Questions about percent contribution

[ROCHA, OSCAR]

Hello all,

 

This is a naïve question from a new user of MaxEnt. I am comparing the potential range of a group of species of plants in the genus Ocotea (Lauraceae). I found differences in the “size” of their suitable niche across species. These differences are reflected in the range on the environmental variables used in the model (minimum and maximum values for eight climate variables), where as expected species that are more widely distributed have wider range for these variable. I also found that widely distributed species have more climatic variable with high percent contribution, while for species with restricted distribution only two or three variables have values that exceed 20-30% and thus add to over 80% of the contribution.

 

My questions are the following:  Is this an appropriate use and interpretation of the percent contribution? Is it always expected that widely distributed species would show wider range for the environmental variables used in the model? Is it correct to justify limited distribution on the basis of a very strong effect of one or a few variables?

 

Thank you for your thoughts

 

Oscar Rocha

 

[desalegn chala]

Hi again,

Here are the sheloford's law of tolerance if they are really helpful to you:

 

§Shelford’s Law of Tolerance: Factors could be limiting both at their maximum and their minimum.

I.Organisms have a wide range of tolerance for one factor and a narrow range for the other

II. Organisms with wide range of tolerance for all factors are the most widely distributed

III. Tolerance limit and range of optima may shift, displace or extended

Example: when there is nitrogen deficiency, the tolerance of grasses to drought will less

IV.The limit of tolerance for reproductive individuals, seeds, eggs, embryos and larvae are narrower than the noon reproductive ones

 For the percentage contribution, I had similar result when I was doing my thesis. I am not sure about the exact answer but it was not challanged on the defense at least. You can check page 22 of the following paper where you see a similar scenario: 
http://www.itc.nl/library/papers_2010/msc/nrm/desalegn.pdf

---

From: "ROCHA, OSCAR" <oro...@kent.edu>
To: "max...@googlegroups.com" <max...@googlegroups.com>
Sent: Wed, April 13, 2011 10:42:46 PM
Subject: Questions about percent contribution

--
You received this message because you are subscribed to the Google Groups "Maxent" group.
To post to this group, send email to max...@googlegroups.com.
To unsubscribe from this group, send email to maxent+un...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/maxent?hl=en.

[Steve Collins]

Apologies for piggy-backing on Oscar's question, but I had a question about the percent contribution and permutation importance output.  The percent contribution depends on the order that Maxent constructed the model, so that variables that contribute little additional training gain show a low percent contribution.  Variables with low percent contribution can still have high permutation importance, however.  As the tutorial warns, correlated variables will skew the percent contributions considerably.  This suggests that percent contribution can't be directly interpreted as which variables are the most important, which I think is how many people have interpreted these values.

So apart from the jackknife plots, how does one really know how much the Maxent model is using each variable?  The idea of variable weights may not be an appropriate concept in Maxent.  Please help me out.

Thanks,
Steve Collins
Graduate Student
Department of Biological Sciences
Texas Tech University
Lubbock, TX 79409-3131
steve....@ttu.edu

[Lucia]

I have the same question as Steve Collins. How to identify which one
is the most factor for a species?
Thanks a lot.

On Apr 13, 5:42 pm, Steve Collins <dcoll...@ufl.edu> wrote:
> Apologies for piggy-backing on Oscar's question, but I had a question
> about the percent contribution and permutation importance output.  The
> percent contribution depends on the order that Maxent constructed the
> model, so that variables that contribute little additional training gain
> show a low percent contribution.  Variables with low percent
> contribution can still have high permutation importance, however.  As
> the tutorial warns, correlated variables will skew the percent
> contributions considerably.  This suggests that percent contribution
> can't be directly interpreted as which variables are the most important,
> which I think is how many people have interpreted these values.
>
> So apart from the jackknife plots, how does one really know how much the
> Maxent model is using each variable?  The idea of variable weights may
> not be an appropriate concept in Maxent.  Please help me out.
>
> Thanks,
> Steve Collins
> Graduate Student
> Department of Biological Sciences
> Texas Tech University
> Lubbock, TX 79409-3131

> steve.coll...@ttu.edu

>
> On 4/13/2011 3:42 PM, ROCHA, OSCAR wrote:
>
>
>
>
>
> > Hello all,
>

> > This is a na�ve question from a new user of MaxEnt. I am comparing the

> > potential range of a group of species of plants in the genus Ocotea
> > (Lauraceae). I found differences in the "size" of their suitable niche
> > across species. These differences are reflected in the range on the
> > environmental variables used in the model (minimum and maximum values
> > for eight climate variables), where as expected species that are more
> > widely distributed have wider range for these variable. I also found
> > that widely distributed species have more climatic variable with high
> > percent contribution, while for species with restricted distribution
> > only two or three variables have values that exceed 20-30% and thus
> > add to over 80% of the contribution.
>
> > My questions are the following:  Is this an appropriate use and
> > interpretation of the percent contribution? Is it always expected that
> > widely distributed species would show wider range for the
> > environmental variables used in the model? Is it correct to justify
> > limited distribution on the basis of a very strong effect of one or a
> > few variables?
>
> > Thank you for your thoughts
>

> > Oscar Rocha- Hide quoted text -
>
> - Show quoted text -