Sampling Designs - Sampling Spatial Units for Agricultural Surveys - page 107

The survey package was used to compute the design-based estimates and their
standard errors (see Lumley 2010 for a review of its features). In this simulated
example, we have assumed that the preliminary necessary operations on the data
have been successfully completed. These include data entry, coding, editing, and
imputation (see Chap. 9 ) . User-specific computer applications are often used to
prepare the cleaned data set for analysis.
For design-based estimation, the function svydesign collects all the neces-
sary information for the sampling scheme (i.e., sampling-design identifiers, stratum
identification variables, cluster identification variables) and the sampling weights or
finite population corrections (i.e., the sampling rates) that are required for sampling
without replacement. Undoubtedly, a full awareness of the complexities of the
actual sampling design is required. The function svytotal estimates the popu-
lation totals, and its output includes the point estimates and their estimated standard
errors, the coefficients of variation, and the design effects (but obviously not for an
SRS design). These statistics are calculated using the information recorded in the
design object. Note that the package also includes different functions to estimate
other univariate population parameters such as svymean to estimate the mean,
svyquantile to estimate the quantiles of the survey variable, and svychisq to
estimate a contingency table from the survey data. A set of functions is also
provided for the proper analysis of survey data, which exploit all the necessary
design information (see Chap. 12 ) . Note that when a qualitative variable is included
in the list of variables to be estimated, svytotal automatically realizes that the
estimation objects are the absolute frequencies of each code of the variable. The
coef and SE utility functions extract the point estimate vectors and their standard
errors from the svytotal output. We can use the interaction function to
produce a cross tabulation of two qualitative variables, which creates all combina-
tions of the two sets of codes (the symbol

... indicates that some redundant output is

dropped).

> dsrs < - svydesign(id ¼ ~1,data ¼ framesrs,fpc ¼ ~rep(n/N,n))
> esrs < - svytotal(~yobs+as.factor(q1obs),dsrs)
> esrs

total SE

yobs 91667 2111.419

as.factor(q1obs)1 270 42.330

as.factor(q1obs)2 350 45.477

as.factor(q1obs)3 380 46.280

> coef(esrs)
...
> SE(esrs)
...
> qesrs < -svytotal(~interaction(q1obs,q2obs), design ¼ dsrs)
> ftable(qesrs, rownames ¼ list(q1obs ¼ c("1","2","3"),
+ q2obs ¼ c("1","2","3","4","5")))