search

SCBI-ForestGEO / McGregor_climate-sensitivity-variation

repository for linking the climate sensitity of tree growth (derived from cores) to functional traits
0 stars 0 forks source link

Final tasks remaining #97

Closed mcgregorian1 closed 3 years ago

mcgregorian1 commented 4 years ago

This may not need to be an issue but I'm putting the remaining things that need to do be done (before submitting again) here so it's easier for me to remember.

I've put these in relative order of importance.

statistics

reviewers

Other formatting

mcgregorian1 commented 4 years ago

A note about Variance Inflation Factor (VIF).

I'm doing this to ensure that our best models indeed have low multicollinearity. When I run the full thing, I notice that height, TWI, and the interaction have absurdly high values. image

To double-check this, I run VIF as though there is no interaction, and I get perfectly good values for both height and TWI. This means that the interaction is responsible for the high values. I haven't been able to find anything about this online. I asked Monika about this as she ran into the same problem for her own research, and she said even with asking Katie Edwards (her advisor), their thoughts were that they didn't know why the interaction did this. Overall though, they acknowledged it and moved forward. image

teixeirak commented 4 years ago

I'm flagging @ValentineHerr to see if she happens to know something about this (Valentine, no need to research this if you don't). My inclination would be the same-- acknowledge and move on.

ValentineHerr commented 4 years ago

an interaction means you include V1 + V2 + V1xV2, so V1xV2 will by definition be highly correlated with V1 and V2. This is not a collinearity issue you should worry about and looking at the VIF in an equivalent model without interaction (the one that givesyou all the ones) is the way to go.

mcgregorian1 commented 4 years ago

Cool, thanks @ValentineHerr !!

mcgregorian1 commented 3 years ago

@ValentineHerr for clarification, is VIF something that's normally reported in paper texts? Or is it meant to be an exploratory thing that is assumed to have been done from the outset?

If it is reported, is it ok to do a one-sentence thing stating something like "All variables had variance inflation factors <1.5 (1 +/- 0.2)." ?

ValentineHerr commented 3 years ago

Yeah one-sentence thing is good.

teixeirak commented 3 years ago

I've added that here:

Analyses focused on testing the predictions presented in Table 1 with $Rt$ as the response variable, and then repeated using $Rt_{ARIMA}$ as the response variable. Models were run for all drought years combined and for each drought year individually. The general statistical model for hypothesis testing was a mixed effects model, implemented in the lme4 package in R [@R-lme4]. In the multi-year model, we included a random effect of tree nested within species and a fixed effect of drought year to represent the combined effects of differences in drought characteristics. Individual year models included a random effect of species. All models included fixed effects of independent variables of interest (Tables 1,3) as specified below. All variables had variance inflation factors <1.5 (1 +/- 0.2). We used AICc to assess model selection, and conditional/marginal R-squared to assess model fit as implemented in the AICcmodavg package in R [@R-AICcmodavg]. AICc refers to a corrected version of AICc, and is best suited for small data sizes [see @brewer_relative_2016].

mcgregorian1 commented 3 years ago

Sounds good. Those were made-up numbers - I need to go back and get the correct ones.

On Wed, Jul 22, 2020 at 11:33 AM Kristina Anderson-Teixeira < notifications@github.com> wrote:

I've added that here:

Analyses focused on testing the predictions presented in Table 1 with $Rt$ as the response variable, and then repeated using $Rt_{ARIMA}$ as the response variable. Models were run for all drought years combined and for each drought year individually. The general statistical model for hypothesis testing was a mixed effects model, implemented in the lme4 package in R [@R-lme4]. In the multi-year model, we included a random effect of tree nested within species and a fixed effect of drought year to represent the combined effects of differences in drought characteristics. Individual year models included a random effect of species. All models included fixed effects of independent variables of interest (Tables 1,3) as specified below. All variables had variance inflation factors <1.5 (1 +/- 0.2). We used AICc to assess model selection, and conditional/marginal R-squared to assess model fit as implemented in the AICcmodavg package in R [@R-AICcmodavg]. AICc refers to a corrected version of AICc, and is best suited for small data sizes [see @brewer_relative_2016].

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/SCBI-ForestGEO/McGregor_climate-sensitivity-variation/issues/97#issuecomment-662523457, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJNRBEKPVVXRTJ4EJJDL23DR44BGFANCNFSM4OQ3Z7EQ .

--

Ian McGregor

Ph.D. Student | Center for Geospatial Analytics

He/Him/His

College of Natural Resources

Jordan Hall 4120 | Campus Box 7106

North Carolina State University

2800 Faucette Dr.

Raleigh, NC 27695 USA imcgreg@ncsu.edu | 714-864-1005 | geospatial.ncsu.edu

teixeirak commented 3 years ago

Okay, please be sure to correct in the text when done.

teixeirak commented 3 years ago

I think we can close this issue once this last task is done. It just references other open issues.

mcgregorian1 commented 3 years ago

Okay, please be sure to correct in the text when done.

This is done.

mcgregorian1 commented 3 years ago

Closing this per @teixeirak's comment