update table in paper

[teixeirak]

@mcgregorian1,

I've proposed changes to the table in the paper. Let me know if you think it makes sense. It will be possible to make this more compact outside of markdown, so hopefully everything will fit reasonably. If not, we could split out the individual years into a separate table.

This new version compares against the best full model with response variable removed (if part of best full model).

If you don't have a way to automatically populate this, you can just create a .csv and link to it.

[mcgregorian1]

The table makes sense, and yes I can automatically code that. I can make that table once I confirm with you on #19

[teixeirak]

@mcgregorian1, let's create two versions of this table: 1- testing against the best full model 2- testing against limited model (i.e., random effects, year, strongly interacting variables such as height) I've updated the manuscript outline to include both of these, and provided a bit more description of each. One will end up in the main paper, and one in the SI material. For now, I've assigned #2 to the SI, but we'll want to see how the results come out to know which is better for clearly communicating the findings.

The null models may shift a bit as we include more variables (e.g., sapwood area ratio, #21 ; percent leaf shrinkage? - hydraulics trait repository, but see this issue; wood density? ; SLA? )

[mcgregorian1]

@teixeirak I've updated the full_mod_all in manuscript to include sap_ratio, which comes out as moderately strong for all years but is only in the top model for 1977.

For the limited model table, do you want to still use height?

For the full model table, our best full model for all years now is "resist.value ~ height_ln*elev_m+tlp+rp+year+(1|sp/tree)".

Also as a reminder (and for myself), "results_full_models_combined_years.csv" shows the table but testing the full best model from before I did leaf collections; in other words, this is what our full table will look like

[mcgregorian1]

@teixeirak there are different results. I was looking back at the data and was realizing the dbh (and thus the height and sap ratio) hadn't merged properly and were all 1 value per tree, instead of different values for each year.

After fixing this, sap ratio is significant for all years except for 1966. I've uploaded this to github (full_mod_all)

[teixeirak]

@mcgregorian1, the results table currently lists "canopy<subcanopy" as the prediction for position, and I don't know how to interpret +/- sign on the coefficients. Could you please update to (1) ensure you're using the 4 categories, (2) if possible, summarize order of coefficients in the table (e.g., "dominant < codominant < intermediate < supressed".

[mcgregorian1]

@teixeirak The coefficient symbols (+/-) refer specifically to our null models (in this case, the simplified version, where we were just testing against null models of height). I believe we were saying that if the coefficient is -, then the having the variable makes a better model. If the coefficient is +, then that's our way of saying we predict the added variable makes a worse model.

I've updated the table to be more clear about the positions.

[teixeirak]

actually, I meant the +/- symbol to indicate the predicted (and observed) sign of the coefficient (i.e., direction of correlation). We hypothesize that all of these variables will make the model better, and the dAIC gives us that info (so no need to repeat). Actually, I don't think we want the coefficients (just their sign) in the final table, but its useful to have it at this stage.

[mcgregorian1]

Ah, I see what you're saying. Yes, I think we talked about this briefly before, noting that the numbers are for us mainly at this stage but not for the final product

[teixeirak]

Closing (obsolete).