Hi All,
Could someone explain the reasoning behind choosing the dependent trait in pairwise logistic regression of two categorical traits. My hypothesis does not explicitly state which trait is the IV as I just need the relationship between traits. To word this another way, I dont understand theoretically why the directioning in the model is important.
I have ~100 traits of a binary (presence/absence) type and am running pairwise logistic regressions on each trait combination using an R package that fits phylogeny as a covariant.
when i=19 and j=20
when i=20 and j=19
So in the end I get an asymmetrical heatmap if I plot out the loglikelihoods from each regression. I take this to mean (for example) that trait1 is more dependent on trait2 than trait2 is on trait1?
My overall plan is to compare the divergence of a phylogenetic model with a standard logistic regression (subtract log likelihoods), but need to understand why swapping the model around gives different results first.
I would appreciate any suggestions here.
Cheers,
Lesley
Could someone explain the reasoning behind choosing the dependent trait in pairwise logistic regression of two categorical traits. My hypothesis does not explicitly state which trait is the IV as I just need the relationship between traits. To word this another way, I dont understand theoretically why the directioning in the model is important.
I have ~100 traits of a binary (presence/absence) type and am running pairwise logistic regressions on each trait combination using an R package that fits phylogeny as a covariant.
Code:
phyloglm(trait[,i]~trait[,j], data=datafile, phy=treedata, method = c("MPLE","IG10"))
Code:
logreg$loglik
[1] -6.276559
Code:
logreg$loglik
[1] -67.33166
My overall plan is to compare the divergence of a phylogenetic model with a standard logistic regression (subtract log likelihoods), but need to understand why swapping the model around gives different results first.
I would appreciate any suggestions here.
Cheers,
Lesley