Discrepancy in Spatiotemporal Model Outcome

[Tatiana Pineda Portella]

Hello,

I ran this model several times about two years ago:

model_final<- inla(casos ~ covariates + f(idarea, model = "bym", graph = g)+ f(idtime, model = "iid")+ f(idtime1, model = "rw1")+ f(idarea1,model="iid", ##type two interaction group=idtime.int, control.group=list(model="rw1")), family = "nbinomial", data = mapa1, E=E, control.predictor = list(compute = TRUE), control.compute=list(dic=TRUE, waic=TRUE, cpo=TRUE, return.marginals.predictor=TRUE))

Now I run the same model with the same data and get completely different results. 

What could have gone wrong?  Has anyone else experienced this issue? Any insights would be greatly appreciated.

Sincerely,

Tatiana

[Abi Riley]

Hi Tatiana,

Have you tried setting a seed in your code? (set.seed(123)

I had this issue, I think it's because INLA is an approximation method and doesn't always converge in the same way. You could check the sort of INLA trace plots or try a more informative prior to guide it slightly.

Maybe someone has a more mathematical answer, but these worked for me!

Thanks,

Abi

[FredrikLAa]

Hi,

Have you updated INLA since last time? If "yes", have you tried using inla.mode = 'classic' ? Now the default is the new computational mode. I do not think they should give too different results, but classic was the default before. The INLA developer team should be able to tell more. 

[Moussa Tao]

Hello, how to verify the efficience of  model us INLA ans SPDE?

--
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/0ec02459-c30a-4c2f-82bc-399060669652n%40googlegroups.com.