Sampling Designs - Sampling Spatial Units for Agricultural Surveys - page 135

At first glance, this sampling plan seems quite cumbersome with many artifi-
cially introduced difficulties, both in management and methodological aspects.
However, after a more detailed analysis, there are some considerable advantages.
The difficulties encountered when developing procedures for the random selection
of units are a small price to pay when compared with the number of potential data
editing improvements that have a significant impact on the quality of the estimates.
Further, the ability to easily integrate information from various surveys encourages
the analysis of relationships between the different aspects that characterize a
statistical unit. In other words, such an integrated system of surveys should be
interpreted as a methodological investment aimed at reducing non-sampling error
and at gaining new information by integrating different data sources.
Moreover, we should consider the overlaps between surveys and also among
different periods of the same survey.
Survey methods focus on studies that are designed to produce a snapshot of the
population at one point in time. However, in practice, we are often interested in
obtaining an estimate of the changes that occur over time. The time dimension can
be introduced by repeating the survey, or by using some form of panel design
(Kalton 2009 ).
Many agricultural phenomena vary over time. In this context, a spatial definition
of the statistical units leads to a population that can be considered closed and
homogeneous, in which there is no possibility for the unit or to enter or to exit.
However, over a certain period of time, it is obvious that the units of the
population can change some of their characteristics. The effects of changes to the
units are generally consistent with each other. Hence, the population that results
from the variations of the individual units is still compact and homogeneous,
although different from the original.
In a first class of methods to study the changes over a defined period of time, it is
sufficient to compare the populations at two different periods. Then, a pair of
surveys are needed, one at the beginning of the transformation and the other at
the end. In this case, we talk about cross-sectional surveys, even if they have been
repeated over time.
A second class of possible changes occurs when the units that represent an open
population (farms, households, individuals, and so on) react to external stimuli in a
strong and different way. Then, the units can follow very different paths. At the end
of this process, the population becomes entirely heterogeneous.
Agricultural, but also economic, social, and demographic changes, often fall into
the first category. However, the second kind of phenomena prevails in practice. For
this reason, traditional statistical methods often fail when trying to follow some
particularly complex dynamic of the phenomenon.
Longitudinal or panel surveys can also produce additional information regarding
the following:
• Flow estimates, measured using the transition probabilities from one state to
another with reference to the type and economic, rural, and social conditions for