It may help if you post what command you are using for the SEM, as well as what command you are using for the spatial weights matrix. Are you estimating the spatial weights matrix based on the coordinates for one period, or for all periods?

On his webpage, Solomon Hsiang proposes a file.ado implementing Conley (2009) correction for spatial auto-correlation. It might help you.

We typically use bigW = identity(T) kronecker W, where T is the number of time periods and W is of dimension n by n, where n is the number of spatial units. This creates a block diagonal bigW matrix, with a W-matrix for each time period on the main diagonal blocks. I don't know how to do this in Stata. In MATLAB it is: bigW = kron(eye(T),W);

In stata I use spmat but I always get the following error message: two or more observations have the same coordinates with is right as I'm using a panel dataset.

What I can do in order to create my spatial weight matrix for panel data?

Thanks a lot for your help

Try to create the weight matrix from the cross section , then use that in panel data. I had similar issue, and solved it following this strategy. If you use in panel data, then same coordinate is used several times (if T=20, then 20 times), and so stata is giving this error. If you use centeroid then coordinates should be one (in fact two : lon and lat). if you are still working it and having issue please let me know. Thanks

Thank you Mr Shafiun I'll try it and let you konw if it works with my or not. I'll do, once again thank you for your help.

Boukhicha, I would be interested to know what strategy you followed to implement Lesage's MATLAB suggestion [bigW = kron(eye(T),W)] in Stata (or in R, if anyone knows an alternative)

Hell Victor: You may find it useful http://www.stata.com/meeting/italy13/abstracts/materials/it13_belotti_hughes_mortari.pdf; and the package "SPCS2XT" does similar to what Prof. LeSage suggested. Please let me know whether it works for you. Thanks 

 Hi Shafiun, thanks a lot for your quick reply. "SPCS2XT" does seem to be what I was looking for. I'm afraid I will need a cmd other than xsmle to estimate models with binary dependent variables, i.e., spatial probit, though. Any idea which stata or r command could do that?

I searched for spatial probit estimation with stata with no luck, However, I know there is R package which can do it (you have to do Bayesian, probably). Here is the R package that can handle probit model not sure whether it can do for panel, however, you should be able to do pooled reg https://cran.r-project.org/web/packages/spatialprobit/spatialprobit.pdf

This is a good ref https://journal.r-project.org/archive/2013-1/wilhelm-matos.pdf

Hey Folks, I have a similar problem with duplicate coordinates. In my case, multiple entities in a cross section are measured at the same spatial location. Is there any way to deal with this? I am also using spmat in stata.

Thanks.

 Hello, 

I think the problem is to calculate the spatial weight matrix we may need to use the coordinates of the centroids of each county (or state, for example). So if for each panel there should be one centroid (in other words latitude and longitude of the center). See this (http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.725.1656&rep=rep1&type=pdf  page 253).