[R-sig-Geo] Question spdep package - Moran's I

[Roger Bivand]

On Mon, 17 Aug 2009, Haenlein, Michael wrote:

> Dear all,
>
> I have a question regarding the spdep package:
>
> Assume that each person within my dataset is characterized by a 
> continuous variable y. In order to test for spatial autocorrelation in y 
> I calculated a Moran's I statistic. The problem is that y is also likely 
> to be influenced by a series of other variables x that are unique to 
> each person in my dataset. To test to which extent the autocorrelation 
> effect is present even when accounting for the effect of x, I first did 
> a regression y = f(x) + Epsilon and then used the regression residuals 
> Epsilon to determine a Moran's I statistic using the lm.morantest 
> function. One of the reviewers of my paper now states that this two step 
> approach (first running a regression and then using the residuals to 
> determine Moran's I) could be inefficient. S/he asks me to correct for 
> the potential impact of x on y in one step when calculating Moran's I.

If the referee is refering to the social networks literature, you will 
have to find out what "one step" means there, if anything. It is not a 
term used in connection with Moran's I for spatial data.

>
> Would anyone happen to know any other way of doing what I'd like to do 
> that only requires one step?

If you believe that the referee's "one step" might be met by fitting a 
model including both the x variables and the dependence in either y or the 
residuals of the regression, look at lagsarlm() and errorsarlm(). Maybe 
check against the Ward/Gleditsch "Spatial regression models" Sage volume 
if need be.

>
> Or could you please provide me with a source I could quote that states 
> that this two-step approach is fine from a statistical perspective?
>

When working across disciplines, nothing constitutes absolute authority in 
this way. Testing residuals ought to be OK, but see Schabenberger & Gotway 
for caveats (especially about concluding that autocorrelation is present 
when the real problem is model misspecification).


Hope this helps,

Roger

> Thanks very much for your help in advance,
>
>
>
> Regards,
>
>
>
> Michael
>
>
>
>
>
> Michael Haenlein
> Professor of Marketing
> ESCP Europe - The School of Management for Europe
> 79, Avenue de la R??publique??|??75011 Paris??| France
>
> 	[[alternative HTML version deleted]]

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no