The elusiveness of representativeness in general population surveys for alcohol.

Population survey research is limited by biases introduced through the exclusion of sub-populations from the sampling frame and by non-response bias. This is a particular problem for alcohol surveys, where populations such as the homeless and the institutionalised-who consume on average more alcohol than the general population-are usually excluded, and where people who respond to alcohol surveys tend to consume less alcohol than those who do not. These biases lead to the underestimation of alcohol consumption at the population level, which can be corrected for by triangulating alcohol consumption data with population data sources (i.e. taxation and production). Other methods which account for the biases inherent in surveys include triangulation with outcomes (e.g. traffic injuries), calculation of estimates for groups which are outside common sampling frames, and combining probabilistic sampling with new methodologies, such as computer-assisted web interviews. In particular, population surveys do not attract sufficient participation numbers for certain groups, such as the marginalised urban male youths. In this situation, it may be helpful to add estimates generated via respondent-driven sampling or non-probabilistic web panels restricted to a specific group to such population surveys. Additionally, computer-assisted web interviews perform better for sensitive questions, such as those about personal alcohol use. In sum, based on the objectives, the future of survey will need to include statistical modelling, adding data from external sources for validation and combining data from various types of surveys.