[ADMB Users] difference between ADMB-RE and R/mgcv in SEs for smoother coefficients in a GAM fitted by maximum likelihood

[Tim Miller (NOAA Federal)]

Yes, Hans has it right. Just a simple linear function of them. There is 
a second sdreport_vector that gives SEs closer to the "correct" ones 
provided by R/mgcv.

         wrong_transformed_betas = bad_U_F * betas + U_R * smooth_reff;

bad_U_F is the same as U_F except the second column is multiplied by -1.

Tim


On 11/28/2012 10:42 PM, H. Skaug wrote:
> It is the following "sdreport_vector" that depends on both
> fixed (betas) and random effects (smooth_reff)
>
> transformed_betas = U_F * betas + U_R * smooth_reff;
>
> that has unexpected variances.
>
> Hans
>
>
> On Thu, Nov 29, 2012 at 3:46 AM, dave fournier<davef at otter-rsch.com>  wrote:
>> Without bothering to read all that stuff just want to know if this is a std
>> error for
>> a fixed effect a random effect or something else.
>> _______________________________________________
>> Users mailing list
>> Users at admb-project.org
>> http://lists.admb-project.org/mailman/listinfo/users
> _______________________________________________
> Users mailing list
> Users at admb-project.org
> http://lists.admb-project.org/mailman/listinfo/users