Sampling strategies for alcohol and drug users: two recent Brazilian experiences
Mauricio Teixeira Leite de Vasconcellos, ENCE/IBGE
Keywords: survey techniques, mobile populations, inverse sampling
Studies with alcohol and drug users (ADUs) pose challenges for using probability samples to survey the population. This paper describes procedures adopted for selecting two population-based samples in Brazil. Study 1 surveyed drivers present at alcohol selling outlets in Porto Alegre, a Brazilian southern state capital. Study 2 is surveying crack-cocaine users across the country. Sampling strategies for both studies comprise stratified three-stage cluster sampling. In Study 1, census enumeration areas (CEAs) were stratified by alcohol outlet (AO) density and selected with probability proportional to size (PPS), size being number of AOs in each CEA. In stage 2 combinations of outlets and shifts (COS) were stratified by prevalence of alcohol-related traffic crashes and sampled with PPS to their squared duration in hours; finally, drivers who drank at the selected COS were stratified by their intention to drive and sampled with inverse sampling. Sample weights were calibrated using post-stratification. In Study 2, 37 geographic strata were defined (27 state capitals, 9 metropolitan regions and all the other municipalities in the country). In this remainder stratum 10 sets of neighboring municipalities were selected. In every sampled municipality an exhaustive list of crack-cocaine consumption scenes was prepared with information on its size and shifts. Shifts were stratified according to the crack-cocaine user numbers, before selecting combinations of scenes and shifts. In the last stage individuals were selected using inverse sampling when they were leaving the scene. In Study 1, 3,118 individuals were approached and 683 drivers interviewed, leading to an estimate of 151,573 drivers who drank at the AOs during the survey period. Study 2 is still in course. Both sampling strategies can be viewed as a step forward towards better surveys with these mobile and hard-to-reach populations.