search

hputnam / Moorea_Sym

Symbiodinium communities of specifist and generalist corals
0 stars 0 forks source link

Finalize CAP analysis #30

Open hputnam opened 6 years ago

hputnam commented 6 years ago

Finalize variable choice - @daniclaar Can you check the new abbreviated version of the CAP on the rmd wrt appropriateness and interpretation?

If necessary - remake the table if necessary. If not, list % explained in the text.

jrcunning commented 6 years ago

PS Danielle: we reasoned that “dip” is not interpretable/useful since the temp data are so fragmented seasonally, and an error in the temp processing was inflating the bimodality. So, without dip we found that the three metrics that differ most among sites are daily mean range, daysover30, and var — they are highly correlated and load equally on PC1. Without attempting to disentangle them, we reasoned we should use these same three metrics in the CAP to determine how much variability in sym composition can be explained by them together, and again without trying to use model selection or any other approach to disentangle them.

On Feb 10, 2018, at 10:34 AM, Hollie Putnam notifications@github.com wrote:

Finalize variable choice - @daniclaar https://github.com/daniclaar Can you check the new abbreviated version of the CAP on the rmd wrt appropriateness and interpretation?

If necessary - remake the table if necessary. If not, list % explained in the text.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/hputnam/Moorea_Sym/issues/30, or mute the thread https://github.com/notifications/unsubscribe-auth/AIKtn4ozwMPWHAJzvitsBXukQwsL3i8mks5tTbcLgaJpZM4SA93p.

Ross Cunning, Ph.D. Postdoctoral Researcher University of Miami, RSMAS 4600 Rickenbacker Csy Miami, FL 33149

daniclaar commented 6 years ago

Seems reasonable re: dip.
A few questions (since I wasn't there for the discussion):
Why did you decide to choose those three variables and run them separately? If they are well-represented by PC1, then modelling phy.f.t.porites ~ PC1 makes sense (and if you run a backwards stepwise analysis on PCs 1-4, only PC1 is retained for Porites). It seems like that is doing what you want it to do. (Side note, when you run the same analysis on Acropora, no PCs are retained and the model is phy.f.t.acropora ~ 1...)
I interpreted the by-variable model selection as a way to look at which specific metrics are driving the Symbiodinium community, so it's slightly different than using the PCs. I think if we're going to do that, we should include all of our potential variables (i.e. what's in the spider plot in Fig 1). By using ordistep, then you could choose the most important variables there. If you guys agreed not to include that, no problem, then I think it's best to only include/present the results using the PCs, and then discuss which variables are coordinated with those PCs. Basically, if we want to include specific variables in the models, I think we should include all potential variables and run backwards stepwise selection, since choosing variables ahead of time might subjectively influence the output.

Thoughts?