Data Reduction for HyperNiche

[Ken]

Hi all,

I am studying the response of a target species to a suite of habitat variables (around 100). I need to reduce the data set for use in HypoerNiche. I could use ordination I believe but... I really want to see the response of the species to specific habitat variables rather than a synthetic variable.

The data mostly consists of the presence/absence of a habitat variable at locations where the target species was found versus randomly located samples (Habitat availability versus habitat use). So, there are a lot of 1s and zeros.

I had thought of using ordination and selecting variables based on the correlation coefficient scores: those variables that show the strongest response along a habitat gradient. But I cannot find where anyone else has done this. Is this valid?

Any suggestions given that I would like to see the response of the species to actual variables and not synthetic?

Thanks,

Ken

[Bruce McCune]

The downside of using that many predictors is that you can fit a model to almost any response variable. But the permutation test will yield a poor p value, revealing the emptiness of that result.

I do think it makes sense to do some filtering ahead of time. Using ordination to identify good variables in your system makes sense. So does eliminating strongly redundant variables. Although it sounds like you don't want to go the route of synthetic variables, considering doing this for subsets of interrelated variables.

The only other remedy that I can think of is increasing the sample size, but that may not be in the cards.

Bruce McCune

--
You received this message because you are subscribed to the Google Groups "HyperNiche and NPMR" group.
To unsubscribe from this group and stop receiving emails from it, send an email to hyperniche+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Ken]

Bruce,

Thank you for your reply.

Yes, I would like to reduce the data to 15-20 variables for use in HyperNiche. In the past I have used synthetic variables, but in this study we really want to know about specific factors.

My big issue is that I have found only a few agency reports and stats websites mentioning using ordination for data reduction by selecting apparently informative variables rather than using site scores/axes for analyses. So I have questions about validity and I cannot help but wonder why this is not a more common method of data reduction.

Is there any reason why this approach is invalid? Any sticking points to using it that you are aware of (other than a lack of works to cite)?

Thank you,

Ken

[Bruce McCune]

The only problem I can see with it is if there is an issue of circularity. For example, say you have ordinated communites and found environmental variables X10 and X22 most important out of 100 randomly generated environmental variables. Then if you use NPMR to regress a community dominant against X10 and X22 -- it will find a strong relationship and, if you have excluded the other 98 variables, it will beat the randomization test.

That circularity would not be as much of an issue if the response variable for NPMR was not a controlling variable in the community ordination.

Bruce