Qualitative data collection: interviews, focus groups, observation

Three methods, three kinds of access

Qualitative methods are routinely commissioned as an undifferentiated block — “and some KIIs and FGDs” — as if the methods were interchangeable. They are not. Each one accesses something the others cannot [1]:

• In-depth and key informant interviews access individual experience, expertise and accounts of processes: how the referral actually works, what happened when the committee collapsed, why an official believes the policy stalled.
• Focus group discussions access shared norms and the reasoning around them: what a community considers acceptable, contested or shameful, and how views shift when challenged by peers. The interaction is the data — a focus group is not eight simultaneous interviews.
• Observation — from structured checklists to open field notes — accesses behaviour and setting directly: what the clinic queue actually looks like, whether the protocol on the wall is the protocol in use. It is the only method that does not depend on what people are willing or able to say [1].

Method selection should therefore start from the evaluation question, not the budget line. Questions about mechanisms and sequences point to interviews (and, where causal claims will be tested against the evidence, to process tracing); questions about norms and acceptability point to groups; questions about practice-versus-policy point to observation. Most designs need two of the three, deliberately paired for triangulation [3].

Key informant interviews

Selection is an argument, not a convenience. Key informants are chosen because their position gives them knowledge others lack — the district officer who approves the permits, the community health volunteer who sees every defaulter, the trader who sets the local price. The selection logic (which positions, why, and how many of each) belongs in the methods section; a list of “people the programme suggested” is a selection bias with a bibliography. Deliberately include informants positioned to be critical — the operator of the rival scheme, the official the programme bypassed [1][3].

The semi-structured guide lists topics and core questions with freedom to reorder and probe. Good guides are short — a page of themes, not forty scripted questions — and open with easy, concrete ground before moving to judgements. Probing technique is where interviews are won: silence held a beat longer, “tell me about the last time that happened”, “you said it usually works — when didn’t it?”. Leading probes (“so the training helped?”) manufacture agreement and belong nowhere [1].

Capture: record where consented, and always take structured notes (attribution, date, place, role — pseudonymised as the protocol requires). An unrecorded, unnoted interview is an anecdote.

Focus group discussions

Composition first. Groups work when participants are peers on the dimensions the topic makes sensitive: separating women and men for gendered topics, residents from officials, users from providers. A senior voice in the room converts a discussion into a hearing [1][2]. Keep groups small enough that every participant speaks — common practice runs from about six to ten — and run several groups per population segment rather than one large one, because the group, not the individual, is the unit of analysis [1].

Facilitation is a two-person job: a moderator who manages the guide, the dominant talker and the silent corner; and a note-taker who logs speaker turns, non-verbal reactions and the seating map, and manages the recorder. Consent in groups has an extra clause: the facilitator can promise their own confidentiality but not that of other participants, and participants must hear that plainly before anything sensitive is discussed [1] — the wider consent duties are covered under ethics and consent. Groups are the wrong instrument for individually sensitive disclosure (violence, stigma, illegality); route those topics to private interviews with referral support [2].

Observation

Observation ranges from structured checklists — is the cold chain equipment functioning, are gloves worn, how long is the wait — which yield countable, comparable data across sites, to open ethnographic notes that capture how a setting works before you know what matters in it. Structured instruments need the same definition discipline as any survey form; open notes need a disciplined habit of separating description from interpretation, written up the same day [1].

The known threat is the observer effect: people behave differently when watched, particularly for compliance-flavoured behaviours. Mitigations are familiar — repeated visits so the observer becomes furniture, unannounced timing where ethical, triangulation against records — but the honest posture is to report the effect as a limitation rather than claim to have abolished it [1][3].

From talk to evidence

Between fieldwork and findings sit four decisions that determine whether the qualitative component produces evidence or vibes:

1. Transcription and translation. Decide, per study, what gets full transcription versus structured notes, and in which language coding will happen. Translating before coding loses nuance; coding in-language and translating quotes is often better, but demands bilingual analysts. Whatever is chosen, record it [1].
2. Coding. Apply labels to segments of text systematically — a starter codebook from the evaluation questions, extended with codes that emerge from the data. Two analysts coding a sample of transcripts and comparing is the affordable reliability check.
3. Saturation as the stopping rule. Fieldwork can stop when additional interviews or groups stop producing new codes and themes for the questions at hand [1]. Claiming saturation requires evidence — typically a simple account of when new codes stopped appearing — not the assertion alone. Where the sample was fixed by budget before saturation could be assessed, say that instead.
4. Analysis that can be inspected. Findings should trace to coded evidence: how many sources support the theme, from which segments, with what exceptions.

Structured story techniques put a formal spine under parts of this pipeline — Most Significant Change turns collection and selection of stories into documented, deliberative steps [4].

Quality markers reviewers look for

• Quotation discipline. Quotes illustrate an analysed pattern; they are not the analysis. Every quote carries an anonymised source tag, and no single articulate informant supplies half the report’s voice.
• Negative-case reporting. The account that did not fit the pattern is reported and interpreted, not dropped. Nothing raises confidence in a qualitative finding faster than seeing its exceptions handled honestly [3].
• A positionality and method statement. Who collected the data, in what language, with what relationship to the programme, and how that may have shaped what was said — stated plainly [2].
• Triangulation made visible. Where interview, group and observational evidence converge, show it; where they conflict, report the dissonance and what was done about it (see mixed methods) [3].

Checklist for a defensible qualitative component

• Each method is matched to a question it is actually suited to answer.
• The purposive selection logic — positions, variation sought, expected numbers — is written down before recruitment.
• Guides are pretested; facilitators and note-takers are trained and rehearsed.
• Consent covers recording, quotation and the limits of confidentiality in groups.
• Transcription, translation and coding decisions are documented; a second coder checks a sample.
• Saturation (or its absence) is assessed and reported.
• Findings trace to coded evidence, with negative cases and dissonance reported.