[R-sig-Geo] rGeo vs. gstat; Question on geostatistical method for specific experimental design

[Nicholas Lewin-Koh]

Hi Christian,
I think in this particular case geostatistics is not entirely
appropriate. Given your design you have a generalized
linear mixed model (glmm) where you have plants nested in plots, plots
nested in blocks, and your flowerstrip effect 
at the block level. Then you have your field effect which in statistical
parlance would be a blocking factor. Since
your data is binary on each plant, you have a binomial error structure.

On the other hand, what might be interesting is you have 16 replicated
grids of 96 plants. Aside from the above design
you may want to look at some spatial aggregation measure of parasitism
events and compare those across plots.

Cheers
Nicholas


Date: Wed, 19 Oct 2005 09:13:46 +0200
From: "Schlatter Christian" <christian.schlatter at fibl.org>
Subject: Re: [R-sig-Geo] rGeo vs. gstat;        Question on
geostatistical
        method for specific experimental design
To: "Edzer J. Pebesma" <e.pebesma at geo.uu.nl>
Cc: r-sig-geo at stat.math.ethz.ch
Message-ID: <F07A3234E1CDF74399E254240638883001E80DAE at bernina.fibl.ch>
Content-Type: text/plain; charset="iso-8859-1"

Dear Edzer, dear list members

Thank you very much for your comments. It made me investigate quite
well.

The answer to the question "rGeo vs. gstat" is answered by your very
helpful article of the DSC 2003 meeting in Vienna (chapter
"introduction"):
http://www.ci.tuwien.ac.at/Conferences/DSC-2003/Proceedings/Pebesma.pdf

As I'm new and not yet so familiar with the customs of [R-sig-Geo], I
did not feel about asking about my personal statistical problems but
more about general statistical questions.

But as you asked for more detail I will gladly describe them (please let
me know if this should not be the place for it):

We are looking at parasitism rates in cabbage pests eggs (Lepidoptera:
mainly Mamestra brassicae, Pieris rapae) in relation to distance effects
from flowering strips. Specifically the following question: In what
relation is the parasitism rate to the distance from the flower source
(which in the case are sown flower strips).

Many practical restrictions led to a somehow "compromisical"
experimental design consisting of four blocks in one cabbage field (two
with flower strips, two without). Each block has two grids of 6x 8
plants (distance in between each plant: 3m) on each side of the flower
strip (cp. Image), 96 plants per block. On each plant we collected pest
eggs to determine parasitism rate. (Addtionaly we sampled on 4 different
days).

The main problem is the low parasitism rate in the field (only about
15-20% of all eggs have been parasitized), consequently many 0 values.

My idea was to calculate for each of the four blocks variogram
parameters and to compare them afterwards (as we had two study sites, we
have 8 blocks all in all, making 4 with flowers and 4 without). With so
many "0" samples difficult to manage.

Now I found in Edzers article and the gstat manual the possibility of
block kriging. Without knowing exactly what it is, I suppose there is
the possibility to keep all four blocks together and defining the blocks
with the rectangular block-defining possibility.

Is this a valid procedure?

Best wishes

Christian Schlatter (Research institute for organic farming,
Switzerland)

-----Urspr?ngliche Nachricht-----
Von: Edzer J. Pebesma [mailto:e.pebesma at geo.uu.nl]
Gesendet: Montag, 17. Oktober 2005 23:43
An: Schlatter Christian
Cc: r-sig-geo at stat.math.ethz.ch
Betreff: Re: [R-sig-Geo] rGeo vs. gstat

Schlatter Christian wrote:

>Dear list members
>
>I'm very new to R but a little informed about geostatistics.
>
>As I was looking for possibilities of geostatistical analysis in R I encountered at least two very interesting packages:
>
>Rgeo and gstat
>
>And of course I'm wondering now about the one which fits better my needs which are the comparison of spatial data (insect parasitism rates) for different fields in which we collected data.
>
>
>
Christian, I too find these two packages very interesting. Another
package that you might
find interesting besides geoR is geoRglm. However, you've told us too
little about what
your problem is to give you advise about which methods to use, and where
to find them.

You will find even more packages for geostatistical analysis on the
spatial task view:
R home page -> CRAN -> CRAN mirror -> Task Views -> Spatial.

Best regards,
--
Edzer