MCMCsummary vs. PROC MCMC in SAS

[sharontuck]

Hi RStudio Community,

I have run a Bayesian GLMM model in SAS (proc MCMC) and have aquired the posterior means and 95% Highest Posterior Density (HPD) intervals. I'm looking to validate this in R but I am struggling to get my results to match (datasets are consistent with each other). I have created models in R using 'brm' and 'jags.model' and then used 'MCMCsummary' to get the HPD intervals, but I just can't get my SAS and R results to match.

Any pointers would be much appreciated. TIA.

[nirgrahamuk]

how different are they ?

[sharontuck]

I wouldn't say they're hugely different but large enough for me not to consider the results validated.

I just wondered if anyone was aware of any examples using both SAS and R and provided the same (or near enough) the same results?

[nirgrahamuk]

How different are the results when you run SAS proc MCMC and compare to another run of proc MCMC?
Are you controlling with a seed to ensure no variation ?

if so , try different seeds and first get a feeling for the variance introduced by the pure random aspects of what you are doing. This has to be a first step before comparing output to any other system or approach.

[sharontuck]

Great question. I am using the same seed in both SAS and R, but I shall do as you suggest. I had assumed that using the same seed would have ensured that both sets of results would have at least remained similar...