Counting Drug Use in a District: Definitions, Ethics and Sampling

The police figure of 340 registered cases counts enforcement, not use: it records people caught, and it rises and falls with policing activity as much as with consumption. The sociological definition would be any non-medical use of a psychoactive substance in a defined reference period, whether or not a case was registered. The gap between the two is the dark figure of unreported and undetected behaviour, and in a district where possession itself is an offence, that gap is likely to be very large, because admitting use to a stranger is admitting to an offence. The de-addiction figure has a different bias: it is a treatment population, so it over-represents severe dependence and those with family support, and its 78 per cent of admissions aged 18 to 30 says nothing about prevalence among 15-year-olds. Choosing self-report in a confidential survey will produce a number far above 340 and closer to actual use; choosing registered cases will produce a defensible administrative figure that understates. The district must therefore be told which question its quoted number answers, or it will treat an enforcement count as a prevalence rate.

The officer's proposal is the more serious. Police accompanying investigators converts a research interview into an interrogation: respondents cannot give free consent when the person beside the interviewer can arrest them, and any disclosure of use becomes self-incrimination. It would also destroy the study's validity, since only those who trust the police would speak, giving a falsely low figure - the proposal fails ethically and methodologically at once. The safeguard is strict separation: no police presence, no names collected, no list of respondents shared with the administration, and a written undertaking to that effect before fieldwork. The NGO's plan has a different, milder problem: people in treatment are a captive and vulnerable group, dependent on the centre for their bed, so consent is compromised unless staff are absent from recruitment and refusal cannot affect treatment. Six weeks at one centre also cannot estimate prevalence. Both plans need informed consent in language the respondent uses, anonymity and secure storage, and a clear rule that a respondent who asks for help is referred without any record linking them to the study.

I would use a two-stage design: select a stratified random sample of villages and urban wards, then use the electoral roll and school and college enrolment lists as partial frames, supplementing with a household listing in each selected cluster to catch those in neither. Target roughly 1,500 completed interviews, which gives usable precision for a district estimate and is achievable in four months with a small team. Unlike a prevalence survey of rare severe dependence, this design does not require huge numbers because the outcome is a proportion in the population, not a count of cases. The online questionnaire fails on two grounds that are fatal in this district: it is self-selected, so it reaches the most connected and most willing, and it systematically misses poorer youth, girls with restricted phone access, and out-of-school young men - probably the highest-risk groups. It would also confound prevalence with the willingness to disclose. The cluster design costs more per interview but delivers a figure the district can defend, and a short anonymous self-completion section protects disclosure.