|Page tools: Print Page Print All|
Wheat Use Survey Design
The development of the Wheat Use Survey (WUS) provides a good example of the challenges in the design of Agricultural surveys. WUS is one component of a program of user-funded surveys aimed at meeting requirements of the Department of Agriculture, Fisheries and Forestry (DAFF) for information on the storage and use of wheat to determine the levels of uncommitted grain in Australia. The program consists of:
The first two of these surveys have been conducted monthly since the start of the program in October 2008, the grain handlers' survey, initially an annual, has been run monthly since March 2009. The program will has funding to September 2011.
The main data items collected are the amount of wheat used during the reference period and the amount stored at the end of the period. A significant issue in the development of the program was the determination of the target population and the assembly of a frame. Initially, the main users and/or storers of wheat were considered to be: manufacturers of foods, animal feed and organic chemicals, beef feedlot operators and pig and poultry farmers. In the initial month a large sampling fraction was used (3,036 from a population of 6,253) since it was expected that there would be significant numbers of dead or out-of scope units on the frame. The data provided by this first survey allowed refinement of the frame and design with a reduction of the sample size to about 1,101 from a population of 5,143.
During the first year of the program it became apparent that there is significant use or storage of wheat by industries not in scope of the existing collection. In particular, significant amounts of wheat appeared to be stored on farm by growers and the dairy industry also uses wheat as stock feed. It was therefore decided to broaden the scope of WUS to include these sectors. A supplement would be selected from the population of dairy and wheat farmers and added to the existing WUS sample. Before this was done a coverage survey was undertaken which provided design data for the allocation of the supplementary sample.
Design of this supplementary sample illustrates another challenge encountered in the design of agricultural surveys - variables available on the frame for allocating units to size groups generally have only a poor correlation with the design variables of interest. In other economic surveys, frame employment or BAS turnover are used to size units. For agricultural businesses, with their extensive use of mechanisation and contract labour, frame employment is not a good indication of agricultural production. Since many businesses with agricultural operations also derive income from non-agricultural activities, BAS turnover also is not a good measure. For this reason another measure of size has been developed for agricultural units - the estimated value of agricultural operations (EVAO). This is a composite index based on the production of each commodity and the land area allocated to that commodity as reported by the unit in the last Agricultural Census. Three-year averages of commodity prices are used to assign a value to the production of that commodity and the associated land area. The EVAO of the unit is then the sum of these values over all commodities it produces.
EVAO is a good measure of the overall production of a farm, but the specific set of commodities of interest in a particular survey may account for only a part of its production. Furthermore, data items other than commodities may also be of interest. Hence it is possible that EVAO may still not correlate well with major survey design variables. For example, in the case of wheat growers, the annual production of wheat may not be a good indication of the amount stored at the end of any particular reference month (the variable of interest here). Similarly, the milk production from a dairy farm is not necessarily well correlated with the amount of wheat used by the farm since other stock feeds may be used. The consequence of this poor correlation is that strata may be quite heterogeneous with respect to the design variable(s). This means that designs will not usually be efficient. In the case of the WUS supplement, sample sizes, set by budgetary constraints, corresponded to overall sampling fractions of about 1 in 40 for dairy and 1 in 50 for wheat. The design RSEs achievable with these sample sizes were 20% at the national level and 40% at state level, which is in marked contrast to the design performance for other business surveys.
Although the impact of some high estimation weights has raised some issues with the BSC the first two months' estimates from the expanded collections have yielded positive results - with the expanded data adding to the industry's ability to more accurately determine the available national wheat supply.
For more information, please contact Jos Beunen on (02) 9268 4647 or email@example.com.
These documents will be presented in a new window.