Sampling Designs - Sampling Spatial Units for Agricultural Surveys - page 125

Fig. 6.5 Sample selected with maximal stratification in 100 strata, one sample unit within each
stratum
6.6 Multi-stage Sampling
Until now we have considered a conceptual list of units, where the units have an
identification variable that is used for sample selection. In the previous section, we
introduced a partition of this list, which is regarded as a valuable tool for grouping a
set of similar units to reduce the sampling variance. However, such a partition may
also require the definition of the statistical unit to be sampled, particularly if it is
performed according to the codes of a hierarchical nomenclature. This is because of
a feature of hierarchical nomenclatures in which the codes at any aggregation level
may be used to define a statistical unit.
There is often a large number of these nomenclatures available in the sampling
frame that are also inherent from the definition of the statistical unit. For example,
families and individuals are two possible levels of a hierarchical classification, as
are firms and establishments. However, in many cases, the most classic geograph-
ical nomenclature that partitions a country in several hierarchical administrative
levels can be used as statistical unit (i.e., regions, counties, municipalities, etc.).
The structure of a hierarchical classification implies that any unit of a level will
include a cluster of units of the lower level. Thus, performing a random selection of
units relative to a fixed aggregation level leaves several options for the units in the