Closed lionel68 closed 4 years ago
You’re using the log of the posterior mean for the INLA code, but the posterior mean of the log in the inlabru code. These should not be expected to be the same. For a proper comparison I would suggest comparing the posterior medians instead, since those are consistent under transformation.
Wow, thanks for your super fast answer. You are right, when comparing the prediction on the link scale (exp), then all falls into line
Inspired by the code to fit LGCP in INLA from: https://becarioprecario.bitbucket.io/inla-gitbook/ch-spatial.html#sec:pointpat, I simulated some data and tried to compare the predictions form INLA and inlabru.
It seems that there is a slight negative bias for the inlabru predictions as compared to INLA, do you have an idea where this could come from?![grafik](https://user-images.githubusercontent.com/7189850/68758039-56deb980-060d-11ea-951f-ccddfacd6941.png)
The R code used: