified. A finding that evidence supports delivery and some intermediate changes but does not resolve the final outcome link may guide monitoring and adaptation. A finding that a programme appears to matter only under stated conditions may prevent inappropriate scale-up. The discipline lies in preserving these limits in the final report rather than reducing them to a favourable overall narrative.
9.17 Timing, resourcing, and use
Contribution analysis is easier to conduct when it is planned before the programme reaches its final reporting stage. At inception, the programme can define the outcomes and intermediate changes it expects to observe, record the conditions that may affect delivery, and retain the records needed to investigate decisions and adaptations. The evaluation does not need to fix the contribution story permanently at that point. It does need a clear enough initial account to identify what evidence would be lost if it were not collected as the programme unfolds.
An early or mid-course contribution analysis can test whether the theory is being implemented as intended. It may show that a target group is not being reached, that an assumed organisational decision has not occurred, or that a necessary resource is unavailable. These findings can support programme adaptation. They do not establish final impact, particularly where outcomes are expected only after a long delay. The report should state whether it is assessing delivery, intermediate contribution, or contribution to a final outcome, and avoid treating evidence from one stage as evidence for another.
The budget needs to reflect the iterative nature of the work. The team requires time to review existing evidence, develop an evidence matrix, collect targeted material, examine rivals, and revise the story. It may need site visits, translation, access to administrative data, and follow-up after an initial finding exposes a gap. A design that allocates almost all fieldwork to a single endline visit will be able to collect accounts and documents; it may not be able to test a disputed contribution claim thoroughly. A smaller but focused contribution analysis is often preferable to a broad claim based on a superficial review of a long theory of change.
Use arrangements should be agreed before the evaluation starts. Programme staff may need timely findings on a weak implementation link, while a funder may need a later assessment of contribution to an outcome. These purposes can coexist if the reports distinguish provisional learning from final causal conclusions. They come into conflict when management expects evaluators to affirm the programme theory regardless of the evidence. The terms of reference should protect the reporting of adverse findings, evidence gaps, and alternatives that remain plausible.
The final product should be usable without concealing uncertainty. A concise contribution story, a link-by-link evidence table, and a clear account of what the programme should change can support decision-making. The technical record should retain the evidence register, theory revisions, source limitations, and rationale for the final judgement. This makes it possible for later evaluators or programme staff to revisit a claim as new data emerge, rather than treating the contribution analysis as a one-off verdict.
9.18 Contribution claims and their limits
Contribution analysis is suited to programme-level questions where the outcome is produced through a chain of activities and contextual conditions, and where a counterfactual comparison is unavailable or addresses only part of the question. It can organise evidence across several sites and explain why a programme appears to have operated differently across them. It cannot convert a plausible theory into proof that the intervention caused the outcome. Claims about an effect require an appropriate comparison; claims about a mechanism may require process tracing; claims about cross-case pathways may require configurational analysis.
Table 9.3 summarises when contribution analysis fits an evaluation question and when another method is required.
| Contribution question | Best-suited evidence base | Causal claim supported | When another method is needed |
|---|---|---|---|
| Is delivery reaching intended groups and are assumptions holding? | Monitoring data, implementation records, and interviews | Evidence about delivery, reach, and assumptions; not contribution to an intermediate outcome on its own | Evidence that the intermediate outcome changed, together with assessment of relevant alternatives, for a contribution claim |
| Under what conditions did the programme contribute across sites? | Cross-site comparison, outcome data, and targeted process tracing | Conditional contribution claim bounded by site conditions | Configurational analysis (Chapter 7) for cross-case pathways |
| Did a specific disputed link operate in one case? | A focused process trace nested in the contribution story | Within-case contribution to that link | Process tracing (Chapter 8) for the within-case detail |
| What share of the outcome is due to the intervention? | Not suitable on its own | None: contribution is not attribution | A counterfactual design (Chapter 6) to estimate an effect |
The next chapter examines realist evaluation, which also works with programme theory but frames explanation through context–mechanism–outcome configurations.