warning message in GeoDaSpace

[Kate Peng]

Dear all,

I created spatial weight (Queen Contiguity) in Geoda, and it worked okay in spatial lag and spatial error model in Geoda. I also checked to gal file, in which every observation has at least one neighbour.

Then I imported this spatial weight into GeodaSpace because I wanted to run SARAR model (i.e. spatial lag+error). But for some reason, when I ran SLR and SARAR using this spatial weight in GeodaSpace, I got the following error message:

"power expansion will not coverage, check model specification and that weight are less than 1."

Could anyone tell me what I should do to fix this? 

Your help is much appreciated. Thank you.

Sincerely yours,

Kate

[Julia Koschinsky]

Hi Kate,

Try loading the spatial file into GeoDaSpace and then create the queen weights directly in GeoDaSpace. 

Sometimes there's problems transferring the GeoDa weights to GeoDaSpace.

Best,

Julia

Julia Koschinsky, Ph.D.

Executive Director and Senior Research Associate

Center for Spatial Data Science

University of Chicago

https://spatial.uchicago.edu/

https://puttingscienceintodatascience.org/

---

From: openspa...@googlegroups.com <openspa...@googlegroups.com> on behalf of Kate Peng <katep...@gmail.com>
Sent: Sunday, April 23, 2023 4:28 AM
To: openspa...@googlegroups.com <openspa...@googlegroups.com>
Subject: [Openspace] warning message in GeoDaSpace

 

--
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 view this discussion on the web visit https://groups.google.com/d/msgid/openspace-list/CAJfEm7GjLoNDA9MhKTRF%2BDn9C5ripdH-axXMdevmeg9N7w8RFw%40mail.gmail.com.

[Luc Anselin]

Actually, I suspect the problem is with the SARAR specification. That is really ill-behaved and

can lead to identification problems. Often the lag coefficient and error coefficients have opposite

signs (which doesn’t really make a lot of sense) and the estimates are very large in absolute value.

This can lead to problems with the GMM/IV estimators when the instruments are poor, e.g., when only

WX is used as an instrument. Unlike ML estimation, GMM/IV cannot guarantee that the estimate for the

spatial coefficient is in the proper parameter space, i.e., less than 1 in absolute value. This will make the

power expansion fail. There isn’t much one can do about this except trying higher order spatial lags of X

as instruments. But it generally points to a problem with SARAR. I personally avoid that specification unless

there is a very good (theoretical) reason.

L.

To view this discussion on the web visit https://groups.google.com/d/msgid/openspace-list/PH0PR11MB49983F1BEE6771211884C282CD669%40PH0PR11MB4998.namprd11.prod.outlook.com.

[Vania Ceccato]

Luc

So, do you suggest the use of lag and error separately?

Best wishes from Stockholm

Vania 

—-•—•—•—•—•—•—

Prof Vania Ceccato

Skolan för arkitektur & samhällsbyggnad

Kungliga Tekniska Högskolan

Teknikringen 10A, 10044 Stockholm, Sweden, Mobil: +46(0)73-6649070

https://www.kth.se/profile/vace/

On 23 Apr 2023, at 19:12, Luc Anselin <lans...@gmail.com> wrote:

[Roger Bivand]

Vania,

As Luc says, the SARAR specification is really ill-behaved, also with ML, and the coefficient estimates seem to lie on a banana ridge from strong negative on one coefficient to strong positive on the other (and in ML the optimisation is properly bounded). A different start point can end up in a different quadrant. Maybe GNM is less problematic, but suffers from the same issues; SDM or SDEM are kinder on the numerics, because the spatial coefficient estimates are not getting in each other's way.

Roger

--
Roger Bivand
Emeritus Professor
Norwegian School of Economics
Postboks 3490 Ytre Sandviken, 5045 Bergen, Norway
Roger....@nhh.no

________________________________________
From: openspa...@googlegroups.com <openspa...@googlegroups.com> on behalf of Vania Ceccato <vania....@abe.kth.se>
Sent: 23 April 2023 20:46
To: Luc Anselin
Cc: openspa...@googlegroups.com; katep...@gmail.com
Subject: Re: [Openspace] warning message in GeoDaSpace

--
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<mailto:openspace-lis...@googlegroups.com>.
To view this discussion on the web visit https://groups.google.com/d/msgid/openspace-list/424089CE-E167-4984-B94E-D1C415DEEBE9%40abe.kth.se<https://groups.google.com/d/msgid/openspace-list/424089CE-E167-4984-B94E-D1C415DEEBE9%40abe.kth.se?utm_medium=email&utm_source=footer>.

[Vania Ceccato]

Many thanks Luc and Roger. I will keep that in mind. I  am having a go with spaceGeoDa. 

Best

Vania 

—-•—•—•—•—•—•—

Prof Vania Ceccato

Skolan för arkitektur & samhällsbyggnad

Kungliga Tekniska Högskolan

Teknikringen 10A, 10044 Stockholm, Sweden, Mobil: +46(0)73-6649070

https://www.kth.se/profile/vace/

On 23 Apr 2023, at 21:25, Roger Bivand <Roger....@nhh.no> wrote:

[Kate Peng]

Dear Luc,

Thank you for your clear explanation. It now makes theoretical sense to me.

To be honest, I chose to run SARAR simply because I saw several papers using these three models (i.e. spatial lag, spatial error and SARAR). As you suggested, the theoretical reason why a specific model is chosen should be clarified. I will think carefully about this. Thank you.

I tried to use GMM and ML estimation to run SARAR but still had the same warning message. I think the possible reason is that I only used WX as an instrument in SARAR, which could be a poor instrument, as you pointed out.

I will try to use higher order spatial lags of X as instruments; more importantly, I will think carefully why SARAR is theoretically necessary to my model.

Thank you for your advice. 

Sincerely yours,

Kate

[Kate Peng]

Dear Julia,

Thank you for this helpful suggestion.

I will give it a try.

Sincerely yours,

Kate

[Kumar, Indraneel]

Good afternoon, everyone,

Thank you, Luc and Roger for the insights! The SARAR discussions are very useful. Is there a reference article related to this?

Thanks a lot, and best wishes,

Indraneel

Indraneel Kumar, PhD, AICP
Principal Regional Planner
Purdue Center for Regional Development
o: 765-494-9485
iku...@purdue.edu

-----Original Message-----
From: openspa...@googlegroups.com <openspa...@googlegroups.com> On Behalf Of Roger Bivand
Sent: Sunday, April 23, 2023 3:26 PM
To: Luc Anselin <lans...@gmail.com>; openspa...@googlegroups.com
Cc: katep...@gmail.com
Subject: Re: [Openspace] warning message in GeoDaSpace

---- External Email: Use caution with attachments, links, or sharing data ----

To unsubscribe from this group and stop receiving emails from it, send an email to openspace-lis...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/openspace-list/SV0P279MB0475366DE65A853E683E8D89EE669%40SV0P279MB0475.NORP279.PROD.OUTLOOK.COM.