Null model - considering the 95% quantile value as the 95% Confidence interval upper limit

[Areej Jaradat]

I'm confused from the statistical approach described below

According to my understanding the confidence interval is different from the quantile. The issue appeared when I was applying the null model approach of Raes & Steege, 2007 (to create a null 'random' model to evaluate the predictive accuracy of maxent 'real' model)

In that paper they wrote [assessed the 95% C.I. upper limit of AUC value, by ranking the 999 AUC values and selecting the 949th value (0.95*999=949)]. Another paper used the same approach wrote [ For each group of 500 null models the average AUC and the 95% Confidence Interval (CI) is calculated. The AUC of each ‘real’ species is then compared with the 95% CI of the ‘random’ species; if the AUC of the real species is higher than the 95% quantile value, this species is significantly different from random.]

If someone can discuss this here it would be great. 

[Adam B. Smith]

I think both papers are doing the same--if they were able to use "standard" statistics (e.g., statistics based on maximum likelihood), they could calculate (without simulation) the 95th CI (or any other CI).  But spatial things like species' ranges and occurrences rarely meet the assumptions inherent in these kind of tests, so Raes & ter Steege proposed a Monte Carlo test, where you randomize aspects of the data 999 times (the observed value is taken to be the 1000th value).  So the 95th quantile in this case is the same as the 95 CI.

Good luck,

Adam

-----------------------------------------------------

A free, online course in best practices in species distribution modeling in R

Using museum and herbarium specimens in ecology and evolution: A free webinar