[ADMB Users] penultimate estimate of selection too high

[Luis Ridao]

ADMB-help,

I have noticed that estimating selection for the last age group (I mean for
the last age group from which selection is fixed=1.0)
in an age-based model becomes too large.
To get around this I have also fixed it to 1.0, f.e.,

PARAMETER_SECTION
  init_bounded_dev_vector log_sel1_coff(1,nages-2,-15.,15.,2)    // nages-2
= 10  ,, age-12
  init_bounded_dev_vector log_sel2_coff(1,nages-2,-15.,15.,2)
  init_bounded_dev_vector log_sel3_coff(1,nages-2,-15.,15.,2)
...............

for (i=1;i<=nyrs;i++)
  {
  if (i<=16)
  {  for (j=1;j<nages-2;j++)
    {
    log_sel1(j)=log_sel1_coff(j);
    }
  }
  else
  {
  if (i<=35)
  {  for (j=1;j<nages-2;j++)
    {
    log_sel2(j)=log_sel2_coff(j);
    }
  }
.................................

    log_sel1(nages-2)=0;
    log_sel1(nages-1)=0;
    log_sel1(nages)=0;

    log_sel2(nages-2)=0;
    log_sel2(nages-1)=0;
    log_sel2(nages)=0;
..................

>From ".cor" file:

 index   name               value
     1   log_sel1_coff    -2.2926e+00
     2   log_sel1_coff    -1.3813e+00
     3   log_sel1_coff    -1.0400e+00
     4   log_sel1_coff    -1.1053e+00
     5   log_sel1_coff    -1.1191e+00
     6   log_sel1_coff    -1.1294e+00
     7   log_sel1_coff    -1.0320e+00
     8   log_sel1_coff    -8.5347e-01
     9   log_sel1_coff    -8.0637e-01
*    10   log_sel1_coff     1.0760e+01*
    11   log_sel2_coff    -3.2898e+00
    12   log_sel2_coff    -1.6544e+00
    13   log_sel2_coff    -1.0402e+00
    14   log_sel2_coff    -7.6604e-01
    15   log_sel2_coff    -8.2858e-01
    16   log_sel2_coff    -9.0862e-01
    17   log_sel2_coff    -9.5011e-01
    18   log_sel2_coff    -9.6055e-01
    19   log_sel2_coff    -8.8192e-01
*    20   log_sel2_coff     1.1280e+01*
..............................................

I have tried to shorten estimates for selection so that instead of
estimating (nages-2)
I end up with (nages.-3) parameters:

PARAMETER_SECTION
  init_bounded_dev_vector log_sel1_coff(1,nages-3,-15.,15.,2)
  init_bounded_dev_vector log_sel2_coff(1,nages-3,-15.,15.,2)
  init_bounded_dev_vector log_sel3_coff(1,nages-3,-15.,15.,2)
.................................

for (i=1;i<=nyrs;i++)
  {
  if (i<=16)
  {  for (j=1;j<nages-2;j++)
    {
    log_sel1(j)=log_sel1_coff(j);
    }
  }
  else
  {
  if (i<=35)
  {  for (j=1;j<nages-2;j++)
    {
    log_sel2(j)=log_sel2_coff(j);
    }
......................................

    log_sel1(nages-2)=0;
    log_sel1(nages-1)=0;
    log_sel1(nages)=0;

    log_sel2(nages-2)=0;
    log_sel2(nages-1)=0;
    log_sel2(nages)=0;
..................

Estimates by this second approach are even more flawed:

index   name               value
     1   log_sel1_coff    -5.3813e-01
     2   log_sel1_coff    -5.1953e-01
     3   log_sel1_coff    -3.2202e-01
     4   log_sel1_coff    -2.2866e-01
     5   log_sel1_coff    -1.1985e-01
     6   log_sel1_coff    -1.4601e-02
*     7   log_sel1_coff     2.0729e-01
     8   log_sel1_coff     5.7745e-01
     9   log_sel1_coff     9.5804e-01
    10   log_sel2_coff    -1.2548e+00
*    11   log_sel2_coff    -8.8436e-01
    12   log_sel2_coff    -3.8221e-01
    13   log_sel2_coff     1.0742e-01
    *14   log_sel2_coff     2.8319e-01
    15   log_sel2_coff     3.8595e-01
    16   log_sel2_coff     4.4048e-01
    17   log_sel2_coff     5.7487e-01
    18   log_sel2_coff     7.2942e-01
....*.....................................................

One question is whether is right or not to let ADMB estimate this last age
group (and because of its high value) and set to 1?
On the other hand selection in such an old age group (10=age-12 ) is quite
noisy and therefore difficult to estimate
(and probably not so bad to fixed it to 1.0)

Tips and comments are welcome,
Thanks in advance,
Luis
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.admb-project.org/pipermail/users/attachments/20110110/83699da9/attachment.html>