|Page tools: Print Page Print All RSS Search this Product|
Robust Modelling of Design Effects for Household Survey Design
To date the ABS lacked a consistent and reliable set of methods to model design effects (DEFFs) from past survey data in order to assess the sampling error properties of proposed household surveys. Therefore a paper was composed that articulated the key challenges in modelling robust DEFFs for household surveys and examined general solutions in the context of two recent sample design applications.
The first sample design application was for the well-established Survey of Income and Housing 2013/14. In this instance, the effect of alternative cluster sizes on sample design considerations was investigated via a comprehensive simulation approach.
The second application was the newly formed Health Services Survey. This instance was one where relatively fine-level stratification was conducted, and the design was therefore made challenging through limited historical data being readily available at such fine levels.
A focus was placed on seeking to consolidate best-practice around design effect modelling using historical survey data, rather than census data. A subsequent assessment of the suitability of proposed solutions was conducted in order to gauge appropriateness as standard methods for ongoing and future application.
The key challenges include:
· modelling design effects under alternate sample clustering scenarios
· addressing the small data problem for specific sub-populations of interest - volatility undermining the design optimisation
· assessing the utility of design effects in meeting a multiplicity of output objectives, summarising across multiple items of interest.
This research was presented to the Methodological Advisory Committee in June 2013.
These documents will be presented in a new window.