Small Area Estimation - Sampling Spatial Units for Agricultural Surveys - page 294

To compute the SEBLUP, we must define an appropriate neighborhood struc-
ture. In our simulated case study, we have 9 SAs, and so the N ( d ) system can be
defined using the following R code.

> domcoor < -cbind(floor(datFH$coddom/10),datFH$coddom-floor
+ (datFH$coddom/10)*10)
> cmatr < - as.matrix(dist(domcoor))
> cmatr[cmatr > 1.5] < -0
> cmatr[(cmatr < 1.5) & (cmatr > 0)] < -1
> cmatr < - cmatr/rowSums(cmatr)

The above spatial weight matrix is binary and row-standardized (cmatr/
rowSums(cmatr)), where two domains are considered neighbors if they share
either a border or a vertex (i.e., queen weight matrix).
The SEPLUB estimates can be computed as follows.

> domSFH < - eblupSFH(yobs ~ auxFH - 1, vardir¼var, proxmat¼cmatr,
+ data¼datFH)

> domSFH

$eblup

[,1]
11 77.48649
12 96.03450
13 75.77453
21 96.05400
22 119.18608
23 96.38841

31 78.38464

32 94.50742

33 82.15607

$fit
$fit$method
[1] "REML"
$fit$convergence
[1] TRUE
$fit$iterations
[1] 10
$fit$estcoef
beta std.error tvalue pvalue
auxFH 1.957659 0.03388053 57.78125 0
$fit$refvar
[1] 17.76237
$fit$spatialcorr
[1] -0.4184711
$fit$goodness