[R-sig-Geo] Anselin Local Moran's I with R

[Roger Bivand]

On Tue, 9 Sep 2014, David Romero wrote:

> Hello Roger,
>
> Thank you for aclarations. Obviously, climatic data are dependent of
> elevation. I wished to calculate local Moran as an exploratory method
> to identifie outliers before computing semivariograms. Could you
> suggest a better method?

Why would local Moran's I indicate outliers - do you mean very large L/H 
or H/L values, in which neighbouring observations are very dissimilar? I 
assume your data are from weather stations, so point support, rather than 
remotely sensed. Shouldn't a map of residuals from fitting a mean model 
(including relevant covariates and/or trend) with carefully chosen class 
intervals be sufficient? If you need contrasts between proximate 
neighbours, maybe just plot the variable of interest against its spatial 
lag, and choose those stations with big differences.

Hope this helps,

Roger

>
> Thank you,
>
> David
>
> 2014-09-09 8:24 GMT-05:00 Roger Bivand <Roger.Bivand at nhh.no>:
>> On Tue, 9 Sep 2014, David Romero wrote:
>>
>>> Hello,
>>>
>>> Could somebody help me with the method to compute the Cluster and Outlier
>>> Analysis using spdep and obtain Local I Index, Z-scores and cluster type
>>> like with the Arcgis tool.
>>
>>
>> Preferably not. The Arcgis tool answers lots of the wrong questions wrongly.
>> If you take multiple comparisons seriously - see ?p.adjust - and treat
>> permutations of all values except i to permutation bootstrap with reserve
>> (use localmoran.sad() or localmoran.exact() instead), you'll realise that
>> the z-scores are deceptive, and the "cluster" types (HH/LL, LH/HL)
>> wrongheaded if you colour only "significant" ones - they are only
>> significant in most cases before correcting for multiple comparisons.
>> Further, if the mean model is mis-specified (yi - \bar{y} uses \bar{y} alone
>> as the mean model), there may well not be any autocorrelation anyway, just
>> omitted spatially patterned covariates.
>>
>>> My data are points with temperature value.
>>
>>
>> In your case temperature should surely be related to elevation, and very
>> likely to urban density before trying to measure local residual
>> autocorrelation. Just because they do this in Arcgis, it doesn't mean that
>> it is an appropriate procedure.
>>
>> Hope this clarifies,
>>
>> Roger
>>
>>>
>>> Thank you,
>>>
>>> David Romero
>>> phD student
>>> Instituto de Geografía
>>> UNAM
>>>
>>>         [[alternative HTML version deleted]]
>>
>>
>> PS. Please only post plain text, HTML is bulkier and can carry harmful
>> payloads.
>>
>>
>>
>>>
>>> _______________________________________________
>>> R-sig-Geo mailing list
>>> R-sig-Geo at r-project.org
>>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>
>>
>> --
>> 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.Bivand at nhh.no
>

-- 
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.Bivand at nhh.no