model identifiability

[Sylvan Benaksas]

Hi INLA Team,

Normally in a model one factor level is dropped into the intercept, and with INLA if you take the automatic intercept this seems to happen. However, all online sources state for SPDE models to drop the intercept and add a specified term instead (y ~ -1 + Intercept + factor ...). Doing this leads to an intercept term and all factor levels having a coefficient as well, I.E no dropped level. Can you advise what is this best route in this situation?

Many thanks,

Sylvan

[Finn Lindgren]

+ myeffect(~ 1+ factor, model="fixed")

In inlabru was made to handle this.

Finn Lindgren
email: finn.l...@gmail.com

--
You received this message because you are subscribed to the Google Groups "R-inla discussion group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to r-inla-discussion...@googlegroups.com.
To view this discussion, visit https://groups.google.com/d/msgid/r-inla-discussion-group/4909fe93-55a2-440e-8bda-669a7ee83fa9n%40googlegroups.com.

[Sylvan Benaksas]

Hi Finn,

Thanks for the reply, is there an option for plain inla, as I do not use inlabru?

Sylvan

[Finn Lindgren]

You can precompute the indicator variables using the same method used by inlabru, with the MatrixModels model.Matrix method.

You can then use that as the A-matrix for a model=“iid” component where you set a fixed precision value. The system is very general, so theres more than one possible way to do it.

For mesh/SPDE models in particular, the inlabru interface was constructed to avoid the need for inla.stack on the user side of the code, and I consider using those models with the raw inla interface to be deprecated/obsolete, so I would encourage you to consider it.

Finn

On 8 Jul 2025, at 14:07, Sylvan Benaksas <sylva...@gmail.com> wrote:



[Sylvan Benaksas]

would dropping the intercept and using the following argument be an acceptable approach? it seems to produce a sensible model for me?

control.fixed = list(expand.factor.strategy = "inla")

Cheers,

Sylvan

[Finn Lindgren]

If you only have one factor variable component in the model, yes.

Finn

On 21 Jul 2025, at 15:48, Sylvan Benaksas <sylva...@gmail.com> wrote:

would dropping the intercept and using the following argument be an acceptable approach? it seems to produce a sensible model for me?

To view this discussion, visit https://groups.google.com/d/msgid/r-inla-discussion-group/775baab7-0d41-47c1-bd5d-ee6b5a2af301n%40googlegroups.com.