[ADMB Users] crappy software

[H. Skaug]

Hi,

Lack of convergence in a few simulated datasets can cause differences
in average estimates. The best approach would be to
do a paired experiment, i.e. to apply R and ADMB to the same
simulated datasets, and look at within replicate differences. Ideally
Zhang et al should be able to reproduce all simulated datasets.
I have run 300 of Dave's simulations in his ADMB program. Convergence does not
seem to be a problem there, though.

Hans


On 10/31/11, Ben Bolker <bbolker at gmail.com> wrote:
>    Dave, you're a monster.
>    I am working on the R code equivalent but don't have that much time
> to devote to this right now.  I think I have working simulation code, so
> it's really just a matter of putting the pieces together.  If anyone
> wants to take my code and push this farther, please let me know (I have
> updated it some since the stuff I posted yesterday). If the current
> versions of lme4 etc. are still off, it would be interesting to me to
> understand why.
>
>   Ben
>
>
> On 11-10-31 10:45 AM, dave fournier wrote:
>> I did 1000 simulations as in the Zhange et al. paper Table 3 (I think).
>> Thus is for n=500 and tau=2.0.
>> The results were:
>>
>>
>>
>>                 Type 1 error   Num    mean b1    std dev
>>                        %             sims                          b1
>>
>>   5 pt AGHQ 0.064      1000    0.975589  0.0272978
>>
>>   Laplace     0.100      1000    1.00432     0.0396829
>>
>>
>> So the 5 pt AGHQ were almost identical to SAS NLMIXED
>> reported in Zhange et al. It would be nice to see what
>> the R stuff would produce.
>>
>>
>> _______________________________________________
>> 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
>