Advances in Sampling Estimation - Sampling Spatial Units for Agricultural Surveys - page 261

1. We can reduce the undercoverage effects, because we assume that a statistical
unit is recorded in at least one of the frames.
2. We can integrate the outcomes of two or more surveys in such a way that the
estimates of common variables are coherent (this is also one of the most
important reasons to design a multi-phase sampling (Sect. 6.7 ) and coordinate
several samples (Sect. 6.8 ).
3. We can more efficiently use administrative data by considering them on a
separate frame and as additional auxiliary information.
4. We can estimate the measurement error of variables observed on different
frames, because interviews on different frames are often based on multiple
modes (de Leeuw 2005 ).
5. Some variables can only be observed in a specific frame, so the only way to get a
complete picture of a phenomenon is to merge several surveys performed on
different frames.
6. A national survey is rarely planned to produce detailed estimates, so it can be
supplemented with smaller, localized surveys to improve accuracy in some areas
(Chap. 11 ) .
7. A specific survey from a frame with a high concentration of units can also be
used to assist a more general survey by estimating a rare population (Sect. 6.10 ).
8. We can obtain several quality parameters regarding the coverage rates of a single
frame as a byproduct of a multiple frame survey.

However, we should not forget that it is more difficult to manage a multiple
frame survey if there are several survey steps and organizational rules, which are
usually different from one frame to the other. This independence makes it very hard
to determine the required links between the surveys.
Moreover, data collection is even more complex, because we always need to
detect and record the domain in which every unit falls. These domains are defined
by the overlap of the frames and are usually unknown before the data collection
phase (see Fig. 10.3 ). In general, we expect that multiple frames are more sensitive
to non-sampling errors for these reasons, and are subject to more types of

non-sampling errors than their single frame counterparts.
It is important to emphasize that the term multiple frames is often erroneously
used to refer to surveys that select a random sample from an available frame (e.g., a
spatial frame) to detect and interview a set of statistical units that are different from
the selected units, but satisfy some rule. For example, we can select a set of random
points over an agricultural area, and perform the survey on the owners of the fields
that contain a selected point. Such designs are necessary, for example, because it
may not be possible to compile a reliable list frame of farms. These designs belong
to the broad category of indirect sampling techniques, and their main target is to
evaluate the unknown inclusion probabilities of the interviewed units as a function
of the known inclusion probabilities of the selected units (for a review, see Lavall´e
2007 ).
One of the fundamental characteristics of a spatial frame in agricultural surveys
(see Chap. 5 ) is that, by definition, it cannot suffer problems of coverage. However,