Jackknife Validation and P-Value Compute Software

[Michael Anderson]

I'm attempting to model Lepilemur spp. potential distributions
throughout Madagascar. My occurrence data is somewhat limited, so I
have decided to use the Jackknife validation (n-1) method that was
presented in the Pearson et al 2007 paper.

I'm having some difficulties regarding the use of the authors'
supplementary P-value compute software. I was hoping someone here may
be familiar with this, or has gone through similar issues.

The p-value compute software asked for a set of 0's and 1's (test
omission rate) that reflect each models success regarding the
prediction of the occurrence point that was left out and a
corresponding p-value which is equivalent to the Fractional predicted
area and p-value that are presented in the html file that is generated
by maxent.

When these values are run through the P-value compute software they
spit out two values. Does anyone know what these two values
represent. I was expecting it to give one p-value, and there are
two.

Example from the help.txt:

If input.txt contains the following data:

Species #1
1.0000 0.2955
0 0.0153
0 0.0213
1.0000 0.0331
0 0.0328
1.0000 0.0238
Species #2
1.0000 0.0383
1.0000 0.0288
0 0.0133
0 0.0281
1.0000 0.0252
0 0.0227
Species #3
0 0.0379
1.0000 0.0666
1.0000 0.0751
1.0000 0.0751
1.0000 0.0819
1.0000 0.0718
0 0.0612
0 0.0670
0 0.0599
1.0000 0.0666
1.0000 0.0604

Running the program with input.txt as the input file will write the
following output to the output file:

Species_#1 0.5 0.001564
Species_#2 0.5 0.000289
Species_#3 0.636364 0.000001

Any advice would be greatly appreciated.

Cheers,

Mike

[Lukas]

Hi Mike,

you have to open the maxent.csv file and go to the column "Minimum
training presence test omission" for the numbers of the first row. But
you have to change the 0 to 1 and the 1 to 0 because Maxent is showing
the test omission and the p-value software using the errors. So just
the other way around.

Then use the values of the column "Minimum training presence area" for
the predicted area.

This works well for me.

Regards,

Lukas

[Newton Pimentel de Ulhôa Barbosa]

Hi people and Lukas,

Sorry for the question, but I'm trying to understand this for now and
this is going difficult!

So, I have only two *.csv files: "speciename_omission.csv" and
"speciename_samplePredictions.csv". And I don't have the column
"Minimum training presence test omission" in any one. So, where I can
find the values to run in pValue?

So, If I want to run a model with few points, I need to ignore the AUC/
ROC analysis? And after run the model I need to run the pValue? And
how? I'm confuse with this steps. If someone could help me, will be
amazing!

Thank you!

Newton P. U. Barbosa
Ph.D. candidate
University of Alberta, Canada / Universidade Federal de Minas Gerais,
Brasil

[Lukas]

Hi,

there should be a column with this title. And yes, as far as I know
from different papers, AUC/ROC shouldn't be used with a small nummer
of samples.

Maybe you can send me the .csv files and i'll have a look if I can
help you somehow with it.

Regards,

Lukas

On 16 Sep., 22:19, Newton Pimentel de Ulhôa Barbosa