Issue with standard priors for Gompertz model

[OBaerenbold-Lumanity]

Hi Gianluca,

Thanks for the great package!

There appears to be an issue with the standard priors for the Gompertz model in version 1.1.2. Line 307 and following in utils_fit_models.R and line 57 in Gompertz.stan define the prior for alpha as Gamma(0,5) which causes an error in Stan due to the shape parameter needing to be strictly greater than 0.

I suspect either line 328 should include "gom" while line 330 shouldn't which leads to a Gamma(0.1,0.1) prior, or alternatively alpha is supposed to be unrestricted and have a Normal(0,5) prior. As a quick fix I currently just run fit.models with 'priors=list("gompertz"=list(a_alpha=0.1,b_alpha=0.1))' instead.

Best,
Oliver

[giabaio]

Thank you for this. Are you on the most recent version of survHE? I have made some substantial refactoring a while back (see here) and have split the package into 3 separate bits. If you install the version on the devel branch of this GitHub repo, you get the frequentist version of the models (+ the basic plotting/printing/summary etc). If you want to do the HCM analysis, you need to install separately the module survHEhmc (from the relevant branch).

If you look at line 228 of https://github.com/giabaio/survHE/blob/hmc/R/runHMC.R, you can see that the prior for the Gompertz model is fixed and correct...

Your proposed solution also would work --- but I would advice switching to the most recent version.
Hope this helps!

[OBaerenbold-Lumanity]

Thanks, that's very helpful! I'll give it a try.

Just to understand: The version on CRAN (1.1.2) is the main branch on the repository and the refactored version is split onto the three branches as described in your linked blog?

[giabaio]

Yes. The refactored version is 2.0.

Thanks