Multiple autocorrelations

Interfaces brms
1

Is there a way to have multiple autocorrelations in a brms model.

I’m looking to add both cor_car and cor_arma (or cor_ar) to my model

Thanks!

  • Operating System: 64bit Windows10
  • brms Version:2.10.0
2

Not right now but this will be possible in the future once autocorrelations can be specified within formulas.

3

Hi Paul, this is a following up question.
Is brms package able to fit AR(1) \otimes AR(1) in the model?
I use

autocor = cor_ar(formula = ~ x | y,p = 1, cov = FALSE)

in the model that it is AR(1) along x which is nested in y. However, y is not AR(1).

Thank you in advance for your help.

4

Not yet, but as soon as we have something like AR(1) models for random effects (not sure when this will come) it should be possible.

5

Hi Paul, thanks for the prompt reply.
I hope the day will come soon.

BTW, there is a package named INLA using the following syntax for AR(1)\otimes AR(1) model

f(x,model=“ar”,group=y,control.group=list(model=“ar”)).

If the AR(1) could work for random effects in brms, it may use a similarly syntax like

autocor = cor_ar(formula = ~ x | y,p = 1, control=cor_ar(p=1))

Or even a simpler way that

autocor = cor_ar(formula = ~ x ):cor_ar(formula = ~ y)

Anyway, thank you for your time and help.

6

Thanks. Please note that with recent brms versions, AR terms are now directly specific within the formula, e.g. x ~ ar(p = 1)

7

Got it. Thank you. I will start using the new syntax.