NMDS and MRPP on unequal sized sample groups

[Daniel Tucker]

Hi All, 

Multivariate beginner here, forgive me if my questions are overly simple. I am wanting to see if there are differences between bryophyte communities for three habitats I have defined in my test ecosystem 1) rocky outcrops, 2) meadows, and 3) epiphytic (on one tree species). I systematically sampled 8 of my epiphytic habitats (trees) to capture a range of trunk sizes that could have potentially different microhabitats and thus contribute to habitat variability. However, I was only able to collect 4 replicates of rocky outcrops, and 4 replicates of meadows due to time constraints.

My questions are: 

1) with an uneven sample size between my habitat groups am I going to run into problems interpreting my NMDS results? Or will my interpretation be logical because NMDS is non-metric and makes no distributional assumptions?

2) Likewise, can I trust the significance of my MRPP and the pairwise comparisons provided by the PC-ORD output if I have unequal sample sizes? I assumed that I can trust these because MRPP is non-parametric, is this correct? 

3) in order to get the NMDS to work with my abundance data I transformed my matrix with hellinger transformation. When I conduct MRPP and ISA should I be using my raw abundance data or the transformed matrix?

Thanks!
Dan

[Bruce McCune]

Dan, (1) I'd say that it's always a good idea to try to balance your design, but definitely not a big problem with the imbalance that you describe. Maybe more important with the n=4, 4, and 8 is that the sample sizes are overall fairly small, which makes it much more likely that a couple of odd sites will have undue influence on the outcome.

Nonmetric scaling doesn't care about how you have defined groups of sample units, since it's just trying to represent the variation among them as points in species space.

(2) By default, MRPP weights groups by their size in calculating the effect size and p value. Balancing a design is a consideration whether parametric or not. But yes, you can use MRPP with your data.

(3) Whatever logic you applied in transforming the data for NMS should also apply to other analyses that are based on a distance matrix. ISA is different in that no distance matrix is used, but the Hellinger transform will affect the results because it is relativizing by sample units (which changes the distribution of values for each species) and taking the square root of the values (which will affect the relative abundances calculated by ISA). This leaves the question about whether this is a good idea, which I can't answer because it would depend on the reason you are using Hellinger in the first place.

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 on the web visit https://groups.google.com/d/msgid/pc-ord/45a52234-2bb9-4d81-a881-c53a4f114e04%40googlegroups.com.

[Moacir Tinoco]

Dear Colleagues,

Hope everyone is fine!

I am currently reviewing a Master’s dissertation on herpetofauna diversity and communities organization and came across an analysis called: (Functional) Principal Components Analysis, never heard of it. I am fully aware about how PCA works, but this variation is new for me. 

Do any of the colleagues have any idea what is it all about? What is it intended to answer? And any suggestion on how to evaluate its application? 

Any comment or advice would be much appreciated! 

Best regards 

Moacir Tinoco

--

--- Moacir Tinôco From my mobile

[Bruce McCune]

This is new to me, but looking it up online, it seems to be a quite specific approach that would demand a special data form. From Wiki, for example,  "First, the order of multivariate data in PCA can be permuted, which has no effect on the analysis, but the order of functional data carries time or space information and cannot be reordered. "

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 on the web visit https://groups.google.com/d/msgid/pc-ord/CACP%2BtQarx5rkfJy63kPu%3Dr_M_atz_y%3D1P0t6jXtL05TheB%3Dr0w%40mail.gmail.com.

[Moacir Tinoco]

Thank you verymuch.

To view this discussion on the web visit https://groups.google.com/d/msgid/pc-ord/CAK1WPPkzpXmHGabRN5pm%3D8n0tiDht8fo%2Bjs4%2BhZMur-5C1%2BKWA%40mail.gmail.com.