dbRDA residuals as response variable for adonis2?

[jkleinkopf]

Hi,

I am working on what I thought was a pretty straightforward question that many have explored before: What environmental variables explain differences between biological communities? I have presence-absence data for about 100 communities, and I have calculated pairwise turnover (Jaccard and UniFrac distances) for every pair of communities. I have a handful of environmental predictor variables, both categorical (e.g., soil type) and continuous (e.g., mean annual temp).

To this point, it makes sense to me that adonis2 would be straightforward to use—something like adonis2(unifrac ~ soil_type + mean_ann_temp). However, because my communities are distributed continentally, I have several other variables, like the sampling year, latitude, and longitude, among others, that I'd like to account for before using the adonis2 function.

My question is, is there a way to account for these variables in adonis2 first and then see the marginal effect of the variables of interest afterward? Alternatively, is it reasonable to first run dbrda() using the variables I want to take into account, and feed the residuals of that into adonis2 with my predictor variables of interest? Thanks in advance!

Here's an example of what I mean, using the mite dataset:

mite.dist <- vegdist(mite)
mite.vars <- cbind(mite.env, mite.xy)

Let's say I want to account for x, y, and Shrub before I look at SubsDens and WatrCont

# Include x, y, and Shrub as part of the adonis2 formula
adonis2(mite.dist ~ x + y + Shrub + SubsDens + WatrCont,
             data = mite.vars,
             sqrt.dist = T,
             by = "margin")

# adonis2(formula = mite.dist ~ x + y + Shrub + SubsDens + WatrCont, data = mite.vars, sqrt.dist = T, by = "margin")
#       Df SumOfSqs      R2      F Pr(>F)   
# x         1   0.4128 0.01912 1.6458  0.078 . 
# y         1   0.7404 0.03429 2.9515  0.003 **
# Shrub     2   1.0105 0.04679 2.0141  0.002 **
# SubsDens  1   0.2185 0.01012 0.8709  0.594   
# WatrCont  1   0.4985 0.02308 1.9871  0.021 * 
# Residual 63  15.8035 0.73183                 
 #Total    69  21.5946 1.00000                 

Alternatively, use dbrda to remove the effect of x, y, and Shrub from mite.dist

d <- dbrda(mite.dist ~ x + y + Shrub, 
           data = mite.vars, 
           sqrt.dist = T)

d.resid <- residuals(d)

# Just looking at the effect of SubsDens and WatrCont on turnover after accounting for x, y, and Shrub
adonis2(d.resid ~ SubsDens + WatrCont,
              data = mite.vars,
              sqrt.dist = T,
              by = "margin")

 # adonis2(formula = d.resid ~ SubsDens + WatrCont, data = mite.vars, sqrt.dist = T, by = "margin")
 #       Df SumOfSqs      R2      F Pr(>F)
 # SubsDens  1   0.3241 0.01923 1.3447  0.121
 # WatrCont  1   0.1916 0.01136 0.7948  0.801
 # Residual 67  16.1485 0.95795              
 # Total    69  16.8574 1.00000              

[jkleinkopf]

I believe this has been sufficiently answered here: https://stats.stackexchange.com/questions/672483/using-the-residuals-of-a-dbrda-analysis-as-a-response-in-an-adonis2-analysis Thanks!