[R-sig-ME] Unrealistic coefficient values from an MCMCglmm mixed model

[Ben Bolker]

  OK.  In that case: did you check the MCMC diagnostics (effective
sample size, trace plots, autocorrelation, R-hat/Gelman-Rubin
statistics)?  Is your burn-in long enough?  What results do you get if
you fit a comparable model in a frequentist framework (lme4::glmer,
glmmTMB, GLMMadaptive, etc.)?  What about HMC/Stan engines (brms, rstanarm)?



On 2019-04-01 12:45 p.m., Ronan James Osullivan wrote:
> Hi Ben,
> 
> In my model, I have set the intercepts for GeneticType so that B isn't
> relative to A (using "-1"). There are only 2 levels to GeneticType.
> Apologies for the confusion.
> 
> The correct model should read:
> model<- MCMCglmm(LRS~GeneticType*NAO+
>                              GeneticType *Temp-1,
>                           random = ~Year_of_Spawning,
>                           family = "poisson",
>                           data = data,
>                           verbose = TRUE,
>                           nitt = 1010000, burnin = 1000, thin = 1000)
> 
> Regards Temp, it is mean-centred whereas NAO wasn't. I re-ran the model
> with a mean-centred NAO variable and it made no difference.
> 
> Cheers,
> Ronan
> 
>