[R-sig-Geo] Model design

[alfreda morinez]

Hi Robert,

Thanks for this,

So my concern was the high levels of spatial auto-correlation. I made
a semi-variogram of the cell's x and y values and tested for
significant difference from predicted value ( which there was).

So what you are saying is to take the average per AREA maybe in the
form of a ANOVA and test for significant differences between the
AREAS?

However my concern with this approach is that the average is not
realistic as two cells next to each other are not independent from
each other due to their proximity to each other. Hence the approach i
am taking?

Alfreda

On Fri, Dec 16, 2011 at 5:28 PM, Robert J. Hijmans <r.hijmans at gmail.com> wrote:
> Alfreda,
>
> It appears that you are comparing entire populations (all cells in each
> AREA), not samples. If that is the case there is no difference between the
> (true) population mean and the (estimated) sample mean. That makes doing a
> statistical test is irrelevant: all differences should be
> considered statistically significant. Whether these differences matter is
> for you to decide.
>
> Given that we (think we) know how many people live in Germany and France
> would you do a statistical test to say which country has a larger
> population?
>
> Robert
>
> On Fri, Dec 16, 2011 at 5:04 AM, alfreda morinez <alfredamorinez at gmail.com>
> wrote:
>>
>> Dear List,
>>
>> I am realtively inexperienced so i apologise in advance and ask for
>> understanding in the simplicity of my question:
>>
>> I have data on the amount of grass per km in a cell ( of which i have
>> lots) "grass" and for each cell i have x/y coordinates - required due
>> to spatial autocorrelation
>>
>> Cells can be classfied in a hierarchical nature into  AREAS and STATES
>>
>> i.e Cell 1, Cell 2, Cell 3 are all in AREA "A"
>>
>> where as Cell 4,5 and 6 are in AREA "B"
>>
>> However both area A + B are in state "S1"
>>
>> I have lots of these (13000) cells which are classfied into ~2000
>> AREA's and ~750 STATE'S
>>
>> So my question is do AREA'S differ in the amount of grass they contain
>> i.e does AREA A contain significantly more grass than AREA B?
>>
>> I have modelled this by
>>
>> area_grass <- gls(grass~AREA, correlation=corExp(form=~x+y), data =
>> grassland
>>
>> I have set the contrasts to options(contrasts = c("contr.treatment",
>> "contr.poly")) as there are no control groups.
>>
>> What i will get ( it is taking ages!)
>>
>> is
>>
>> AREA A:     -0.12.... **
>> AREA B:      0.17....*
>> AREA C..
>>
>>
>> So can i then say AREA A has significantly less grass than the
>> average, AREA B significantly more and AREA C is not significantly
>> different?
>>
>> Thanks
>>
>> Alfreda
>>
>> P.S - Il will then repate this fot STATE instead of AREA.
>>
>> _______________________________________________
>> R-sig-Geo mailing list
>> R-sig-Geo at r-project.org
>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>
>