time autocorrelation

[Sabira Smaili]

Hi

I am currently working on modeling cumulative vaccination rates by week using an INLA approach, and I would like to account for spatio-temporal interactions in my analysis. My dataset covers 1.5 year, and I aim to capture both spatial and temporal dependencies.

My question is to address autocorrelation over time, would specifying the weeks as an AR process be sufficient, or should I consider a more complex structure?
For context, my goal is to assess the dynamics of vaccination uptake across regions and over time. I want to ensure that the model accurately reflects the temporal dependencies without overfitting.

If you have any recommendations for model specification, relevant literature, or examples, I would greatly appreciate your insights.

Thank you in advance for your time and guidance.

Sab

[Helpdesk (Haavard Rue)]

Hi,

I do not think there is 'one' answer here. yes, AR processes are key to try,
like model="ar1" or model="ar",order=2, for 2nd order, or higher.

I guess you aim to try to add weekly additive time-effect, meaning that the
time-effect is constant for all locations. you can also try, if needed, a space
varying time-effect, where the time is AR1 or higher order, but the effect of
this depends on location. this will cost more computationally...

I think you just have to try and check how simple model you can get away with.

Best
H

> --
> 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/4ecf07e1-3c63-48f5-a309-5a4094fe0740n%40googlegroups.com
> .

--
Håvard Rue
he...@r-inla.org