Designing an Expert Interview Guide for Policy Research

Start from the research need

Before drafting a single question, write one sentence naming what the research must inform: an appraisal, a delivery review, a design choice, a consultation response. Then list the propositions the study needs to describe or test, and mark which of them depend on practitioner knowledge rather than published documents.

This step prevents a common misallocation. Use published documents to establish what the written process says, then use the interview to examine the expert's direct experience and interpretation: how the process works in practice, where discretion is exercised, which steps absorb time, and what remains unresolved. The question map below keeps those distinctions explicit.

The question map

Each row carries one research need, one opening prompt, the probes that pursue it, and the record that would let a reader check the resulting claim. A guide whose rows cannot be completed in that form is usually still a topic list rather than a research instrument.

Write neutral opening prompts

Neutrality is a design property, not a matter of tone. A prompt is neutral when it does not embed the answer, a causal claim, or an evaluation. Consider a fictional study of implementation barriers in a licensing process. No real law, country or agency is described.

A leading version might read: "Why does the licensing process impose unnecessary delays on small applicants?" That question presupposes delay, presupposes that it is unnecessary, and invites the expert to defend or dispute the premise rather than describe the process.

A neutral revision asks for the sequence first: "Walk me through what happens between an application being submitted and a decision being issued. At each stage, tell me who acts and what they check." A probe can then follow: "Where in that sequence do applications most often stop, and what is happening at that point?" The probe still invites a frequency claim, so the guide should add: "How do you know that — from case records, from your own caseload, or from colleagues’ accounts?"

Four habits support neutrality:

• Ask for sequence, actors and documents before asking for explanation.
• Use the expert's own terminology after the first mention, rather than substituting the study's vocabulary.
• Avoid adjectives that carry judgement, such as slow, burdensome, effective or fragmented, unless the expert introduces them.
• Keep "what happens" separate from "why it happens", and label the second as interpretation when it arrives.

Separate experience, interpretation and speculation

Expert interviews mix three kinds of material, and a guide should make them distinguishable in the notes. Experience covers what the expert did, saw or handled. Interpretation covers their explanation of why something happened. Speculation covers what they expect, what others might do, or what would follow from a change.

Useful prompts for each are different. For experience: "What did you observe directly in that case?" For interpretation: "What is your reading of why that happened?" For speculation: "What would you expect to change if that rule were altered, and what would you need to see to be confident?"

Tag each answer accordingly during the interview or immediately afterwards. The tag does not settle whether a claim is true; it records the basis on which the expert offered it. That record matters when findings are drafted, because an interpretation presented in a findings section as an observed fact is a defect that no amount of later editing fully repairs. The UK government's Quality in qualitative evaluation guidance asks whether design and sampling are justified, coverage limits are explained, and analysis is transparent. This question map is a suggested format for applying those concerns; the framework does not prescribe its exact columns.

What belongs in a follow-up

Follow-ups should be planned, not improvised. Four types earn a place in the guide. A definition follow-up asks what a term means in the expert's setting: "When you say pre-application advice, which part of the team provides it?" An instance follow-up asks for a case: "Can you describe one case that illustrates that?" A boundary follow-up asks where a practice stops ("When does that practice not apply?"), and a record follow-up asks whether a shareable written record exists ("Is that written down anywhere that can be shared?").

A follow-up should not ask the expert to confirm the researcher's hypothesis, and it should not combine two questions. "Was the delay caused by staffing, and did that affect small applicants?" produces an answer that cannot be attributed to either part. Split it, and ask the second only if the first produces relevant material.

Build the corroborating record into the guide

Each row of the question map should name what would corroborate the answer: a public procedural document, published processing data, an internal record the expert can share, or a second account from a different part of the process. Naming this in advance changes how the interview runs, because the researcher can ask for the record while the expert is still present.

The record column also exposes gaps. Where no record exists, the finding should be reported as a single account rather than as an established fact. This is a limitation to state plainly in the write-up, not a defect to conceal. Guidance for policy makers on engaging with academics is relevant here, particularly on framing questions and agreeing what evidence will be usable at the end. For appraisal and evaluation questions, the Aqua Book sets out expectations for the analysis that policy decisions draw on.

Order, length and piloting

There is no universal interview length, and no correct number of questions. A practical constraint is more useful: the guide should be shorter than the time available, so there is room to clarify an answer and examine evidence that the opening questions did not anticipate. A guide that fills the whole session leaves no room for the follow-up that the best answer requires.

Order questions from concrete to evaluative. Sequence and role questions are easy to answer and establish shared vocabulary. Judgement questions come later, once the expert has described the process in their own terms. Closing questions should invite material the guide did not anticipate: "What should I have asked about that I have not?"

Test the guide with a colleague or a suitable pilot participant (subject to the project's ethical arrangements). Time it, remove questions that duplicate each other, and rewrite any prompt that a pilot respondent answers with a question of their own.

Boundaries worth stating in the guide

A guide should not solicit confidential client material, unpublished personal data, or legally privileged material. Where an expert offers such material, the researcher should be able to decline it without disrupting the interview. Consent, attribution and any anonymity arrangement should be agreed before the interview begins, and readers should verify legal, ethical and institutional obligations in their own jurisdiction rather than relying on general practice.

A guide also cannot substitute for judgement in the room. It fixes the questions that must be asked in comparable form; it does not decide when to depart from them.

A fictional worked example

The following excerpt is fictional and describes no real law, country or agency. It illustrates the four-part map in use for a study of implementation barriers in a licensing process.

A leading version of the opening, "Why is the licensing process so slow?", was replaced because it embeds an evaluation and invites a defence of the process rather than a description of it. The revised opening produced the sequence first, and the probes then located the wait within it. The fictional study would report the resulting account as one expert's observation, corroborated only if a record or a second account supported it.