|Page tools: Print Page Print All|
Modelling Small Size Units Using a GREG Estimation Technique
Traditionally the scope of ABS agricultural surveys, including the agricultural census, has included all businesses undertaking agricultural activity above a minimum size cut-off relating to their agricultural operations. Since 1993-1994 this cut-off has been set at $5,000 based on the size variable used in agriculture surveys, Estimated Value of Agricultural Operations (EVAO) or an equivalent proxy based on tax data.
A National Agricultural Statistics Review was undertaken in 2013 and 2014 by the ABS and the Australian Bureau of Agricultural Resource Economics and Sciences (ABARES). In light of the findings from the review, the ABS has reviewed the scope cut-off for its agricultural collections and decided to raise the minimum scope cut-off value for the 2015-2016 agricultural census and Land Management Practices Survey (LaMPS) to $40,000. This change in scope will achieve several benefits, including
· improved coherence of ABS data with data released by ABARES, through aligning the ABS definition of an agricultural business with the definition used by ABARES
· significantly reduced respondent burden (with approximate savings of up to $3 million in provider time)
· improved efficiency of collection processing and timeliness of outputs.
It is anticipated that the increase in the scope cut-off will have minimal impact on the majority of estimates in the 2015-2016 agricultural census and LaMPS. In order to ensure comparability and coherence with data from previous agricultural censuses and surveys can be retained, the ABS has investigated modelling options to account for the impact of the scope change on estimates.
In particular, a Generalised Regression (GREG) estimation methodology has been proposed incorporating a GREG model based on the size variable. The model would be implemented for the small size units just above the new cut-off, with the benchmark totals accounting for those units below the cut-off as well as the units just above the new $40,000 cut-off. In this way, the units above the new cut-off would represent the units below the cut-off through the GREG weight adjustment. The modelling essentially assumes the same model of variable by size exists above and below the new cut-off.
An advantage of the GREG approach is that it does not require the creation and maintenance of separate models for individual items. Additionally, the model can be incorporated into current ABS estimation systems, and the quality of the modelling will be able to be reflected through the production of Relative Standard Errors (RSEs) output by the system.
Results from testing the method on three recent agricultural collections have shown that the GREG modelling approach could be a viable option for accounting for the increased scope cut-off in agricultural estimates. Further work will be undertaken to refine the approach and to determine the best use of these outputs in informing users of agricultural census and LaMPS data.
These documents will be presented in a new window.