Skip to content

Standard Error vs Standard Deviation of AUC in Maxent/Boosted Regression Trees (package 'dismo')

1 message · Johannes Radinger

Original
Johannes Radinger
# 
Hi,

I am running two different Species Distribution Models (SDMs):
1) MaxEnt and 2) Boosted regression trees (gbm.step) in R via
the package 'dismo' with a 10-fold cross validation. For both models provide
an value for AUC/ROC either in textfile (maxent) or in the resulting
object (gbm.step). Addtionally the two models give also a parameter of the variation of the AUC, but unfortunately two different ones.
Maxent provides in its text file the standard deviation (SD) of the AUC while
gbm.step provides the standard error (SE) of the AUC.
As I want to get a comparable parameter I only want to use either SE or SD.

So my first question: What is the commonly used statistical parameter?
And secondly: How is that calculated for the model where it is missing? Which  parts of the results are the basis for the calculation so that I get a comparable statistical parameter?

Does anyone have experiments with calculating the SD of the AUC for a k-fold gbm.step result or calculating the SE of the AUC for a k-fold Maxent Result?

Thank you very much,

Best regards,
Johannes