Closed teixeirak closed 4 years ago
I'm also realizing, as I go to fill in Table 1, that we're now missing our tests of the individual-level significance of each non-trait variable. (For instance, a negative effect of height is supported in all top models, but it's only a statistically significant improvement if there are no models within ∆AICc =2 of the top model that exclude the term.
This could be solved by showing any models within ∆AICc =2 in Tables 5/ S5. That gets a little ugly, so could be an SI table, with the main text table showing just the top model.
I've updated findings in Table 1, but this cannot be finalized until we solve the issues mentioned above.
@mcgregorian1 , what we do with respect to the first comment in this issue is pretty subjective/ a matter of presentation. I don't think the DBH test would be missed, and the test of canopy position without height is not needed, and I'm not even sure if its a fair comparison. So, I'd probably drop both of those.
Then the solution is simple: output tables parallel to 5 and S5 for all models within AIC=2. These should probably both go in the appendix, and the main text should have only the top model for each drought year-- could potentially include both Rt and Rt_ARIMA_ratio.
Quick question on this, are we officially still going with dAIC<2? I thought we were going to stick with 1
On Thu, 9 Jul 2020, 17:05 Kristina Anderson-Teixeira, < notifications@github.com> wrote:
I've updated findings in Table 1, but this cannot be finalized until we solve the issues mentioned above.
@mcgregorian1 https://github.com/mcgregorian1 , what we do with respect to the first comment in this issue is pretty subjective/ a matter of presentation. I don't think the DBH test would be missed, and the test of canopy position without height is not needed, and I'm not even sure if its a fair comparison. So, I'd probably drop both of those.
Then the solution is simple: output tables parallel to 5 and S5 for all models within AIC=2. These should probably both go in the appendix, and the main text should have only the top model for each drought year-- could potentially include both Rt and Rt_ARIMA_ratio.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/SCBI-ForestGEO/McGregor_climate-sensitivity-variation/issues/99#issuecomment-656350890, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJNRBEJMP64OH4V43GCWRGLR2YWITANCNFSM4OV4572Q .
1 is the cutoff for mode candidate variables. 2 is for significance.
Gotcha. I can hopefully start looking more at this tomorrow
On Thu, 9 Jul 2020, 18:00 Kristina Anderson-Teixeira, < notifications@github.com> wrote:
1 is the cutoff for mode candidate variables. 2 is for significance.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/SCBI-ForestGEO/McGregor_climate-sensitivity-variation/issues/99#issuecomment-656371693, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJNRBEKJICI3YFS2TVYUZLDR2Y4WPANCNFSM4OV4572Q .
I've updated findings in Table 1, but this cannot be finalized until we solve the issues mentioned above.
@mcgregorian1 , what we do with respect to the first comment in this issue is pretty subjective/ a matter of presentation. I don't think the DBH test would be missed, and the test of canopy position without height is not needed, and I'm not even sure if its a fair comparison. So, I'd probably drop both of those.
Given canopy position and height are somewhat correlated (plus we need one to understand the other), I agree with you on that.
Regarding DBH, it had a correlation of 1 with height according to the correlation plot. I think that's good enough evidence for only including height and disregarding DBH in this case (we could also argue the opposite but of course we're using height for other things anyway).
Then the solution is simple: output tables parallel to 5 and S5 for all models within AIC=2. These should probably both go in the appendix, and the main text should have only the top model for each drought year-- could potentially include both Rt and Rt_ARIMA_ratio.
Sounds good! I'll get on this.
1 is the cutoff for mode candidate variables. 2 is for significance. ...Then the solution is simple: output tables parallel to 5 and S5 for all models within AIC=2
So to make sure we're fully on the same page based on your comments above:
Is this correct?
Regarding DBH, it had a correlation of 1 with height according to the correlation plot. I think that's good enough evidence for only including height and disregarding DBH in this case (we could also argue the opposite but of course we're using height for other things anyway).
To be clear, I'd never include them both in the same model. I'd follow similar structure as before, where we first say, yes larger (DBH) trees have greater growth reductions during drought, now we want to dig into why.
So to make sure we're fully on the same page based on your comments above:
* 1 is the cutoff for the candidate variables (Table 4/S4) and the model runs themselves (Table 5/S5) for the main manuscript. * 2 is the cutoff for the model runs for the appendix parallels of Table 5/S5.
Is this correct?
That's mostly correct.
1 is the cutoff for whether a trait variable can qualify for inclusion in the main model. Just that-- it is never a test of significance. Regarding the main text table 5, we could potentially use 1 as a cutoff, but I think it would make more sense to show just the top model with astricks to indicate significance. The one thing this wouldn't capture, though, is the evidence that top models for 1999(Rt) or 1977(Rt_ARIMA) have either TLP and PLA (not both). I think that's worth somehow representing in the main doc, but not sure how.
2 is the criteria for significance (following the accepted convention). So, yes, 2 is the cutoff for the model runs for the appendix parallels of Table 5/S5.
Regarding table 5 in the main doc, let's go ahead and produce the SI tables with the AIC=2 cutoff, and then decide how to present in the main text.
Regarding DBH, it had a correlation of 1 with height according to the correlation plot. I think that's good enough evidence for only including height and disregarding DBH in this case (we could also argue the opposite but of course we're using height for other things anyway).
To be clear, I'd never include them both in the same model. I'd follow similar structure as before, where we first say, yes larger (DBH) trees have greater growth reductions during drought, now we want to dig into why.
Gotcha. I guess what I'm referring to is I think some people would have seen a correlation of 1 in the raw data, and used that as the justification to never look at DBH from the start. But that's fine.
Regarding table 5 in the main doc, let's go ahead and produce the SI tables with the AIC=2 cutoff, and then decide how to present in the main text.
Sounds good, I've updated Table S5 for the AIC=2 threshold. Table S4 of course was not updated as it's already operating on the AIC=1 cutoff.
Gotcha. I guess what I'm referring to is I think some people would have seen a correlation of 1 in the raw data, and used that as the justification to never look at DBH from the start. But that's fine.
Oops, I'm not sure I was clear. I don't think we need to look at DBH. I was just clarifying that I wouldn't consider including it in the full model along with height. Still, presenting it adds little to the paper; let's just drop it.
Sounds good, I've updated Table S5 for the AIC=2 threshold. Table S4 of course was not updated as it's already operating on the AIC=1 cutoff.
Thanks!
Oops, I'm afraid my instructions were confusing....I see you updated the ARIMA analysis with the AIC=2 threshold, but not yet the main analysis. I need to see those results before giving guidance on what to put in the new Table 5. Could you please run that/ update table 5? We'll take it from there.
Just checking, does your code also allow for the other variables (H, CP, etc) to not be included in top models? I'm a little surprised that CP comes out in all top models.
Oh, no it does not - I've been assuming this whole time that we kept the base models the same from the single variable to the multi variable analysis..
On Sun, 12 Jul 2020, 07:33 Kristina Anderson-Teixeira, < notifications@github.com> wrote:
Just checking, does your code also allow for the other variables (H, CP, etc) to not be included in top models? I'm a little surprised that CP comes out in all top models.
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/SCBI-ForestGEO/McGregor_climate-sensitivity-variation/issues/99#issuecomment-657209686, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJNRBEJSJJIFHI3RR7BOGETR3GNSLANCNFSM4OV4572Q .
No, we want effectively the same as we had before. The single-variable model is just for testing traits individually. (On a related note, I think we selected that because it's what came out in the top model with no traits, right? )
From what I remember, height was in every top model so that was a no brainer. Crown position was in a number of top models <2 dAIC, and TWI was in slightly more top models than crown position.
Ok. I will be starting to work on this about 9 so I'll rerun this allowing for all combinations
On Sun, 12 Jul 2020, 07:41 Kristina Anderson-Teixeira, < notifications@github.com> wrote:
No, we want effectively the same as we had before. The single-variable model is just for testing traits individually. (On a related note, I think we selected that because it's what came out in the top model with no traits, right? )
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/SCBI-ForestGEO/McGregor_climate-sensitivity-variation/issues/99#issuecomment-657210375, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJNRBEIC264MFJ36PJPKNPDR3GOONANCNFSM4OV4572Q .
I'm going through this - just wanted to clarify that when I ask for all possible combinations of variables in the model, I'm counting TWI*height interaction as a standalone, and not including TWI and height alone? Or do we want that?
We still want height and/or TWI when their interaction is not significant.
Sorry this took so long - I had to get code out of the archive code I set aside from earlier.
This is what the manuscript table5 (Rt) looks like allowing for all possible combinations, with dAICc <=2. I will do the same for arimaratio.
Arimaratio has been updated as well.
Okay, thanks. Looks about as expected. I'll look it over.
Actually, maybe best to take out the TWI*height interaction from the single-variable tests, as it's not significant in most models? Hopefully that doesn't change the variables selected, but I doubt it will; that seems to be pretty stable.
I assume you mean to include TWI and height separately then? I can try this.
Yes, each individually but not their interaction.
When I do this, this is when we get RP as a top candidate again:
Right, but it goes in opposite directions and therefore is disqualified as a candidate for full modes. It’s also important to note that we only have two diffuse porous species (LITU and FAGR), and they are at opposite extremes in terms of drought resistance.
Ah gotcha. You're right
So, let's update former tables 4 and S4 (now S4 and S6) with these results (single-variable trait tests without including height x TWI interaction). This won't affect Table 5.
I'm doing this now
These have been updated
@mcgregorian1 , I like our analysis reform (#95 ), but am now realizing that it has dropped two tests of predictions in Table 1: 1- effect of DBH (as opposed to height) 2- relative importance of height vs canopy position.
I'm currently not sure if we really need either of those-- something to think about as we work through revisions.
The DBH test in particular probably wouldn't be missed.
The test of height vs canopy position ties in with the issue that the two are collinear (mentioned in #94 , #95)