On Tue, 16 Dec 2025, Aymard Loïc KWEKEU KWEKEU wrote:
First, you do not show the versions of software being used, including the
platform, versions of rgdal, sp, sf, spdep, mgcv, R2BayesX, possibly
others. The versions of GDAL used in rgdal and sf may also matter, as GDAL
also changes over time (the precision of coordinates in the same ESRI
Shapefile may differ, artefacts in polygons bay be corrected differently).
Comments inline.
*Initial code with rgdal's readOGR function: *
geo <- readOGR("./Data/Carto/COMMUNE_CARTO.shp")
Without an example file (all the files making up the ESRI shapefile, at
least *.shp, *.shx, *.dbf, preferably with *.prj), there is no way to
establish anything.
geo2 <- subset(geo, geo@data$INSEE_COM %in% selected_com)
Never refer to S4 slots using @ in workflows - geo$INSEE_COM has been
sufficient now for 20 years.
DATA2BND <- sp2bnd(geo2, geo2@data$INSEE_COM)
plot(DATA2BND)
list_names <- as.data.frame(names(DATA2BND)) %>%
group_by(names(DATA2BND)) %>%
summarise(n = n())
sp2bnd is a function in R2BayesX and is strictly superfluous to this
question.
nb <- mgcv:::pol2nb(DATA2BND)$nb
mgcv:::pol2nb is a function in mgcv and is strictly superfluous to this
question.
W.nb <- poly2nb(
geo2,
row.names = rownames(geo2@data$INSEE_COM)
)
w <- nb2mat(W.nb, style = "B")
This is the core of the question - why no warnings or errors?
*New code with sf and sp: *
geo_sf <- st_read("./Data/Carto/COMMUNE_CARTO.shp", quiet = TRUE)
Coercing to an sp-class is superfluous.
geo <- as(geo_sf, "Spatial")
geo2 <- subset(geo, geo@data$INSEE_COM %in% selected_com)
DATA2BND <- sp2bnd(geo2, geo2@data$INSEE_COM)
plot(DATA2BND)
list_names <- as.data.frame(names(DATA2BND)) %>%
group_by(names(DATA2BND)) %>%
summarise(n = n())
nb <- mgcv:::pol2nb(DATA2BND)$nb
It is possible that the coordinate values of the polygons in geo2 differ
here by a very small amount if the versions of GDAL used to install rgdal
and sf differ, and also because sf's implementation for reading geometries
is more modern and cleaner than that in rgdal (>20 years old).
W.nb <- poly2nb(geo2, row.names = rownames(geo2@data$INSEE_COM),
queen=T)
Warning messages:
1: In poly2nb(geo2, row.names = rownames(geo2@data$INSEE_COM), queen =
T):
some observations have no neighbors;
if this seems unexpected, try increasing the snap argument.
Recent changes in spdep warn about observations with no neighbours, and
about subgraphs. Please read the relevant vignette:
https://cran.r-project.org/web/packages/spdep/vignettes/subgraphs.html
No-neighbour observations may be detected when apparently contiguous
polygons are separated by >10mm (default value) if the input polygons are
an sf object, otherwise sqrt(.Machine$double.eps), (so the sp case is
unchanged).
2: In poly2nb(geo2, row.names = rownames(geo2@data$INSEE_COM), queen =
T):
neighbor object has 7 sub-graphs;
if this sub-graph count seems unexpected, try increasing the snap
argument.
w_SMD_2025 <- nb2mat(W.nb, style = "B")
Error in nb2listw(neighbors, glist = glist, style = style, zero.policy =
zero.policy):
Empty neighbor sets found (zero.policy: FALSE)
If you accept a neighbour object with subgraphs (probably no-neighbour
observations), set zero.policy=TRUE in nb2mat. This may however cause
problems later on.
I suggest rather avoiding sp, subsetting geo_sf, and using that object in
poly2nb. You also need to check whether the polygons are planar or
spherical, as this affects the default value of snap.
If you need to share data, do not attach to posts on this list, but rather
raise an issue on https://github.com/r-spatial/spdep/issues/, where the
data and code for a minimal reproducible example may be attached.
Hope this clarifies,
Roger
--
Roger Bivand
Emeritus Professor
Department of Economics, Norwegian School of Economics,
Postboks 3490 Ytre Sandviken, 5045 Bergen, Norway.
e-mail: [email protected]
_______________________________________________
R-sig-Geo mailing list
[email protected]
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
