How to calculate standard errors for population estimates based on Australian National Health Survey data.

摘要

The National Health Surveys (NHS), conducted regularly by the Australian Bureau of Statistics, are designed to obtain benchmark information on a range of health-related issues and to enable the monitoring of health trends over time. Confidentialised Unit Record Files (CURFs), containing detailed individual-level NHS data, are available from the ABS. When using the CURF data to make an estimate relating to the Australian population, the precision of the estimate should also be calculated, but due to the complexity of the survey design and the limited survey design information in the CURF, this is not straightforward. This paper explains three different methods for calculating standard errors for population estimates obtained using CURF data: (i) using the estimates of relative standard errors published by the ABS; (ii) using the replicate weights provided by the ABS in the CURF; and (iii) assuming the survey design is a simple random sample. The most appropriate method will depend on which survey is being analysed and the type of population estimate; this is discussed in detail. Worked examples for each method using data from the 2001 NHS CURF are provided, together with Stata 9 code illustrating the implementation of each method.