Thanks Peter,
summary.permat are plot.permat are very useful, thanks for that.
If I understand correctly from your help page and code, the problem you
describe is primarily with sequential algorithms, where the "null"
matrix obtained from a given run can actually be quite similar to the
one obtai
Dear Etienne,
As Jari Oksanen pointed out, we found that quantitative null models
can be really odd, and I view them as a last resort for doing
community analyses. Their value must be judged by using independent
methods. Just an example that worth mentioning: the "swap" and
"quasiswap" methods in
Etienne,
We used Chi-square statistic to inspect randomness, see the code in
summary.permat.
To test the convergence of sequential algorithms, you can use time
series and MCMC diagnostic tools.
Peter
2010/1/7 Etienne Laliberté :
> Many thanks Jari for your input.
>
> I'll have a look at this
Many thanks Jari for your input.
I'll have a look at this backtracking method and see how I could
implement it.
Ensuring that the null matrices are indeed random is clearly good
advice, and I'd need to do this, but do you have any suggestions on how
to do this in practice? A copy of the script yo
On 07/01/2010 21:48, "Etienne Laliberté" wrote:
> Many thanks again Carsten.
Yes, you're right that care must be taken to
> ensure that a decent number
of unique random matrices must be obtained. I
> don't think it would be a
problem in my case given that transforming my
> continuous abundance
Many thanks again Carsten.
Yes, you're right that care must be taken to ensure that a decent number
of unique random matrices must be obtained. I don't think it would be a
problem in my case given that transforming my continuous abundance data
to count by
mat2<- floor(mat * 100 / min(mat[mat > 0
Hi Etienne,
I'm afraid that swap.web cannot easily accommodate this constraint.
Diego Vazquez has used an alternative approach for this problem, but I
haven't seen code for it (it's briefly described in his Oikos 2005
paper). While swap.web starts with a "too-full" matrix and then
downsamples,
