Building an Evidence-Based Policy Brief
Build a decision-focused policy brief by separating evidence from inference, comparing options and explaining sources, uncertainty and limitations.
An evidence-based policy brief answers a named decision question, keeps what was observed separate from what the author infers, compares options a decision-maker could realistically choose between, and records the limits of its sources and the uncertainty in its conclusions. Format serves those four commitments; it does not substitute for them. The method set out here is an editorial approach for researchers preparing a short, decision-relevant public research sample. It is not an official standard, and it does not replace statistical review, legal advice or ethics review where those are needed.
Begin with the decision, not the topic
A topic describes an area of concern. A decision question names a choice, the person or body facing it, the timeframe, and the point at which the brief stops being useful. "Skills shortages in social care" is a topic. "Whether a fictional regional authority should extend a pilot training subsidy to a second district in the next budget cycle, given the evidence available on take-up and completion" is a decision question. The second version tells the author what to include, what to leave out, and when to stop reading.
Turning a topic into a decision question
Four prompts usually convert a topic into a workable question:
- Who decides, and on what instrument? A funding renewal, a procurement specification, a regulatory threshold and a ministerial submission are different decisions with different evidence needs.
- What is the choice set? Renew, extend, reduce, consolidate, or defer pending further evidence. If the options are not named, the brief cannot compare them.
- Over what period and for which population? A conclusion about a twelve-month pilot in three sites does not transfer automatically to a national rollout or to a different age group.
- What would change the answer? Stating this in advance prevents the brief from becoming a defence of a preferred option.
Write a scoping statement before drafting
A short scope paragraph near the top saves the reader from guessing. It should specify the jurisdiction, the period covered, the population or organisations in view, the sources consulted, and the questions the brief does not address. The last item matters most: a brief that appears to cover a national question while relying on three local interviews misleads even when every sentence is accurate.
Keep observation, inference and recommendation apart
Most weak briefs fail at the boundary between what a source says and what the author concludes. Three tiers, signalled in the language of the text, make the boundary visible.
Observation. What a source records, expressed in terms that match the source's scope: "the evaluation report recorded higher completion rates among participants who received the additional advisory session"; "two of the four interviewees described difficulty recruiting staff with the required certification."
Inference. What the author concludes from observations, with the reasoning made explicit: "this suggests the advisory session may have contributed to completion, although participants were not randomly assigned and motivation may differ between groups."
Recommendation. What should be done, by whom, subject to which conditions: "on current evidence, a conditional extension for one year is defensible if intake records are standardised so that the next review can compare sites."
The tiers are not a hierarchy of importance. A brief with strong observations and no recommendation can be entirely appropriate. The failure mode is inference written in the grammar of observation — "the pilot improved completion" rather than "completion rose during the pilot period, and the evaluation could not isolate the pilot's contribution."
A claim–evidence–limitations table
The most useful working tool is a table that forces each material claim to carry its own evidence and its own caveats. Build it while drafting, then decide which rows belong in the body and which belong in an annex.
| Claim, stated precisely | Type | Evidence and source | Source quality and date | Limitations and uncertainty | What would change it |
|---|---|---|---|---|---|
| e.g. "Applications processed at co-located sites required fewer return visits" | Observation | Case-record extract from the pilot's own management system | Administrative data, single system, twelve-month window; not independently audited | Two of three sites kept incomplete records; definition of a "return visit" may differ by site | A standardised record audit across all sites |
| e.g. "Co-location is likely to reduce repeat contact for applicants" | Inference | Comparison of case records with staff interviews | Mixed: administrative data plus four semi-structured interviews | No comparison group; staff accounts may reflect expectations as much as experience | A matched comparison with a non-co-located site |
| e.g. "Extend for one year with standardised intake" | Recommendation | The rows above, plus the decision-maker's stated constraints | Judgement, not evidence | Depends on cost data not available within the drafting window | Cost per case exceeding the current centralised service |
Three rules make the table useful rather than decorative. Write the claim as a testable sentence, not a theme. Name the specific document, dataset or interview round rather than "stakeholders" or "research". Record the date of the evidence, because a source that was current two years ago may no longer describe the operating environment.
Recording source quality and uncertainty
Source quality
Provenance matters more than the prestige of the publisher. For each source, ask who produced it, under what mandate, using which method, on what population, over what period, and whether the version you hold is complete or an extract. A national statistical release, an internal monitoring return, a stakeholder interview and a news report answer different questions and carry different weight. None is automatically superior; a single well-documented interview can be the best available evidence on how a process operates in practice.
Published professional frameworks support the general expectation of transparency. The UK Policy Profession Standards Annex, published for policy work in the UK civil service, sets expectations about the use of evidence and analysis within that jurisdiction; it is a reference framework for professional practice, not a briefing template, and its existence does not indicate that any particular brief complies with it. The European Code of Conduct for Research Integrity, published by ALLEA, addresses honesty in reporting and transparency about methods across European research institutions; it speaks to research conduct rather than to the structure of a policy document. Both are useful for justifying why a brief should expose its method. Neither endorses the method described here.
Types of uncertainty
Recording uncertainty is not a single act of hedging. Distinguish at least five kinds, because each calls for a different response from the reader:
- Measurement uncertainty: sampling error, response rates, instrument design. Where it has been quantified, report the range rather than the point estimate alone.
- Coverage gaps: missing regions, periods, sites or groups. State which parts of the decision question the evidence cannot reach.
- Contested interpretation: sources that describe the same events differently. Present the disagreement rather than averaging it away.
- Assumption dependence: conclusions that hold only if take-up, cost or behaviour follows a particular path. Name the assumption and test the conclusion against a plausible alternative.
- Genuine unknowns: questions no available source addresses. Saying so is more useful than filling the gap with inference presented as fact.
For each, note whether the uncertainty is directional: whether resolving it could reverse the recommendation rather than merely adjust a margin.
Comparing options without hiding value judgements
An options comparison fails in two opposite ways: it presents a single recommendation with no alternatives, or it aggregates everything into a composite score that conceals the weights used. A defensible comparison names the options, names the criteria, shows the evidence against each criterion, and states the judgements that remain.
Structure the comparison
Include the status quo as an option; it is usually the actual baseline against which a decision is taken. Keep the option set small enough to compare honestly — three or four distinct choices is usually more useful than eight variations. Then assess each against criteria the decision-maker recognises, with a column that records who bears the costs and who receives the benefits.
| Criterion | Option A: renew as is | Option B: extend to two further sites | Option C: consolidate into a central service |
|---|---|---|---|
| Reach to target population | Known from pilot records; limited to three sites | Uncertain; depends on site selection | Wider nominal coverage, lower accessibility for some groups |
| Evidence strength | Moderate: consistent records, no comparison group | Weak: extrapolation from three sites | Weak: no local data |
| Distributional effect | Concentrated in pilot areas | Depends on which sites are chosen — a value judgement | Likely to disadvantage those with least travel capacity |
| Reversibility | High | Moderate; expansion creates expectations | Low in the short term |
| Cost per case | Not established within drafting window | Not established | Not established |
| Judgement required | Whether pilot-period conditions persist | Whether equity or coverage should carry more weight | Whether accessibility outweighs administrative simplicity |
Make judgements visible
Choosing the criteria is a value judgement. So is weighting them, setting the time horizon, deciding whose costs count and choosing how much risk to accept. These should appear as stated positions, not as technical inevitables. A composite score is acceptable only if the weights are shown and the ranking is tested against a plausible alternative weighting; otherwise, present the criteria separately and let the decision-maker see the trade-off.
Fictional outline: question, findings, options, limitations
The following outline is fictional. It is included only to show the structure; it does not describe a real programme, jurisdiction, organisation or finding, and its statements should not be read as evidence about any real service.
Decision question. Should a fictional regional authority renew a two-year pilot that places advisory staff in three public libraries, extend it to further sites, or return the service to a single central office?
Findings. Case records show that applicants seen at library sites made fewer repeat visits than applicants seen centrally. Two of the three sites kept incomplete records for part of the period. Staff interviews described faster resolution of straightforward queries and more difficulty with complex cases. No comparison group was established, and cost data were not available within the drafting window.
Options. Renew as is; extend to two further sites; consolidate centrally. The comparison uses reach, evidence strength, distributional effect, reversibility and cost, with cost marked as unresolved rather than estimated.
Limitations. The absence of a counterfactual means the reduction in repeat visits cannot be attributed to co-location. Inconsistent records limit site-level comparison. Interview accounts are not representative of applicants. Cost per case is unknown, so any recommendation is conditional on that figure remaining within the authority's existing service budget.
Conditional recommendation. Renew for one year, with a requirement to standardise intake records and to report cost per case, so that a later review can compare sites on a consistent basis.
What to leave out, and how long to make it
Length follows the decision, not a house convention. A single-page note may be sufficient for a narrow question with strong evidence. Four pages with an annex may be needed where several options and a contested evidence base must be presented. Whatever the length, three omissions improve most drafts: background that does not bear on the choice, secondary reporting where the primary document is available, and recommendations that outrun the evidence recorded in the table.
Questions that arise in practice
Does a brief have to end with a recommendation? No. Where the evidence does not support a single choice, a brief can set out the conditions under which each option would be preferable. If the commissioner has asked for a recommendation, give one, state how confident it is, and identify the evidence that would change it.
How should interview material be used? Attribute by role rather than by name unless the participant has agreed otherwise, separate what was said from what it means, and note the limits of your notes or transcripts. Confidentiality obligations may restrict what can be published at all; the note on confidentiality for expert consulting sets out questions worth settling before material is reused.
What if two credible sources conflict? Report the disagreement, give each source's provenance, and explain what additional evidence would resolve it. A brief that silently prefers one account is harder to challenge and less useful to the decision-maker.
A short pre-submission check
- Does the first paragraph state the decision question and the scope?
- Is every material claim in the table assigned to observation, inference or recommendation?
- Are source dates and coverage gaps recorded next to the claims they qualify?
- Are the options genuinely feasible, and is the status quo among them?
- Are the criteria, weights and remaining judgements stated rather than implied?
- Does the brief say what it does not cover?
Researchers assessing whether their own expertise suits project-based policy, regional or social research can review the public policy expert area for the kind of work involved, then test a recent piece of their own against the check above. If the table cannot be completed from the evidence at hand, that is usually the most useful finding the draft can report.