nb object in spdep from categorical attribute

On Tue, 19 Jun 2018, Dexter Locke wrote:

Dear list,

Does anyone know how to make nb objects (to eventually make spatial
weights) with spdep using an attribute in a shapefile's data frame?

Consider, for example, all property parcels on the same street segment
should be defined as neighbors. In the parcel polygons I have the
associated street segment name.

KNN forces sparsely settled areas to become neighbors, which is

undesirable

for the subsequent analyses.

Distances between parcels vary, so distance-based neighbors are also
undesirable in this case.

Contiguity-based methods do not connect parcels on opposite sites of the
same street segment.

Is there a way to use an attribute to define what constitutes a neighbor?

Yes, spdep::nb2blocknb() is a specific case where say observations without
positional data are known to belong to an aggregate entity for which we
have positional data. The output is an nb object with diagonal blocks (if
the observations are sorted by block ID, not in the case below), as this
example shows. You'll need the development version to run it, as released
spdep::aggregate.nb() failed with empty aggregate sets - the nc median
poish grid has such empty sets:

# devtools::install.github("r-spatial/spdep")
library(spdep)
data(nc.sids)
ncCC89.nb[sapply(ncCC89.nb, length) == 0L] <- 0L
image(as(nb2listw(ncCC89.nb, zero.policy=TRUE), "CsparseMatrix"))
anb <- aggregate(ncCC89.nb, interaction(nc.sids$M.id, nc.sids$L.id))
image(as(nb2listw(anb, style="B"), "CsparseMatrix"))
# when the neighbour object is not empty
bnb <- nb2blocknb(anb, as.character(interaction(nc.sids$M.id,
   nc.sids$L.id)))
image(as(nb2listw(bnb, style="B"), "CsparseMatrix"))
# and when the neighbour object is empty
anb1 <- lapply(anb, function(x) x <- 0L)
attributes(anb1) <- attributes(anb)
bnb1 <- nb2blocknb(anb1, as.character(interaction(nc.sids$M.id,
   nc.sids$L.id)))
image(as(nb2listw(bnb1, style="B"), "CsparseMatrix"))

This isn't your case, but crafting an nb object by hand (adding
attributes after creating the list) then using spdep::union.nb or similar
to combine with a positional nb (such as poly2nb) may work. For parcels
across a stream or street, maybe look at the snap= argument if the streets
are narrower than frontages. Could you provide a small reproducible
example?

Hope this clarifies,

Roger

All ideas welcome,
Dexter

      [[alternative HTML version deleted]]

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; e-mail: Roger.Bivand at nhh.no
http://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en