Apologies for the long data strings and code. I would just like people to be able to reproduce my analysis. I am having trouble getting what I consider to be good estimates of the parameters in a mixed-effects model run in stan. df <- data.frame(id = c(1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 3, …

A couple of things to try: Instead of 95% I’d use 50% is more computational stable and makes for a better comparison. Are the priors the same in the stan and brms model. In brms you can get_prior and set_prior. I think nmle essentially uses uniform priors?

Thank you Ara Winter. If it is the priors making all the difference then why are the brms and nlme models so similar? brms does not use uniform priors as default. I chose 95% because I wanted to emulate the intervals returned by brms and nlme.

Not sure yet but can you make the brms priors match the stan priors and then we can go from there. I’ll try and get the code copied over into R on my side tomorrow for a run through.

Thank you @Ara_Winter I really appreciate the help. I actually found a statement in the ‘brms_overview’ vignette that said “The default prior is an improper flat prior over the reals” but when I run stancode(brmsMod) this seems not to be the case (though I must confess the stancode is well beyond m…

Let me see what I can do. I’ll get this up and running this morning. Hopefully others folks will pile in on the code end of thing.

A couple of things looking at your model calls. I’d recommend 4 chains, 2000 iter, and 500 for the warmup to start. When I run your Stan model I end up with a number of divergent transitions.

@Ara_Winter I tried that but the width of the 95% HPDIs do not change

Well the divergent transitions tells me there is something up with the model or data. I am not sure what yet. Sorry haven’t had much time to dive into the details.

[image] Divergent transitions - a primer General This post is outdated, see Runtime warnings and convergence problems for an improved and expanded version So you got: Warning: There were XXXX divergent transitions after warmup. What does it mean? What to do? Divergent…

Estimates from hierarchical longitudinal analysis in stan much less certain than estimates from same model in brms

Modeling

Ara_Winter September 24, 2020, 5:42pm 10

Topic		Replies	Views
Trend analysis: Full stan model does not match resutls from same model in brms Modeling rstan , techniques	5	567	October 1, 2020
Hierarchical random intercept model - Stan vs brms Modeling specification , brms	5	231	May 6, 2025
Reproducing brms results in rstan yields different estimates brms	7	916	July 30, 2020
Translating/comparing brms parameter estimates to rstan estimates brms rstan	9	1137	May 12, 2020
One chain not moving: sampling for complex nonlinear mixed effects model brms techniques , fitting-issues , specification , ecology	14	3640	November 6, 2020

Estimates from hierarchical longitudinal analysis in stan much less certain than estimates from same model in brms

Related topics