This result is more interesting than that of the weighted Delaunay MEM. The
AICc of the best model, obtained with a threshold of 2 m, is À99.85 with a model
consisting of 4 MEM variables. Let us see if we can improve this result by weighting
the connections by an inverse distance function.
# 3b. Variant: same as above, but connections weighted by the
#
complement of the power of the distances, 1-(d/dmax)^y.
#
Again, after transformation of the distances by function f2,
#
values near 1 are attributed to pairs of sites with easy
#
exchange, values near 0 mean difficult communication.
mite.thresh.f2
MEM.autocor = "positive",
f = f2,
y = 2:10,
dmax = max(unlist(nbdists(x, as.matrix(mite.xy)))),
xy = as.matrix(mite.xy)))
# Lowest AIC, best model
mite.f2.minAIC
# Smallest AICc (best model among the 10)
min(mite.f2.minAIC)
# Number of the model among the 10
(nb.bestmod <- which.min(mite.f2.minAIC))
# Actual dmax of best model
(dmax.best <- mite.thresh.f2[nb.bestmod][[1]]$all[1, 2])
Hints The actual d max value found by the function is often smaller than the d max provided
to the function by means of the vector of user-selected threshold distances,
because the output of the function shows the largest actual distance within the
limit provided by each threshold value. In the example, the 6th value in vector
thresh10, which contains the list of user-selected threshold distances, is
2.671639. There is no such distance in the mite geographic distance matrix; the
function found that the largest distance smaller than or equal to that threshold is
2.668333.
In some applications, the computations are followed by one or more warnings:
1: In nb2listw(nb, style = "B", glist = lapply(nbdist, f),
zero.policy = TRUE) :
zero sum general weights
This means that one or more points have no neighbours under the definition of
neighbourhood provided to the analysis. This may or may not be important
depending on the context of the analysis.
7.4 Eigenvector-Based Spatial Variables and Spatial Modelling
339
AICc of the best model, obtained with a threshold of 2 m, is À99.85 with a model
consisting of 4 MEM variables. Let us see if we can improve this result by weighting
the connections by an inverse distance function.
# 3b. Variant: same as above, but connections weighted by the
#
complement of the power of the distances, 1-(d/dmax)^y.
#
Again, after transformation of the distances by function f2,
#
values near 1 are attributed to pairs of sites with easy
#
exchange, values near 0 mean difficult communication.
mite.thresh.f2
MEM.autocor = "positive",
f = f2,
y = 2:10,
dmax = max(unlist(nbdists(x, as.matrix(mite.xy)))),
xy = as.matrix(mite.xy)))
# Lowest AIC, best model
mite.f2.minAIC
# Smallest AICc (best model among the 10)
min(mite.f2.minAIC)
# Number of the model among the 10
(nb.bestmod <- which.min(mite.f2.minAIC))
# Actual dmax of best model
(dmax.best <- mite.thresh.f2[nb.bestmod][[1]]$all[1, 2])
Hints The actual d max value found by the function is often smaller than the d max provided
to the function by means of the vector of user-selected threshold distances,
because the output of the function shows the largest actual distance within the
limit provided by each threshold value. In the example, the 6th value in vector
thresh10, which contains the list of user-selected threshold distances, is
2.671639. There is no such distance in the mite geographic distance matrix; the
function found that the largest distance smaller than or equal to that threshold is
2.668333.
In some applications, the computations are followed by one or more warnings:
1: In nb2listw(nb, style = "B", glist = lapply(nbdist, f),
zero.policy = TRUE) :
zero sum general weights
This means that one or more points have no neighbours under the definition of
neighbourhood provided to the analysis. This may or may not be important
depending on the context of the analysis.
7.4 Eigenvector-Based Spatial Variables and Spatial Modelling
339