Residual spatial fields

[Tim Meehan]

Hi all,

I am working on a model in inlabru with count data sampled on 9 

different occasions. I'd like to demonstrate a general spatial pattern 

in the data using a spatial field. But I'd also like to account for variation in the spatial 

field across sampling occasions. So I am considering having two 

spatial field terms in my model statement, one for a 'temporally

averaged' spatial field and one grouped version for the 'residuals' 

around the temporally averaged term (see below). I suppose I could 

just do a single grouped field, but I am hoping that the way I am 

formulating the model below will make it easier to get predictions 

for the average field. Does this approach make sense?

cmp1 <- count ~ 

    # global intercept

    Intercept(1) + 

    # 'average' zero-centered spatial field?

     field(geometry, model = spde1) +  

    # 'residual' zero-centered spatial fields?

     field_residuals(geometry, model = spde2, group = date_int,
                  control.group = list(model = 'iid', hyper = pc_prior))

# given

spde1 <- spde2 <- inla.spde2.pcmatern(
  mesh = mesh1,
  prior.range = c(10000, 0.5),
  prior.sigma = c(1, 0.01)
)

pc_prior <- list(prec = list(prior = "pc.prec", param = c(1, 0.01)))

predict(object=fit1, newdata=pix1, formula=exp(Intercept + field))

[Finn Lindgren]

Hi Rim,

Yes, in principle this makes sense. You just need to look out for potential nonidentifiability (the common field is confounded with the average of the residual fields). With 9 time points, this is probably ok, but you should check if the average of the residual fields has a noticeable spatial pattern. If it does, you might want to include that average pattern in the interpretation of the posterior common pattern.

Finn

On 22 Aug 2024, at 17:56, Tim Meehan <tme...@gmail.com> wrote:

Hi all,

--
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 on the web, visit https://groups.google.com/d/msgid/r-inla-discussion-group/b7e77461-fb54-4f2c-980a-5fcf67574883n%40googlegroups.com.

[Tim Meehan]

Thanks, Finn!