mean.descendants
Closed this issue · 2 comments
I was wondering if you could comment on the interpretation of the 'mean.descendants' option for ilr.weights in the philr function and when its use may be appropriate. My thinking is that this weighting would be giving more weight to nodes/balances deeper in the phylogeny. Is that correct?
Related to this, I am trying to interpret a data set where including branch length weightings in the philr calculation reduces the explanatory power of a model. I have a distance-based model (using euclidean distance of the philr transform) of some microbiome data that fits much better (r2 = ~0.5) when using ilr.weights='uniform' than it does when using ilr.weights='blw.sqrt' (r2 = ~0.25). I was initially thinking that this was the because most predictive balances were closer to the tips of the tree, but I found that results using ilr.weights='mean.descendants' are very similar to ilr.weights='uniform' .
First off, my apologies for the delay in the response.
mean.descendants does end up giving more weight to nodes that have large subtrees descending from it. It is the same as sum.descendants but then further adds the mean distance from each direct child to the tips that descend from it.
Regarding the second question. This could be some biology at play but first I would check that its not some trivial technical issue. For example, I could imagine seeing this if you are including a ton of taxa that are very rare.
Justin
closing due to inactivity