Turn the links in your theory of change into monitoring questions, evidence and decisions about what to improve.
A theory of change helps monitoring and evaluation focus on whether the expected pathway to change is happening, why it is happening and where it is breaking down. Monitoring follows selected measures during implementation. Evaluation examines questions about the work’s implementation, results and value, using an appropriate design.
The model supplies questions to investigate, not a conclusion to confirm. If you need to build the model first, start with the theory of change introduction.
| Expected connection | Monitoring question | Evaluation question |
|---|---|---|
| Training → demonstrated skill | Are people attending and passing relevant assessments? | Which parts of the training support skill development, and for whom? |
| Skill → use at work | Are people getting opportunities to apply the skill? | What enables or prevents application? |
| Application → improved practice | Is the intended practice changing? | How much of the change can reasonably be linked to the intervention? |
A count of sessions addresses delivery, not the whole pathway. Include evidence about the connections between delivery and change. A small number of well-chosen questions can be more useful than monitoring every box equally.
| Element | What to record |
|---|---|
| Outcome | The change, affected group and expected timing. |
| Indicator | Definition, unit, calculation and relevant disaggregation. |
| Starting point | Baseline source, period and limitations. |
| Evidence | Survey, assessment, observation, interview or authorized record. |
| Schedule | When evidence is collected and when it is reviewed. |
| Responsibility | Collector, reviewer and decision owner. |
| Decision | What finding would trigger a change in the work. |
Keep the plan proportionate. A new monthly survey may add little if reliable routine records already answer the question. Conversely, routine activity data will not answer how people experienced the service unless that experience is actually collected.
An illustrative team trains 80 supervisors. Sixty-four complete the course; 50 respond at follow-up; 30 of those report using a new coaching practice. Completion is 64/80, or 80%. Reported application among follow-up respondents is 30/50, or 60%. Neither number tells you the application rate for all 80 supervisors.
Interviews suggest that some supervisors have no protected time for coaching. The model assumed that time would be available. The response may be to work with managers on scheduling, not simply deliver more training. Check whether nonrespondents face similar barriers before treating the interview findings as representative.
Write assumptions as observable conditions: staff have access to equipment, participants can reach the service, or partners use the agreed referral route. Assign a source and review point for the assumptions most likely to affect the result.
Look for negative and unexpected effects. A faster process may create extra work elsewhere or exclude people who cannot use a digital form. Evidence that challenges the theory is useful; it tells the team where the explanation or implementation needs attention.
A before-and-after improvement is compatible with several explanations. Other services, economic conditions, changes in who responded or measurement differences may contribute. Decide what claim is needed and choose evaluation methods that can support it.
Qualitative accounts can clarify mechanisms and alternative explanations. Comparison designs may help estimate what would have happened otherwise. Neither a diagram nor an AI summary substitutes for an appropriate evaluation design. BetterEvaluation explains how to make the theory explicit.
Use consistent definitions and preserve the date, source and version of each measure. Where the question concerns individual change, link observations appropriately to the same person. Where it concerns a population trend, retain the sampling context instead.
Documents and interviews often explain why a number moved. Store the relevant evidence with its source rather than treating an uploaded file as an answer in itself. AI-assisted classification can support review, but people remain responsible for checking interpretation and deciding what changes.
Do not quietly redraw the pathway to make every result appear successful. Keep a record of what the team originally expected and what it learned. The CDC evaluation framework provides a broader structure for planning and using evaluation.
No. Prioritize the results and assumptions needed for the decisions and evaluation questions. Explain any important gaps.
Match review timing to when change could occur and when decisions can be made. A quarterly review may suit one program; another needs a different rhythm.
Yes. Document the evidence and reasoning, retain the prior version and distinguish a learning-driven revision from a changed target.
Usually monitoring alone does not establish causal impact. It describes progress and helps identify questions for evaluation.
