This problem of sparse data has been regarded as intractable in the past; for example the confidence intervals associated with smoking prevalence in up to 40% of remote regions are suppressed due to unacceptably high standard errors. This means that regions miss out on the evidence of what works in their community. These are typically regions that are coping with multiple levels of disadvantage, for example higher levels of social exclusion and higher levels of risky behaviour such as smoking.
After a review of the state of the art of small area estimation, we will build a multilevel statistical model for the estimation of smoking prevalence in small areas across the whole of Australia. This will lead to mapping the prevalence of smoking prevalence with increased precision in sparsely sampled regions of Australia. A third part of the project will evaluate the impacts of interventions implemented at the small-area level (rather than a State or national) level on smoking prevalence over time.
While tobacco smoking is declining on average over time on both Indigenous and non-Indigenous populations, the prevalence is spatially diverse and distributed unevenly throughout the population. This variation makes monitoring the situation particularly difficult. Yet there is a tremendous demand for such information to ensure proper planning and resource allocation. This is particularly true for regions where policy interventions have taken place with little ability to use data to evaluate their success, and little ability to identify areas of best practice.
