Qualitative research methods in public health

In short

Qualitative research in public health studies meaning, behaviour and context — why people do what they do about health — using words, observation and documents rather than counts. It answers "why", "how" and "for whom" questions that a survey cannot: why mothers skip the measles-rubella second dose, how a TB patient experiences stigma, why a PHC nurse bypasses a protocol. MPH and MD (Community Medicine) syllabi examine it in the Research Methods paper (methods, sampling, analysis, rigour, ethics), and it is tested in interviews at NIE/NIRT and other ICMR institutes, IIPS, PHFI/IIPH, NIHFW, state health societies, and in programme and M&E roles at WHO-India, UNICEF, BMGF-funded units, Jhpiego, PATH, IHAT, Piramal Swasthya, Karuna Trust, SEARCH Gadchiroli and SNEHA Mumbai. Day to day it is used for formative research before a campaign, process evaluation during it, and explaining unexpected quantitative findings after it.

Explained

What it is, and when to choose it

• Definition: systematic enquiry into the meanings, experiences, practices and social contexts of health, producing textual or visual data analysed interpretively rather than statistically.
• Philosophical base: quantitative work rests on a positivist/post-positivist view (one measurable reality); qualitative work rests on interpretivism/constructivism (reality is socially constructed, multiple) and sometimes critical theory (power, inequity) or pragmatism (mixed methods).
• Logic: mostly inductive — theory emerges from data; quantitative work is largely deductive (hypothesis tested). Qualitative studies state research *questions* and objectives, not hypotheses, and do not calculate power.
• Choose qualitative when the phenomenon is poorly understood, the question is about process/meaning/acceptability, you need to develop an instrument or intervention, or you must explain a quantitative result.
• Do not choose it to estimate prevalence, coverage, risk, effect size or any population parameter — it has no sampling frame and no generalisable denominator.
• Common study designs: phenomenology (lived experience of illness), grounded theory (build theory of a process, e.g. treatment-seeking), ethnography (culture of a community or a ward), case study (one programme, one outbreak response), narrative research (life stories), qualitative description (most applied programme research), participatory action research (community acts on its own findings).

Sampling and sample size

• Non-probability by design. Types: purposive (information-rich cases), maximum variation, homogeneous, typical case, extreme/deviant case, critical case, confirming and disconfirming, theoretical (grounded theory — sampling driven by emerging categories), snowball/chain referral (hidden groups: sex workers, people who inject drugs, undocumented migrants), convenience (weakest), and quota.
• Key informant = someone with special knowledge of the system (ANM, ASHA, MO-PHC, panchayat member, chemist, faith healer), not someone typical of the population.
• Sample size is governed by saturation, not formulae. Data saturation = no new codes/themes from further data collection; theoretical saturation = categories and their properties are fully developed.
• Practical guides used in protocols (indicative, not rules): 6–12 IDIs per reasonably homogeneous group; 4–8 FGDs per population stratum (saturation often reached by 3–6 FGDs per stratum); case studies 4–10. Always justify the number by saturation plan plus strata, and say you will add interviews if saturation is not reached.
• Stratify by what matters: sex, age group, caste/tribe, rural-urban, literacy, cadre, exposure to the programme. Each stratum needs its own saturation.

Data collection methods

• In-depth interview (IDI): one-to-one, 45–90 min, semi-structured topic guide with open questions and probes; for sensitive, personal or deviant experience. Variants: narrative interview, life history, cognitive interview (to pre-test survey questions), exit interview at a facility.
• Focus group discussion (FGD): 6–10 (6–12) participants, homogeneous in the ways that affect speech (never mix mothers-in-law with daughters-in-law, or ASHAs with their supervising ANM), 60–90 min, led by a moderator with a note-taker; produces norms, shared language, debate and range of views — not individual prevalence or private experience. Risks: dominant speaker, social desirability, group conformity.
• Group interview / community group discussion: larger, less interactive; used in rapid assessments.
• Participant and non-participant observation: structured or unstructured; observation checklists for facility readiness, infection prevention, IMNCI practice, waiting time, provider-patient interaction; recorded as field notes (descriptive + reflective).
• Free listing, pile sorting, ranking and rating: elicit local illness terms and categories (e.g. names and causes of childhood diarrhoea); semi-quantitative, from cognitive anthropology.
• Participatory tools (PRA/PLA): social mapping, resource mapping, transect walk, seasonal calendar (disease and income seasonality), daily activity clock, Venn/chapati diagram of institutions, problem tree, wealth ranking, body mapping.
• Document and content analysis: case sheets, HMIS registers, ASHA diaries, programme guidelines, media and social-media content, policy documents.
• Other: photovoice, video and audio diaries, mystery client/simulated client studies (STI drug dispensing, abortion services), vignettes, Delphi for expert consensus, verbal autopsy narratives (the open narrative part), social autopsy of a maternal death, and community-based monitoring jan samvad records.
• Rapid approaches: RAP (rapid assessment procedures), RARE, rapid qualitative appraisal for outbreak and emergency response (used extensively in COVID-19 vaccine-hesitancy work in India).

Instruments, fieldwork and data management

• Topic guide: 8–15 broad domains, each with open lead question and probes; opens with easy rapport questions, sensitive items in the middle-late, closes with "anything we have not asked". Pilot it and revise — guides are living documents.
• Probes: silence, "tell me more", echo, elaboration, clarification, contrast ("how is it different for a boy?"), hypothetical. Avoid leading, double-barrelled, closed and jargon questions.
• Language: develop in English, translate to the local language, back-translate to check, pilot for idiom (for example a local word for "depression" or "white discharge" rather than a clinical term).
• Recording: digital audio with consent; verbatim transcription in the language of interview, then translation; keep an audit trail — guides, field notes, memos, coding decisions, version-dated transcripts.
• Interviewer effects: sex-matched interviewers for sexual and reproductive health; avoid interviewing staff in front of superiors; interview adolescents away from parents where ethically permitted.
• Data security: de-identify at transcription, use participant codes, store audio and transcripts encrypted, destroy identifiers per protocol. Indian privacy law in transition: the Digital Personal Data Protection Act, 2023 is enacted and the DPDP Rules, 2025 are notified, but the Data Fiduciary duties (notice, consent, breach intimation, erasure) apply only from 13 May 2027 — until then IT Act section 43A and the SPDI Rules, 2011 govern sensitive personal data, including health data.

Analysis

• Steps common to most approaches: (1) transcribe and clean; (2) immerse — read and re-read; (3) open coding line by line; (4) group codes into categories; (5) develop themes and sub-themes; (6) build a codebook (code name, definition, inclusion/exclusion, example quote); (7) charting/matrix display by case and theme; (8) interpret — link themes, look for deviant cases, build explanation; (9) report with illustrative, attributed-by-code quotes.
• Coding types: inductive (emergent) vs deductive (a priori framework, e.g. Health Belief Model, socio-ecological model, COM-B, Andersen's model, WHO building blocks); in-vivo codes use participants' own words.
• Named approaches:
  • Thematic analysis (Braun & Clarke's six phases) — the default for applied public health.
  • Framework analysis (familiarisation → identifying a thematic framework → indexing → charting → mapping and interpretation) — suits policy and programme work with teams and deadlines.
  • Grounded theory — open, axial and selective coding; constant comparison; memo writing; theoretical sampling until theoretical saturation.
  • Content analysis — can be quantitative (counts of codes) or qualitative (latent meaning).
  • Narrative analysis, discourse analysis, interpretative phenomenological analysis (IPA).
• Software (CAQDAS): NVivo, ATLAS.ti, MAXQDA, Dedoose (paid); QDA Miner Lite, Taguette, RQDA/quanteda in R (free). Software codes and retrieves — it does not analyse or "find themes" for you.
• Inter-coder agreement: two coders code a subset independently, compare, reconcile; Cohen's kappa or percentage agreement may be reported where a positivist audience expects it, though many qualitative traditions prefer consensus coding and reflexive discussion.

Rigour, ethics and reporting

• Trustworthiness (Lincoln & Guba) replaces validity/reliability: credibility (≈ internal validity) — prolonged engagement, triangulation, member checking, peer debriefing, negative-case analysis; transferability (≈ external validity) — thick description so a reader judges fit; dependability (≈ reliability) — audit trail, stepwise replication; confirmability (≈ objectivity) — reflexive journal, data showing the interpretation.
• Triangulation: data source (patients, providers, family), method (IDI + FGD + observation), investigator (multiple coders), theory.
• Reflexivity: document how the researcher's cadre, caste, gender, language and funder shape questions, rapport and interpretation — expected in a methods section, not optional.
• Ethics in the Indian setting: review by an Institutional Ethics Committee registered with CDSCO/NEC under the New Drugs and Clinical Trials Rules, 2019 where applicable, following the ICMR National Ethical Guidelines for Biomedical and Health Research Involving Human Participants, 2017; informed consent (written, or witnessed thumb impression with an impartial witness for non-literate participants; separate consent for audio recording), confidentiality and anonymity, voluntariness, no coercion, reimbursement of time and travel but not inducement, assent plus parental consent for minors, special care with vulnerable groups, and a referral pathway when an interview uncovers distress, violence or untreated illness. Community consent/permission from the panchayat or gatekeeper does not replace individual consent.
• Reporting checklists: COREQ (32 items, interviews and focus groups), SRQR, ENTREQ (qualitative evidence synthesis), CASP qualitative appraisal tool, GRADE-CERQual for confidence in findings from qualitative evidence synthesis; CONSORT/TIDieR for the quantitative side of mixed-method trials.
• Mixed methods designs: convergent parallel (QUAN + QUAL together), explanatory sequential (QUAN → QUAL to explain), exploratory sequential (QUAL → QUAN to measure), embedded (QUAL inside a trial as process evaluation); report integration explicitly, not as two separate papers.

Key points

• Qualitative research - systematic study of meaning, experience and context; answers why and how, never how many.
• Interpretivist paradigm - multiple socially constructed realities; hence no single "true" value to estimate.
• Inductive logic - themes emerge from data; no hypothesis, no power calculation, no p-value.
• Purposive sampling - information-rich cases chosen deliberately; representativeness is not the aim.
• Theoretical sampling - grounded-theory sampling driven by categories emerging during analysis.
• Saturation - the stopping rule for sample size; data saturation (no new codes) vs theoretical saturation (categories complete).
• IDI - private, deep, individual experience; FGD - group norms, shared language, range of opinion; they are not interchangeable.
• FGD composition - 6-10 homogeneous participants; mixing power hierarchies silences the weaker party.
• Key informant - knowledgeable about the system (ASHA, ANM, MO), not typical of the population.
• Topic guide - open lead questions plus probes, piloted, revised during fieldwork.
• Thematic analysis - familiarisation, coding, generating themes, reviewing, defining, writing up.
• Framework analysis - familiarisation, framework, indexing, charting, mapping and interpretation; good for teams and policy deadlines.
• Grounded theory coding - open → axial → selective, with constant comparison and memos.
• Trustworthiness - credibility, transferability, dependability, confirmability replace validity and reliability.
• Triangulation - cross-checking across sources, methods, investigators and theories to strengthen credibility.
• Reflexivity - explicit account of how the researcher influenced data and interpretation.
• COREQ - 32-item reporting checklist for interview and focus-group studies; SRQR and ENTREQ are the alternatives.
• ICMR 2017 ethical guidelines + registered IEC - the governing framework in India; non-literate consent uses thumb impression with an impartial witness.
• CAQDAS - NVivo/ATLAS.ti/Taguette organise and retrieve coded text; the analysis remains the researcher's.

Formulas and examples

Qualitative research has few formulae; those examined are agreement and the semi-quantitative measures from free listing and pile sorting.

Example 1 — inter-coder agreement. Two coders independently code 120 text segments from TB-stigma interviews; they agree on 102. Each coder applied "anticipated stigma" to about 40% of segments and "no stigma code" to 60%.

• Po = 102/120 = 0.85.
• Pe = (0.40 × 0.40) + (0.60 × 0.60) = 0.16 + 0.36 = 0.52.
• κ = (0.85 − 0.52) / (1 − 0.52) = 0.33 / 0.48 = 0.6875 ≈ 0.69 → substantial agreement.
• Reporting 85% agreement alone would overstate rigour; kappa corrects for chance.

Example 2 — saturation across interviews. New codes per IDI on measles-rubella refusal: I1 = 14, I2 = 9, I3 = 6, I4 = 3, I5 = 2, I6 = 1, I7 = 0, I8 = 0, I9 = 0.

• Cumulative codes = 14, 23, 29, 32, 34, 35, 35, 35, 35 → total 35 codes.
• By I6, 35/35 = 100% of final codes had appeared; after I4, 32/35 = 91.4%.
• No new codes for three consecutive interviews (I7-I9) → saturation declared at interview 6, with 3 confirmatory interviews. Report "saturation reached at 6 IDIs, 9 conducted per stratum".

Example 3 — free listing and salience. Four mothers free-list local causes of child diarrhoea. "Nazar/evil eye" appears in 3 of 4 lists, at rank 2 of 5, rank 1 of 4 and rank 3 of 6.

• Frequency = 3/4 × 100 = 75%.
• Per-respondent salience: (5 − 2 + 1)/5 = 0.80; (4 − 1 + 1)/4 = 1.00; (6 − 3 + 1)/6 = 0.667; non-mentioner = 0.
• S = (0.80 + 1.00 + 0.667 + 0)/4 = 2.467/4 = 0.617.
• Interpretation: a highly salient lay cause — any BCC message on ORS must engage it rather than dismiss it.

Example 4 — sample size and field budget for a protocol. Study of ANC drop-out in two districts of Maharashtra; strata = pregnant women who dropped out, women who completed 4 ANC visits, ASHAs, ANMs/MOs.

• IDIs: 10 drop-out + 8 completers = 18 women; key informants: 8 ASHAs + 6 ANM/MO = 14. Total IDIs = 32.
• FGDs: 3 per district with mothers-in-law = 6 FGDs × 8 participants = 48 participants.
• Total participants = 32 + 48 = 80.
• Transcription at 4 hours per hour of audio, average IDI 60 min and FGD 90 min: audio = 32 + 9 = 41 hours → 164 person-hours of transcription.
• Cost at ₹600 per audio-hour for transcription plus translation = 41 × 600 = ₹24,600; two research assistants for 60 field days at ₹1,200/day = ₹1,44,000; travel ₹35,000; audio recorders and encrypted storage ₹20,000; software (free CAQDAS) ₹0 → indicative total ₹2,23,600. (Rates vary by state and agency.)

Example 5 — choosing the method from the question.

Where it is used

• Formative research for national programmes: message and material development for Mission Indradhanush/U-WIN, NTEP (TB) stigma and adherence studies, POSHAN Abhiyaan feeding-practice research, Ayushman Arogya Mandir (HWC) service-uptake studies.
• Process evaluation and implementation research: ICMR institutes (NIE Chennai, NIRT, NIRRCH, NIN) and implementation-research grants use IDIs, FGDs and observation to explain why a proven intervention under-performs in the field; standard content of ICMR-NIE short courses.
• Social and behaviour change communication (SBCC) agencies and NGOs: SNEHA, SEARCH, Karuna Trust, PATH, Jhpiego, Piramal Swasthya, CARE India — free listing, pile sorting and PRA tools before campaign design.
• Job roles that require it: Research Associate / Project Officer in MPH programmes, Monitoring & Evaluation Officer, Implementation Research Fellow, Qualitative Research Coordinator, consultant to State Health Societies under NHM, social scientist posts in ICMR and NIHFW advertisements.
• Market and policy research for health products: acceptability of HPV vaccine, menstrual cups, ORS-zinc co-packaging, telemedicine (eSanjeevani) usability studies — user interviews and usability observation.
• Outbreak and emergency response: rapid qualitative assessment of rumours and vaccine hesitancy (COVID-19, Nipah in Kerala), community perception tracking feeding risk-communication and community-engagement (RCCE) plans.

Interview questions

• Q: When would you choose a qualitative design over a survey? A: When the question is about meaning, process, acceptability or explanation of an unexpected finding, or when the phenomenon is too poorly understood to write closed questions for.
• Q: How do you decide sample size in a qualitative study? A: By saturation — you plan an indicative number per stratum, monitor new codes as data come in, and continue until further data yield no new codes or categories; there is no power calculation.
• Q: What is the difference between data saturation and theoretical saturation? A: Data saturation is no new codes or themes from additional data; theoretical saturation is when the properties and dimensions of a theoretical category are fully developed, which usually needs more and targeted data.
• Q: When is an FGD the wrong method? A: For sensitive, stigmatised or private topics (sexual behaviour, abortion, violence, mental illness), for individual life histories, and for anything needing prevalence — group pressure distorts disclosure and FGDs cannot produce individual-level estimates.
• Q: How many participants in an FGD and why homogeneous? A: 6-10 (up to 12); homogeneity on the attributes that affect speech ensures participants speak freely without hierarchy — mixing an ANM with her ASHAs, or mothers-in-law with daughters-in-law, silences the junior party.
• Q: How do you ensure rigour in qualitative work? A: Through trustworthiness — triangulation and member checking for credibility, thick description for transferability, an audit trail for dependability, and a reflexive journal with data-grounded interpretation for confirmability.
• Q: What is triangulation, with an example? A: Cross-checking a finding across sources, methods, investigators or theories — for example confirming reported infection-prevention practice in a labour room through staff interviews, direct observation and register review.
• Q: Explain thematic analysis in steps. A: Familiarisation with the data, generating initial codes, searching for themes, reviewing themes against the data, defining and naming themes, and writing up with illustrative quotes.
• Q: How is framework analysis different from thematic analysis? A: Framework analysis adds an explicit indexing and charting stage into a matrix of cases by themes, which makes team working, case comparison and policy-deadline analysis easier while keeping the link to raw data.
• Q: Does NVivo analyse your data? A: No — it stores, codes, retrieves and visualises; identifying and interpreting themes remains the researcher's judgement, and the same is true of free tools such as Taguette.
• Q: How do you take consent from a non-literate participant? A: Read the information sheet aloud in the local language, obtain a thumb impression attested by an impartial witness, record the process as required, and take separate consent for audio recording, following the ICMR 2017 national ethical guidelines and the registered IEC's approved procedure.
• Q: An interviewee discloses intimate partner violence and suicidal thoughts. What do you do? A: Stop probing, respond supportively, and follow the protocol's pre-agreed referral pathway — a named counsellor, One Stop Centre or health facility — which the ethics committee requires to be in place before fieldwork.
• Q: What is reflexivity and why does it matter in Indian fieldwork? A: It is the documented account of how the researcher's gender, caste, language, cadre and funding shape rapport and interpretation; a doctor in a white coat interviewing a patient about provider behaviour will hear a different answer from a community researcher.
• Q: Can qualitative findings be generalised? A: Not statistically; they offer transferability — the reader judges applicability to another setting from thick description of context, participants and process, and analytic generalisation to theory is legitimate.

Common mistakes

• Reporting percentages of respondents ("60% of the 10 women said...") as if a purposive sample were a survey sample.
• Writing a null hypothesis, a power calculation or a sample-size formula into a qualitative protocol; or claiming "random sampling" for what is purposive.
• Using FGDs for sensitive topics, or mixing hierarchies in one group and then puzzling over silent participants.
• Confusing codes with themes — presenting a long list of descriptive codes, or a quote dump, in place of interpreted themes with explanation of relationships.
• Treating transcription and translation as clerical: losing local idiom, kinship terms and tone destroys the data; and translating before coding without back-translation checks.
• Saying "saturation was reached" without any evidence — no record of new codes per interview, no stated stopping rule.
• Omitting reflexivity, the audit trail and ethics detail (IEC approval number, consent for recording, referral pathway), which COREQ and examiners specifically look for.
• Claiming causation or population-level effect from qualitative data, or dismissing the method as "soft" and anecdotal rather than systematic.

Related topics

Public health research: protocol writing and ICMR ethical guidelines · Epidemiological study designs · Biostatistics: sampling and estimation · Health promotion and behaviour change communication · Programme monitoring and evaluation · Implementation research and health systems research