How to write evaluation questions that get answered
Oct 12 2026 · Method
Good evaluation questions are answerable, bounded and tied to a decision. Here is the test to apply before you sign off your terms of reference.
A good evaluation question is answerable with evidence you can realistically collect, bounded to a clear scope and time frame, and tied to a decision somebody will actually make. Write them in plain language about your own intervention, using the OECD-DAC criteria — relevance, effectiveness, efficiency, impact, sustainability — as a prompt list rather than a questionnaire. Avoid double-barrelled questions such as 'was it effective and efficient?'. Test each one by asking what evidence would answer it, whether you can collect that evidence, and who will act on the result. If there is no evidence and no decision, cut the question.
What makes an evaluation question a good one?
An evaluation question earns its place when three things are true: you can answer it with evidence, its scope is clear, and somebody will use the answer. Questions that fail one of those tests tend to sit unanswered in the final report.
This matters most when you are writing a terms of reference or an evaluation plan, because questions written early shape what you collect later. The structure at /store/evalplan applies the same test from the start.
What are the three tests an evaluation question must pass?
- Answerable — the evidence exists, or can be collected within your budget and timeline.
- Bounded — it names the population, the period and the sites in scope.
- Consequential — a named person or group will act on the answer.
If a question fails the third test, it is usually curiosity rather than evaluation. That is not a reason to ban curiosity, but it is a reason not to put it in the terms of reference.
How do you turn the OECD-DAC criteria into evaluation questions?
Treat the criteria as a prompt list, not a questionnaire. Use them to check you have not missed a dimension, then write questions in plain language about your own intervention.
| OECD-DAC criterion | What it prompts you to ask | Example question |
|---|---|---|
| Relevance | Did the intervention address the right problem for the right people? | To what extent did the activities match the priorities of the groups involved? |
| Effectiveness | Did it achieve what it set out to achieve? | Which intended outcomes were met, for whom, and which were not? |
| Efficiency | Was it delivered in reasonable time and at reasonable cost? | How did delivery costs and timelines compare with the original plan? |
| Impact | What changed beyond the immediate outputs? | What changes occurred beyond the intended outcomes, positive or negative? |
| Sustainability | Will the benefits last? | What capacity and funding remain once external support ends? |
| Coherence | How does this fit with other interventions and policies? | How did this work alongside other actors delivering in the same area? |
You do not need a question for every criterion. On a small project, three criteria may be enough; say explicitly which ones you are not assessing and why.
What is the difference between a weak and a strong evaluation question?
Weak questions are usually too broad, too compound or too vague to answer. The fix is almost always to narrow the scope and name the evidence.
| Weak question | Why it fails | Stronger version |
|---|---|---|
| Was the project effective and efficient? | Double-barrelled. You cannot answer two things in one question. | Was the project effective? Then, separately: was it delivered efficiently? |
| Did the training work? | 'Work' is undefined, and the group and period are unspecified. | To what extent did participants apply the training in their roles in the following year? |
| What was the impact? | Unbounded. Impact on whom, over what period, measured how? | What changes in participants' incomes were reported, and what else could explain them? |
| How satisfied were stakeholders? | Satisfaction is not the same as results, and it is often hard to act on. | Which parts of the delivery did partners find most and least workable? |
| Did we reach the target? | This is monitoring, not evaluation. The indicator already answers it. | Why did reach differ between sites, and what does that imply for the next phase? |
Why do double-barrelled questions produce useless answers?
'Was it effective and efficient?' invites an answer to whichever half the respondent finds easier, and you will not know which one you got. Split it. If you genuinely need both, treat them as two questions with two evidence plans.
The same applies to questions joined by 'or'. A question with two possible subjects usually produces a report section that answers neither.
How many evaluation questions should you have?
Fewer than most drafts contain. For a single project evaluation, three to five primary questions is usually workable, each with one or two sub-questions. Beyond that, the evidence collection spreads thin and the report becomes descriptive.
If you have twelve questions, you probably have four questions and eight things you already know. Move the known items into background.
How do you check that the evidence exists before you commit?
Run this before the questions are fixed, not after fieldwork has started.
- Name the decision the answer will inform, and the person who makes it.
- Name the evidence that would answer the question, and confirm it exists or can be collected.
- Check the scope is explicit: which population, which period, which sites.
- Read the question aloud and split it if it contains 'and'.
- Check it is not already answered by routine monitoring data.
- Cross-check against your indicator set. If no indicator speaks to it, add one or drop the question.
- Draft the one-sentence answer you expect. If you cannot imagine one, the question is too vague.
If a question cannot change a decision and cannot be evidenced, it does not belong in the evaluation. Cut it early, when cutting is still cheap.
How should evaluation questions link to your logframe and indicators?
Questions should sit above the logframe, not replace it. The logframe tells you what you intended and what you will track; the questions tell you what you want to understand and judge.
A practical approach: map each question to the indicators that will feed it, and note any gap. Gaps are where you need a survey, interviews or a review of records. A starting structure is at /blog/logframe-template-google-sheets, and indicator definitions are covered at /store/indicator.
For data collection, mobile forms such as those set up with /store/kobo keep field data consistent with your indicator definitions. A wider set of practical tools sits at /tools.
When should you cut a question?
Cut it when the evidence is out of reach, when the answer will not change anything, or when it duplicates something already covered. Cutting at the design stage costs nothing; cutting at reporting stage wastes fieldwork.
If you are unsure how to judge reach, /which-ai can help you think through what is realistic for your budget and timeline. For a fuller walkthrough of designing an evaluation from scratch, see /book.
If you need to build the plan around the questions, the evaluation pack at /store/evalplan and the evaluation plan builder at /store/evalplan are the shortest route. Templates and ongoing support are available through /membership.
Written for M&E professionals. Not a guarantee of accuracy — always verify every claim against its source, and let a named human sign off. Get the free handbook.