[R-sig-ME] Constraining error variance to 0 in a mixed model.

[Rolf Turner]

On 1/12/19 1:40 AM, Mollie Brooks wrote:

> Hi Rolf,
> 
> Thanks for the enthusiasm. I believe the ability to force the variance 
> into the RE in glmmTMB was Kasper Kristensen’s idea.
> 
> I had a look at your example and it seems that it could be a false 
> positive coming from nlminb. The maximum gradient component is 2.4e-6, 
> so it looks to me like the model converged.
> 
> In the future, the glmmTMB developers will reconsider the convergence 
> check that relies on nlminb’s convergence codes.
> 
> if (!is.null(fit$convergence) && fit$convergence != 0) warning()

Thanks. That's encouraging.

cheers,

Rolf

-- 
Honorary Research Fellow
Department of Statistics
University of Auckland
Phone: +64-9-373-7599 ext. 88276