I can estimate multilevel regression and post-stratification (MRP) using Stan. however, HMC is too slow. it will takes several hours to retry the codes. so, can INLA estimate MRP? I believe INLA should be faster than HMC.
Thank in advance.
The answer must be yes, as long as your problem conforms to the restricted class of models that can be estimated using INLA; however, it is probably worth investigating why HMC is “slow” – in many cases it should be scalabe to large data sets and complex models.
Have you tried the same model with other methods? I don’t have a lot of hands-on experience with INLA, but as far as I can remember you will need to obtain a MAP estimate to build the gaussian approximations, so you could check if point estimates like MAP, MLE, or simpler samplers get you ballpark estimates in a reasonable time as a sanity check.