spatial weight matrix: observation without neighbors
415 views
Skip to first unread message
Kate Peng
Jul 21, 2017, 11:33:07 PM7/21/17
to Openspace List
Dear all,
I tried to calculate the spatial weight matrix with 'contiguity' setting in GeoDa.
But given the Japan map with 47 prefectures, there are two prefectures (circled in red; see the photo below) located outside the mainland and thus...no contiguous neighbors. Even I tried to use the 'precision setting', still can't do.
Is there any way to find 'neighbor' for those disconnected regions?
Any feedback or help is deeply appreciated.
Regards,
Kate
Luc Anselin
Jul 21, 2017, 11:38:33 PM7/21/17
to openspa...@googlegroups.com
The main concept to keep in mind is what is a “neighbor”. If these are indeed “islands” then they have
no neighbors. In other words, they are not involved in any interaction with neighbors. If that is desired,
that is fine. If not, then one needs to create artificial neighbors. One way to do this (but not always ideal)
is to use a k nearest neighbor definition of neighbors. This will always guarantee k neighbors for every
observation, but may be a bit too brute force. Another alternative is to edit the GAL file. This is a text file,
so it can be easily edited. You would have to decide which are the proper “neighbors” for the islands and
edit the GAL file accordingly. You enter the the number of neighbors and their IDs for the islands and make
sure to add the former island IDs to the new neighbors to ensure that the neighbor information is symmetric.
There is nothing wrong with true islands or isolates, they are simply not involved in any “spatial interaction”
aspects of the analysis.
L.
> --
> You received this message because you are subscribed to the Google Groups "Openspace List" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to openspace-lis...@googlegroups.com.
> To post to this group, send email to openspa...@googlegroups.com.
> Visit this group at https://groups.google.com/group/openspace-list.
> For more options, visit https://groups.google.com/d/optout.
no neighbors. In other words, they are not involved in any interaction with neighbors. If that is desired,
that is fine. If not, then one needs to create artificial neighbors. One way to do this (but not always ideal)
is to use a k nearest neighbor definition of neighbors. This will always guarantee k neighbors for every
observation, but may be a bit too brute force. Another alternative is to edit the GAL file. This is a text file,
so it can be easily edited. You would have to decide which are the proper “neighbors” for the islands and
edit the GAL file accordingly. You enter the the number of neighbors and their IDs for the islands and make
sure to add the former island IDs to the new neighbors to ensure that the neighbor information is symmetric.
There is nothing wrong with true islands or isolates, they are simply not involved in any “spatial interaction”
aspects of the analysis.
L.
> You received this message because you are subscribed to the Google Groups "Openspace List" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to openspace-lis...@googlegroups.com.
> To post to this group, send email to openspa...@googlegroups.com.
> Visit this group at https://groups.google.com/group/openspace-list.
> For more options, visit https://groups.google.com/d/optout.
Kate Peng
Jul 23, 2017, 9:32:05 PM7/23/17
to Openspace List
Dear L.,
Thank you for so much reply.
anselin於 2017年7月22日星期六 UTC+8上午11時38分33秒寫道：
Your explanation solves the problem. I didn't realize we can 'manually' create neighbours for observation in .gal file -- which is quite useful to take into account those island counties.
Sincerely yours,
Kate
anselin於 2017年7月22日星期六 UTC+8上午11時38分33秒寫道：
Reply all
Reply to author
Forward
0 new messages