MRPP on Raw or relitivized (Hellinger) data

[David McNear]

Hi, 

I Hellinger transformed my main matrix of PLFA bacterial biomarker concentrations and then performed an NMDS.  To evaluate the significance of the group separation I usually use MRPP.  Is it ok to perform the MRPP on the transformed matrix, or should it be performed using the raw data?  

Thanks in advance for your help 

[Bruce McCune]

MRPP doesn't do any built-in relativization, so you should feel free to apply whatever data transformations or relativizations are necessary or desired before running MRPP.

Bruce McCune

--
You received this message because you are subscribed to the Google Groups "PC-ORD" group.
To unsubscribe from this group and stop receiving emails from it, send an email to pc-ord+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/pc-ord/e0f62a83-10ea-4135-b6ee-ae5bf1f9e124n%40googlegroups.com.

[David McNear]

Is it ok to show the NMDS generated using the relativized main matrix data, but perform the MRPP on the raw data?  The MRPP on the raw data seems to explain the NMDS generated using the relativized main matrix data better than using the relativized main matrix data (if that makes sense).  

[Bruce McCune]

I suppose that is possible, but usually it is the other way around. People wonder why one analysis isn't concordant with another when the two are based on different relativizations of transformations. If you chose the same data adjustments and  distance measure for the ordination and the MRPP, I would expect them to be at least roughly concordant. The one common exception to that is that sometimes an ordination will show only vague separation of groups but MRPP will say they are rather different. This frequently happens because MRPP looks at all dimensions of the community space, while ordination is looking at just a couple of the strongest synthetic dimensions.

Bruce McCune

To view this discussion visit https://groups.google.com/d/msgid/pc-ord/751ea90f-b999-4d84-9360-f6f745b8d804n%40googlegroups.com.

[David McNear]

We normally correct our MRPP results for multiple comparisons using the Benjamin-Hochberg (1995) method with a FSD of 0.05.  Maybe the cutoff is too stringent....?  

[Bruce McCune]

David, Although this method of controlling false discovery rates seems reasonable, it still presumes that we are more concerned with type I errors than type II. I have never been clear on why that makes sense in ecology. Regardless of that, your manuscript reviewers are more likely to be obsessed with type I than type II errors, so maybe it's ok.

Bruce McCune

To view this discussion visit https://groups.google.com/d/msgid/pc-ord/0da8e7de-c8fd-4387-873f-f4639b46a995n%40googlegroups.com.