Dear Tom,
I think your idea, using distances instead of ordination scores is
useful. I prefer remaining as close to raw data as possible; and using
distances instead of ordination scores follows this rule.
However, your solution in the original form works only when there is
only one reference site. If there are multiple reference sites, we
should define distance of each regenerating site to the set of reference
sites. It can be the mean or minimum of pairwise distances. I think both
(i.e. min and mean) may be meaningful. If the set of reference sites is
heterogeneous, I would choose minimum, while otherwise mean.
Zoltan
2016.01.18. 6:58 keltezéssel, Philippi, Tom írta:
Conny--
Note that Jari's surface fitting is using ordination scores on the
right-hand predictor size of the formula, with some z as the response.
If you need something about species composition as your _response_ variable
in a linear model (e.g., with time, disturbance type, and treatment as
predictors, and perhaps site as a random effect), why not use each stand's
dissimilarity/distance from your reference forest sites? The trend line
would be compositional distance or dissim v. time, with
color/symbols/whatever for different treatments. That would have the
advantage of being easily & directly interpretable. [The use-case where
that would fail is >>100% turnover so lots of 0 similarities to the
reference forests, so step-across or nmds might help put those large
distances in order.] You might be able to set up the equivalent to your
GLM in adonis to get permutation significance tests.
I hope that this helps, or at least gives you a different way to think
about your problem, or else is so stupid that Jari gets annoyed and blasts
it with a valid solution.
Tom 2
------
Tom Philippi
Quantitative Ecologist & Data Therapist
National Park Service
On Sun, Jan 17, 2016 at 8:04 PM, Conny <constanze_k...@hotmail.com> wrote:
Thanks a lot for all the helpful responses and info.
But I’m actually still not sure how to use both NMDS axes as a response
(y) in a regression model - is this even possible??
My overall goal is to model species compositional change over time in a
restoration project (is the system getting more similar to the reference
forest). I would like to create a trend line here in a graph, rather than
just using an ordination plot.
I thought about using the fitted values returned by ordisurf(), but as I
understood it (please correct me if I’m wrong) it will use my restoration
time again as a response and my axes scores as predictors.
So the z values will represent fitted age values rather than my sample
scores (?) – so it would make no sense to plot it against my restoration
time…
I’m sorry if this is getting a bit confusing.
Cheers,
Conny
_______________________________________________
R-sig-ecology mailing list
R-sig-ecology@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
[[alternative HTML version deleted]]
_______________________________________________
R-sig-ecology mailing list
R-sig-ecology@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
--
Botta-Dukát Zoltán
--------------------------------
Ökológiai és Botanikai Intézet
Magyar Tudományos Akadémia
Ökológiai Kutatóközpont
--------------------------------
2163. Vácrátót, Alkotmány u. 2-4.
tel: +36 28 360122/157
fax: +36 28 360110
botta-dukat.zol...@okologia.mta.hu
www.okologia.mta.hu
Zoltán BOTTA-Dukát
--------------------------------
Institute of Ecology and Botany
Hungarian Academy of Sciences
Centre for Ecological Research
--------------------------------
H-2163 Vácrátót, Alkomány u. 2-4.
HUNGARY
Phone: +36 28 360122/157
Fax..: +36 28 360110
botta-dukat.zol...@okologia.mta.hu
www.okologia.mta.hu
_______________________________________________
R-sig-ecology mailing list
R-sig-ecology@r-project.org
https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
