[R-sig-Geo] R squared in lagsarlm

[Roger Bivand]

On Thu, 21 Mar 2019, Leonardo Matheus Servino wrote:

> When a linear regression is used, usually the degree of freedom, F-value,
> p-value are exposed, in text or in a table. In a lagsarlm, what parameters
> we should expose?
>

None of these make any sense in this case. This model is fitted by maximum 
likelihood, so likelihood-based measures may be appropriate, but the model 
is also non-linear in the spatial coefficient, so it is simply not like 
OLS. However, you could represent OLS in its maximum likelihood form, 
correcting the t-values not to subtract k. The provided measure is a 
likelihood ratio test of the model fitted with and without the spatial 
coefficient (equivalent to a test of the spatial coefficient). You can run 
your own LR tests against other alternatives, and the Nagelkerke measure 
is also likelihood-based. STSLS may provide other measures, but they are 
not OLS-based either, being IV. Just because a supervisor or referee wants 
the same measures as OLS, it doesn't mean they can get them. Certainly you 
should avoid p-values as they give little guidance.

Roger

>
>
> <https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>
> Livre
> de vírus. www.avast.com
> <https://www.avast.com/sig-email?utm_medium=email&utm_source=link&utm_campaign=sig-email&utm_content=webmail>.
> <#DAB4FAD8-2DD7-40BB-A1B8-4E2AA1F9FDF2>
>

-- 
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; e-mail: Roger.Bivand using nhh.no
https://orcid.org/0000-0003-2392-6140
https://scholar.google.no/citations?user=AWeghB0AAAAJ&hl=en