# Welcome — the question that can do harm A 45-minute session for teams who collect data from people about things that are dangerous to disclose: violence, debt, caste discrimination, sexuality, migration status, health. This session is about the decisions taken before a questionnaire is finalised. It is not legal advice and it does not replace an ethics review. Mark each step **done**, or **stuck** to bring the facilitator to you. # The asymmetry When you ask a person a sensitive question, the risk is not shared. You carry a professional risk: a difficult interview, an awkward finding, a delayed report. They may carry a risk to their safety that begins the moment the household notices what was asked. That asymmetry should drive the design, and in most survey planning it does not appear at all. The design is driven by the analysis plan, and the ethics section is written afterwards to match it. Start from the other end for forty-five minutes. # The safety decisions, before the questionnaire The WHO's ethical and safety recommendations for research on violence against women, first published in 2001 and extended to intervention research in 2016, set out a standard that has become the reference point in this field. Several of its provisions constrain the *sampling design*, which is why it cannot be read after the instrument is finalised. Worth knowing before you write a question: - Only one woman per household is interviewed about violence — asking two creates a risk that one will be asked what the other said. - The interview happens in genuine privacy, and the fieldwork plan has to make that achievable rather than aspirational. - The survey is introduced under a broader framing than "a survey about violence", because the title itself can be dangerous. - Interviewers are trained, and are women. - Every respondent, regardless of what they disclose, leaves with referral information. - Interviewer wellbeing is planned for, not left to individuals. Read those and note which ones your last data collection would have failed. # Your own instrument Take a sensitive question from an instrument you use. Answer four things in writing. 1. **What happens to this respondent if a household member learns she answered?** Be concrete about the household, not the average household. 2. **Where does the answer live?** Device, server, shared drive, a printed roster with names in a bag. Trace it to the end. 3. **If she discloses, what do you do?** Name the service. If there is no functioning referral within reach, note that — it changes whether you should ask at all. 4. **What is the analysis you would lose by not asking?** Name it precisely. Sometimes the honest answer is that the question serves the report rather than the respondent, and that is a reason to cut it. #[quiz] Safety by design ## Interviewing only one woman per household about violence is primarily intended to: - [ ] Reduce fieldwork cost - [ ] Improve the representativeness of the sample - [x] Prevent one respondent being questioned by the household about another's answers - [ ] Simplify the sampling frame ## A referral pathway should be established: - [ ] After the first disclosure occurs - [ ] At the analysis stage - [x] Before any question about violence is asked in the field - [ ] Only where local services already exist # When the indicator itself is unsafe A separate failure, and a common one: choosing an outcome indicator that rewards silence. Consider a programme that trains front-line workers to ask about violence and refer. Someone proposes "reported prevalence falls by 5 points in 18 months" as the headline outcome. Work through what that indicator rewards. Successful work of this kind usually raises disclosure first — women learn what counts, and where to go. An indicator measured on falling reports penalises exactly the trust the programme was built to create. Better candidates are things the programme controls: whether a woman who disclosed reached a service, the completeness of referral follow-up, response quality on re-contact. Prevalence is a population statistic on a stable instrument over long periods. It is not a short-run programme indicator. #[quiz] Indicators that reward the wrong thing ## Using falling reported IPV prevalence as an 18-month programme outcome is risky because: - [ ] Prevalence data is always unreliable - [x] Work that succeeds often raises disclosure first, so a fall may record less reporting rather than less violence - [ ] Eighteen months is too short for any health indicator - [ ] It requires a randomised design # Close One thing to take back: for the most sensitive question in your current instrument, write down what happens to the respondent if her answer is discovered, and where the answer is stored. If you cannot answer both, the question is not ready to go to field — whatever the analysis plan says. **Going further.** Module 8 of [Gender-Sensitive MEL](https://www.impactmojo.in/courses/gender-mel/) covers this in depth, and [Safeguarding & PSEA 101](https://www.impactmojo.in/101-courses/safeguarding-psea.html) covers the organisational duty that applies to a research team in a village for three weeks.