Sampling Designs - Sampling Spatial Units for Agricultural Surveys - page 130

> points(tstd[[2]]$xc,tstd[[2]]$yc, pch ¼ 1, cex ¼ 2)
> points(framepop$xc,framepop$yc, pch ¼ 19, cex ¼ 0.5)
> dataest < - cbind(tstd[[2]],prob2 ¼ tstd[[2]][,"Prob_ 2 _stage"],prob1
+ ¼ rep(0.25,n))
> d2stg < - svydesign(id ¼ ~strataid2+id,data ¼ dataest,fpc ¼ ~prob1+prob2)
> e2stg < - svytotal(~yobs+as.factor(q1obs),d2stg,deff ¼ T)
> e2stg

total SE DEff

yobs 89992.573 6483.776 12.2305

as.factor(q1obs)1 223.000 45.441 1.4483

as.factor(q1obs)2 389.000 51.209 1.3768

as.factor(q1obs)3 320.000 43.555 1.0742

6.7 Multi-phase Sampling

Agencies and institutions that produce agricultural statistics devote considerable
effort into the harmonization of surveys. In particular, they try to avoid surveys with
overlapping aims and objectives, or that request the same information from the
same statistical units. This task reduces the so-called statistical burden when the
statistical units are physical or legal bodies and, even more importantly, allows
different estimates to be integrated with each other. A set of different surveys
performed by the same and/or different institutes generally diverges after a short
period of time. This results in significantly incomparable archives, methods, and
techniques. Furthermore, it is often the case that the responsibility of the survey
belongs to different agencies and institutions, so they may use different definitions
of survey variables, statistical units, and target populations.
We can assume that these inconsistencies are avoidable, if we coordinate
activities and share methodological choices. However, the problem of having a
multiplicity of information collected for the same statistical units would remain,
and we would need multivariate analyses.
These considerations have often led agencies and institutions, particularly the
NSIs, to study complex survey structures. They have investigated the logic of
coordinating samples between surveys, and between periods of the same survey
(see the next Sect. 6.8 ). They have also taken subsamples from a master sample (the
first phase sample, see Sect. 5.1 ) that then assumes the role of the reference frame.
Because data collected in the master sample represent a basis for further surveys,
we should devote particular care to identifying statistical units, their typological
classification, and managing the addresses database. The expensive operations that
constitute the proper management of a frame of statistical units are usually
neglected in sample surveys. It is especially important to pay close attention to
them during the master survey, to avoid biases in the derived surveys. When using