Pseudo R^2 in Spatial simultaneous autoregressive error model estimation by GMM

[Alysson Luiz Stege]

I'm using the package "spdep" to run my spatial regressions. I am using the "GMerrorsar" function. Using the summary, it does not display the value of R2 adjustment. I know the "errorsarlm" function calculates the value of R2, but the summary function "GMerrorsar" do not appears. I wonder if the "GMerrorsar" function calculates the value of R2? Thank

[Roger Bivand]

You asked this question on the R-sig-geo list a day or so ago. I did you
the kindness of checking whether the equivalent Stata spreg variant
reports R^2 - it does not. The ML Nagelkerke R^2 is not comparable with
what you want, a naive OLS coefficient of determination. These siplicities
are just not readily available, and if they were provided, would be
misintpreted. Before rushing to claim that your estimation results explain
anything, compare the OLS and error model beta estimates. If the OLS
estimates appear not to be central in the distribution of the ML/GMM ones,
you have big-time trouble, and R^2, pseudo or otherwise gets you nowhere
fast. If your suporvisor wants R^2, maybe mention that if something isn't
provided, maybe there are reasons for that?

Roger

PS. The GMerrorsar implementation was originally contributed by Luc
Anselin, so I would defer to his better judgement on this.

--
Roger Bivand
Department of Economics, Norwegian School of Economics,
Helleveien 30, N-5045 Bergen, Norway.
voice: +47 55 95 93 55; fax +47 55 95 91 00
e-mail: Roger....@nhh.no