The above examples define the various populations associated with the sets of core data. Table 1 provides an overview of how populations can be defined to establish sampling and reporting units. There are essentially four populations: farms in the household sector, enterprises (farms in the non-household sector), households, and land areas. The sampling frame is the listing of farms, households, enterprises, and land areas that also define the sampling units. The sampling unit is the unit of selection from the sample frame. The listing of population units can be determined in one stage or in two stages. When two stage methods are used, the first stage is to define area units such as census EAs or administrative areas, or simply large blocks of land identified geographically. Then, listings of farms, households, or segments of land need only be identified in the selected PSUs. The second-stage sampling units are the farms, households, or land segments defined in the respective PSUs. The reporting unit establishes what data must be associated with the sampling unit. The probability of selecting a unit of land, animal, or person is the probability of selecting the sampling unit.
TABLE 1.1
Review of Sampling Frames
Sampling frame Sampling unit Reporting unit Items of interest
The frame is the listing of farms from agricultural or population census/ administrative registers Single-stage: farm-agricultural holding is sampling unit
Farm or holding identified by name or location
Land, livestock, economic, and social variables associated with the holding
Two-stage: Listing of census EAs or administrative areas (PSUs). List of farms from census/administrative registers in selected PSUs. Farm (agriculture holding) is the sampling unit.
Same as above Same as above
Two-stage: Listing of census EAs or administrative areas (PSUs). List of farms from ad hoc exercise of canvassing the selected PSUs- Farm (agriculture holding) is sampling unit.
Same as above Same as above
Listing of households from agricultural or population census/ administrative registers Single-stage: Household is sampling unit. Farm (agricultural holding) associated with the household
Land, livestock, economic, and social variables associated with the household-holding Two-stage: Listing of census
EAs or administrative areas (PSUs).
List of Households from agricultural or population census/administrative registers in selected PSUs. Household is sampling unit.
Same as above Same as above
Two-stage: Listing of census EAs or administrative areas (PSUs). List of households from ad hoc exercise of canvassing the selected PSUs. Household is sampling unit.
Same as above Same as above
Listing of segments of land or points
Single-stage: Land segment or
point is sampling unit.
Land segment or point and data collected by observation or measurement
Crop areas, livestock on land segment or parcel associated with the point
Farm or holding, household usually associated with land segment or parcel containing the point
Land, livestock, economic, and social variables associated with the farm or household- holding
Two-stage: Listing of large blocks of land (PSUs). List of segments of land or points in selected PSUs. Land segment
or point is sampling unit.
Same as above. Same as above
List of commercial agricultural enterprises or large land holdings
Enterprise (due to the small number, usually a single-stage national listing is arranged.)
Enterprise is sampling unit.
Enterprise Crops,livestock and economic variables associated with enterprise
Single-stage sampling designs offer the greatest variety of sampling methods, but are probably the most expensive to construct and maintain. For a given sample size, sampling errors will be smaller with a single-stage design; however, this results in greater frame development and data collection costs, because the sample is more widely disbursed than when clustering in two-stage designs is used. It is also difficult to keep the large lists of names updated, which means that the population coverage declines over time.
The choice of sampling frames must take into consideration the structure of agriculture, especially the linkage between farms and the farm and non-farm households.
An area frame is a listing of land areas that can be compiled in a single stage or in multiple stages. The land areas in the single-stage or multiple-stage selections are described by geographic boundaries or by geo-referenced boundaries. If satellite imagery is available, the land areas can be classified by land cover, which enables cultivated land to be separated from woodland and urban areas. The listing of land areas is not dependent on any census or administrative data. An area frame provides a means to mount a survey program in the absence of previous agricultural or population census results. The final selection of land areas (segments) or points becomes the sampling units.
The final sampling unit for an area frame is either a segment of land or a point that will be associated with a tract of land. When the final stage of sampling is an area segment, all farms and/or households with land in the segment are included in the sample. The probability of selecting each farm (household) is, simply, the selection probability of the segment. However, the link between farms and the sampled segment is complicated where the physical location of the land operated may cross segment boundaries; therefore, the farms could have multiple probabilities for selection, which requires the use of rules of association. The “open” approach links all land in a farm to the segment containing the farm headquarters or farm operator’s household. Using the “closed” method, the reporting unit is the portion of the farm that is in the sampled segment. The open and closed approaches each have strengths and weaknesses, but both have the advantage of eliminating double reporting. The “weighted” approach allows for multiple reporting: all land area and activities associated with the farm or agricultural holding is reported whenever a segment contains some of its land. The values are then weighted by the fractions of the farm’s total area within the sampled segment. By design, the inclusion of all land in the frame ensures that it provides complete coverage of the population. However, this also depends on choosing reporting units that are also inclusive of the population. The probabilities of selecting farms are based on land characteristics, not on the relative sizes of farms or the presence of items found on only a small subset of farms. Data on crop areas can be observed visually and measured using mapping materials; this enhances data quality, especially if the farm operator is not familiar with area measures.
List frames are lists of farms and/or households obtained from agricultural or population censuses and/or administrative data. It is possible to compile a complete register for the entire country. More often, samples of EAs are first selected, and listings of farms/households are then derived from these sampled units. The frame is complete at the first stage of selection if all EAs have a known probability of being selected. The frame is also complete at the second stage if the name registers are frequently updated in the selected PSUs. However, it is difficult to maintain and update the registers over time, which eventually results in the incompleteness of the list frame coverage. The ideal list frame listing of farms and households will also include information on crop areas and livestock inventories. These measures provide valuable information that can be used for stratification and other types of sample selection. This is especially valuable if the population includes farms that vary widely in terms of the size and coverage of the various crops or livestock.