Building a Theory of Change for an Evaluation
Set out how a programme is expected to produce change, identifying assumptions, rival explanations and evidence that could test the pathway.
A theory of change for an evaluation is a causal argument, not a diagram. It states how an activity is expected to produce immediate changes, how those changes are expected to contribute to a longer outcome, which conditions must hold for each link, and what evidence would count for or against the pathway. The evaluation plan then uses that argument to decide what to measure, which rival explanations to investigate, and what findings would justify revising the programme or the evaluation question.
Begin with the claim the evaluation must test
An evaluation plan is organised around a decision: whether to continue, adjust, scale back or stop an intervention, or whether a claim about its effect is credible. The theory of change should be scoped to that decision. A whole-programme theory can be useful for strategy, but an evaluation usually needs a narrower pathway that can be examined within the available access, time and ethical constraints. A theory that is too broad becomes a description of everything and a test of nothing; one that is too narrow may omit the conditions that determine whether the central claim holds.
The Magenta Book, UK central government guidance on evaluation, distinguishes process, impact and value-for-money evaluation and describes a theory of change as setting out expected causal pathways and assumptions (GOV.UK). That distinction matters because implementation questions, attributable change and questions about benefits and costs require different evidence from the same pathway.
A logic model and a theory of change are sometimes used interchangeably. For evaluation planning, the useful distinction is that a logic model arranges the components of an intervention, while a theory of change explains why the links between them are expected to operate and what would weaken or break them. The explanation is the part that can be tested.
What connects activity and outcome
The connection is a mechanism: a change in knowledge, capability, motivation, behaviour, relationship or institutional response that the activity is meant to produce. Training does not produce employment. It may improve interview fluency, which may change how a candidate performs in interviews, which may increase offers; the candidate must then be able to accept and sustain work. Each step depends on conditions outside the training room.
Assumptions are the propositions about people, organisations, context or evidence on which the pathway depends. A risk is an event that may occur; an assumption is a condition that must hold for the pathway to proceed. The two overlap in practice, but separating them helps the evaluation plan. If an assumption fails, the pathway may break. If a risk materialises, it may disturb delivery. The evaluation should test the assumptions that are both consequential and uncertain, not every condition that can be imagined.
Conditions can be necessary or supporting. A necessary condition is one without which the pathway cannot operate, such as employers with vacancies if the outcome is paid employment. A supporting condition makes the pathway more likely but is not strictly required. The worksheet asks the analyst to mark which is which, because interpretation depends on it: if a necessary condition is absent, a null result says little about the mechanism.
A worksheet for the pathway
The following format is a suggested working aid, not a validated instrument or an institutional requirement. Use one row per distinct pathway, not one row per activity. An activity that triggers several mechanisms needs several rows; otherwise the evaluation will average different causal stories into one.
| Element | Question to answer |
|---|---|
| Activity | What is delivered, to whom, by whom, and with what intensity? |
| Mechanism | What change in knowledge, capability, motivation, behaviour, relationship or institutional response is expected, and why would this activity produce it? |
| Immediate change | What short-term change should be observable in participants or organisations? |
| Longer outcome | What change in status, behaviour or condition matters to the decision? |
| Assumption | What must be true for the link to hold? Mark necessary or supporting. |
| Rival explanation | What else could produce the same immediate change or longer outcome? |
| Observable evidence | What record, measure or account would show the pathway operating, failing or being absent? Name the source and timing. |
Write the longer outcome first and work backwards. Then write the rival explanation before the evidence, because the evidence should discriminate between the pathway and its rivals. In the evidence column, name the data source, who holds it, whether it can be linked, and what pattern would be consistent with the pathway. Participant reports are weaker than a combination of programme records, employer confirmation and administrative outcome data, but access and ethics decide what is possible.
Mark the assumptions that are both consequential and uncertain. Those are the candidates for evaluation questions. An assumption that is certain or trivial will not improve the plan.
A fictional worked example
The following example is fictional. It shows the format and the kind of reasoning required; it does not report a real programme or a measured effect.
| Element | Example |
|---|---|
| Activity | A six-session interview skills workshop for unemployed residents, delivered by a fictional employment-support programme. |
| Mechanism | Participants rehearse answers, receive feedback and revise examples; this is expected to improve interview fluency and confidence. |
| Immediate change | Participants attend interviews and can give structured examples of their skills. |
| Longer outcome | Sustained employment, defined in this example as at least 16 hours per week for six months. |
| Assumption | Necessary: local employers have vacancies that match participants' skills, participants can travel to interviews, and participants can accept work without losing essential support. Supporting: workshop content matches local recruitment. |
| Rival explanation | Volunteers are more job-ready or more motivated; a local employer opened new vacancies; participants already had interview invitations from earlier applications; transport support improved both attendance and outcomes. |
| Observable evidence | Attendance records; mock-interview ratings; interview invitations recorded by participants; employer confirmation; administrative employment records at three and six months; interviews with participants who did not complete. |
Course completion is not proof of employment impact. Completion is an output or participation measure. It may show that the activity was delivered and that participants stayed, but it does not show that interview behaviour changed, that employers responded, or that employment was sustained. If completion rises while interview invitations and employment do not, the pathway may be incomplete at the employer-response link. Alternatively, the outcome may be measured too early, the comparison may be confounded, or the programme may be reaching people who face other barriers.
Reading the row as a chain also exposes a common error: attributing a longer outcome to the activity when an intermediate link has not been observed. The worksheet does not require every link to be measured, but it should record which links are assumed and which are evidenced. A rival pathway can be written as its own row with the same longer outcome and a different mechanism, which makes the evidence needed to distinguish the two explicit.
What an incomplete pathway might show
An incomplete pathway is a pattern in which some links hold and others do not. Activity may be delivered without the expected immediate change: the mechanism may be wrong, delivery may be too weak, or the change may take longer than the observation window. Immediate change may occur without the longer outcome: the link to employers, institutions or context may be broken. The longer outcome may change without exposure to the activity: selection, external shocks or concurrent policies may explain it. Effects may appear in one group and not another: the pathway may depend on a condition that the average hides. Evidence may exist only for participants who completed: attrition can make the pathway look stronger than it is.
Each pattern is a prompt to locate the break, not a verdict on the programme. Process evaluation can establish whether the activity was delivered and reached the intended group; impact evaluation can examine whether change is attributable; value-for-money evaluation can compare benefits and costs. A theory of change organises those inquiries but does not replace their designs.
Turning the pathway into an evaluation plan
Convert the rows into evaluation questions. For each pathway, ask whether the activity reached the intended group, whether the mechanism operated, whether the immediate change occurred, whether the longer outcome changed, what else changed, and what the intervention cost. Assign methods to questions rather than to the theory as a whole. Programme records may answer reach; interviews may illuminate mechanism; administrative data may measure outcomes; comparison groups may address rival explanations where feasible and ethical. Where a method cannot answer a question, say so in the plan and explain how the limitation will affect interpretation.
Specify in advance what pattern of evidence would lead to continuing, refining or stopping. A theory of change makes those decision rules discussable before findings are known. Date the theory and record why a link was added, removed or reclassified. An evaluation that treats the theory as fixed will miss what the evidence is telling it.
Researchers assessing whether a project fits their expertise can use the worksheet to test the brief's causal claims before accepting it; the published guide on before you accept an expert project sets out related questions about scope, access and expectations. Organisations commissioning an evaluation can use the same worksheet to check whether the brief is evaluable and whether the necessary data can be obtained (Policy Expert Network for organisations).
Limitations
A theory of change is a set of claims, not proof of causality. A diagram can clarify a pathway, but it cannot establish that the pathway operated or that the outcome would not have occurred otherwise. The theory may be incomplete, contested or shaped by the interests of those who commission it. Participants, delivery staff and funders may hold different theories; the evaluation should record those differences rather than resolve them by assertion.
Over-specification creates false precision, while under-specification hides the assumptions that matter. The right level depends on the decision, the existing evidence and the feasibility of data collection. Where access is limited, the theory can still guide interpretation, but it cannot supply a missing counterfactual. Ethical constraints on randomisation, data linkage or the use of personal information limit what can be inferred; they do not make an untested causal claim stronger.
Begin with one central pathway and one rival pathway. Complete the worksheet for each, mark the assumptions the evaluation can observe, and identify the evidence that would change the decision. If the central assumptions cannot be observed within access and ethical constraints, revise the evaluation question before the plan is fixed.