Computing associations due to environmental covariates

[Leitemfa]

Hello,

I see that Hmsc has a function computeAssociations to investigate the residual correlations, but I would like to see also the associatios due to covariates, as they seems to be the shared environmental response. Is there any way to extract this information now?

Thanks, Marcio

[ovaskain]

Hi,

There is no specific functionality for this (yet); at some point we could indeed implement a function that computes associations due to responses to covariates. One thing that can be done is to estimate both the raw associations (from an intercept-only model) and residual associations (from model with covariates), the difference between these two tells about the associations due to covariates.

Otso

On 30/10/2019 10:06, Márcio Fernandes Alves Leite wrote:

Hello,

I see that Hmsc has a function computeAssociations to investigate the residual correlations, but I would like to see also the associatios due to covariates, as they seems to be the shared environmental response. Is there any way to extract this information now?

Thanks, Marcio

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHubhttps://github.com/hmsc-r/HMSC/issues/28?email_source=notifications&email_token=AEIYMZWW4CSQEFD732W6OXDQRE6ALA5CNFSM4JGVBE22YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HVKFZYQ, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AEIYMZW76TF5CVWUZXJVTL3QRE6ALANCNFSM4JGVBE2Q.

-- Otso Ovaskainen Professor Department of Biosciences University of Helsinki Finland

otso.ovaskainen@helsinki.fimailto:otso.ovaskainen@helsinki.fi

[janinepdasilva]

Hello,

I would like to know it there were any developments regarding this question. I have used the difference between raw and residual correlations to justify the associations due to the covariates, but the reviewers made some pretty tough comments on this, so I was wondering if I could do something to clarify this part of my manuscript.

Thanks, Janine

[gtikhonov]

Still no specific functionality has been implemented for this purpose at this point. Indeed, the way how Otso has suggested to fit two models - with covariates and without and look at the difference of residual associations is one potential option. Another potentially viable option to get the "X-induced-only" covariance matrix, which does not require deep hacking within the package, would be to sample with replacement from the covariates' joint combinations of the training data, run predict(..., predictEtaMean=TRUE) function for such "new" covariates data, and simply calculate the empirical distribution across the predictions. @ovaskain or @jarioksa , could you comment on such proposal?

[janinepdasilva]

The difference between raw and residual correlations is very useful, except in the case associations pop up in both models - as it is the case with my data. Thank you so much for your suggestion, I'll try to manage this issue that way.