Advances in Sampling Estimation - Sampling Spatial Units for Agricultural Surveys - page 249

total SE DEff
yobs 95235.54 594.28 0.6195
> diff < - weights(dppsg)/weights(dpps)
> summary(diff)
Min. 1st Qu. Median Mean 3rd Qu. Max.
-1.33100 -0.62820 0.01796 0.39150 0.99170 5.69000
> par(mar ¼ c(2,2,1,1),mfrow ¼ c(1,2),xaxs ¼ "i",yaxs ¼ "i")
> hist(diff,breaks ¼ 50,xlim ¼ c(-1.5,1.5),main ¼ "")
> text(1,8,"GREG",cex ¼ 2)
> dppsg < - calibrate(dpps,~xc+yc+xc2+yc2, calfun ¼ "logit",
+ bounds ¼ c(0,100),maxit ¼ 50,epsilon ¼ 1e-10,totpop)
> estg < - svytotal(~yobs, dppsg, deff ¼ TRUE)
> estg
total SE DEff
yobs 95548.52 595.57 0.5859
> diff < - weights(dppsg)/weights(dpps)
> summary(diff)
Min. 1st Qu. Median Mean 3rd Qu. Max.
0.05057 0.06542 0.13900 0.45370 0.28550 10.58000
> hist(diff,breaks¼100,xlim¼c(0,1.5),main¼"")
> text(1,27.5,"Logit",cex ¼ 2)
10.4 Adjusting for Nonresponses
Nonresponse is an increasingly important issue in sample surveys. It is generally
due to non-contact, refusal, or an inability to respond to the survey. It can become a
source of bias if not appropriately considered, because non-respondents are often
extremely different from respondents, with respect to the variables of interest. For
this reason, we always need to address the consequences of nonresponses, in
particular by examining and reducing the bias.
Nonresponses can be considered as a particular case of the general topic of
statistics represented by missing data. However, it is useful to classify the type of
nonresponse into unit nonresponse (a selected unit is not observed or interviewed),
and item nonresponse (a unit responds or is observed with respect to some of the
data items in the survey, but we have a nonresponse for one or more items). We can
reasonably assume that item nonresponse does not exist when dealing with spatially
defined units, because the main source of nonresponse is the physical impossibility
of observing a statistical unit.