I think it will help you. 

Thank you Nilotpol!

dear Casey you can try to assess the suitaility of your own database (extracting absolute values with GIS software, I reccomend ArcGis) and analyze the correlation between the suitability and the environmental variables for the entire database in each species using paired t-tests.

Jose Benavent-Corai, yes I used ecosim to calculate the values of niche overlap. It worked great. But I don't know how to see if two niche overlap values are significantly different.

Here is an example:

Species A and B have overlap of 0.6.

Species B and C have overlap of 0.4

I want to know if the values 0.6 and 0.4 are different from one another, so then I can state whether A and B have more overlap than B and C, or if there is no difference between the overlaps.

You can use ENMTools (http://enmtools.blogspot.com.br/).

The program will automatically measure overlap using two different statistics –

Schoener’s D (Schoener 1968) and the I statistic (see Warren et al. 2008 for additional details). 

Right now I am not aware of any specific function to derive confidence intervals for Pianka's niche overlap, nor for its parent (MacArthur-Levin's).

However, from your data description it seems to me that perhaps you could take advantage of jackknifing or, perhaps better, bootstrapping, as Macedo's answers imply/suggest. If your species' occurrences have been recorded one by one, why not pseudo-sample them? You can use almost any bootstrap software, or do it yourself in Excel. 

For jackknifing, you should drop in turn one subsample from your dataset, and repeatedly calculate the resulting overlap. The tricky part here (and what depends on how you collected your data) would be how to define "subsample".  Assuming that your data have been recorded individually (occurrence-by-ocurrence) a jackknifed subsample could simply be your dataset minus one of the occurrences. That however would possibly yield extremely biased (narrow) confidence intervals. Another approach could perhaps be to stratify your dataset in some sensible way (e.g. grouped by date of collection) and exclude certain ranges in turn. It's up to you, who know the data, to decide whether such approach would work. In practice, a [sufficiently large, but smaller than your total dataset] random subsample could work as well.

Bootstrapping could possibly render a better service. In that case you should just resample your data with replacement. The probability for an occurrence to enter the bootstrap sample should be proportional to the species-site frequency. Your bootstrap samples would have the same number of occurrences as your original dataset. This should be done a large number of times; the literature tends to round it up to 500-100 replicates.

In either case, you'd get an overlap measure for each random (bootstrap or jackknife) sample, and thus obtain an average overlap for each pair together with its variance and confidence interval, thereon to be used through an appropriate (but perfectly common) statistical test to determine significance.

Please remember that jackknifed and bootstrapped estimates tend to be somewhat biased, though. They can't be as accurate as a proper set of replicas.

Update for those still following this question. Thank you to Macedo and Arino for your suggestions to use randomization/permutation testing to derive confidence intervals around my values of niche overlap. I ended up using jackknifing because it worked better for my data. My data consisted of visitation rates by multiple species at many sites, with site being the basic unit of analysis. To jackknife these data, I dropped one site at a time from the data set and recalculated niche overlap. This procedure returned a variance that I used to produce confidence intervals and then from there test whether two values of overlap were statistically different from one another. Probably not the most elegant solution, but the best option I was able to find.

Did you try with the ecospat R package?? this pack. bring the same test of warren (Similarity and equivalence) applied to environmental space. Cheers!

As this question still appears to be of interest I am adding my 2 cents here, too. I agree with Newton and suggest to use ENMTools. In particular I recommend the most recent version (the R package) from github; it can be installed via devtools.

Then, instead of Shoener's D in geographic space (across the cells of the two suitability surfaces) and in line with what Ricardo recommends, do a correlation in environmental space based on latin hypercube sampling. Look up the env.overlap function in the ENMTools package. (You will need the two actual models in R not just the suitability rasters.)

Below is why, but I can't remember where I read that. I am quite sure it was from Dan Warren himself though. Also I might put it somewhat wrong. Anyway:

[Both D and I are overestimates if the number of unsuitable habitat cells is large, i.e. when modeling two species specialized to different but rare conditions across the landscape as they well agree in what is not suitable.]

https://github.com/danlwarren/ENMTools

You can use Schoener's D metric in R =library (ENMeval)