Error message when using XSMLE-Command

[Johannes Blum]

Dear Stata-Community,

I want to estimate a strongly balanced spatial panel data model using the XSMLE command with a weighting matrix which even has no islands. Unfortunately, the output stops with the following error:

Warning: e(V) matrix is not positive definite.
Spatial effects Std. Err. will be computed using a modified
positive definite matrix (Rebonato and Jackel, 2000).
estimates post: matrix has missing values
r(504);

It seems that the not positive definite matrix causes the problem. In the blog of Federico Belotti, who invented the xsmle command, he recommends using the option -nose-. However, introducing this option doesn't solve the problem either.

Does anyone have some advice for me?

I would really appreciate that and thank you very much in advance!

Best,
Johannes

[Nick Cain]

Dear Johannes,

I am using the XSMLE command for analysis as well, however my model operates smoothly. How are you generating the spatial weights matrix? What type of weight-matrix are you using? Do you have the spwmatrix command installed?

-nc

[Federico Belotti]

Dear Johannes

usually the error you are getting is due to the failure of the procedure used to compute standard errors of direct, indirect and total effects. In most cases the -nose- (undocumented) option should solve the issue. Another possible solution is to exclude their computation using the -noeffects- option.

Hope this helps
Best

[Johannes Blum]

Dear Nick,
I could deal with the problem in the meanwhile which was related to a wrong variable I used for clustering the standard errors. No idea, why STATA gave me this error message.
Thank you very much and best regards!
Johannes

[Johannes Blum]

That helped a lot! Thank you very much, Federico!

[Martin Hulenyi]

Johannes, i had the same issue with using -robust- option. Everything worked fine after removing it.