[Openspace] Non-square weight matrix
koschins at uiuc.edu
Thu Nov 1 16:09:32 CDT 2007
The issue is that GeoDa is designed for spatial AUTOcorrelation, i.e. for correlations between values of X at location i and neighboring locations j. Your setup is about the correlation between y (farms) and x (suppliers) within a given distance band. So you're regressing y on WX, which you can do using regular OLS, while GeoDa is designed to estimate Wy or We, which you can't do with regular OLS.
You can compute WX in a regular GIS, e.g., using buffers and the spatial join function (or scripts with loops for overlapping buffers).
---- Original message ----
>Date: Wed, 17 Oct 2007 21:33:18 -0500
>From: "Kolympiris, Christos (UMC-Student)" <cknq6 at mizzou.edu>
>Subject: [Openspace] Non-square weight matrix
>To: <openspace at sal.uiuc.edu>
>I am wondering how (if possible) i can create a distance based weight matrix that is not square and allows for the diagonal elements to be 0.
>Here is an example. Say that i define as neighbors farms and food suppliers when their distance is less than a threshold level and i then have as the dependent variable some measure of profitability for the farm. I then put the farms as the rows and the food suppliers as the columns. If i have different number of observations for farms and suppliers, then the matrix isn't square and most of the diagonals are 0. If GEODA had the option of specifying 2 sets of coordinates (long and lat for the farm and for the supplier), then i could do it.
>I think that due to matrix algebra multiplication rules both agents need to have the same number of observations but (hopefully) i am missing something. Even if the matrix is square is it even possible to create a weight matrix with 0 diagonals? If so how - in GEODA or outside and then insert it?
>Any help would be very much appreciated.
>Department of Agricultural Economics
>University of Missouri - Columbia
>Openspace mailing list
>Openspace at sal.uiuc.edu
More information about the Openspace