ADMB-help,<br><br>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)<br>in an age-based model becomes too large.<br>To get around this I have also fixed it to 1.0, f.e.,<br>
<br>PARAMETER_SECTION<br>  init_bounded_dev_vector log_sel1_coff(1,nages-2,-15.,15.,2)    // nages-2 = 10  ,, age-12<br>  init_bounded_dev_vector log_sel2_coff(1,nages-2,-15.,15.,2)    <br>  init_bounded_dev_vector log_sel3_coff(1,nages-2,-15.,15.,2)<br>
...............<br><br>for (i=1;i<=nyrs;i++)<br>  { <br>  if (i<=16)<br>  {  for (j=1;j<nages-2;j++)                                  <br>    {<br>    log_sel1(j)=log_sel1_coff(j);<br>    }<br>  }<br>  else<br>  {<br>
  if (i<=35)<br>  {  for (j=1;j<nages-2;j++)                                  <br>    {<br>    log_sel2(j)=log_sel2_coff(j);<br>    }<br>  }<br>.................................<br><br>    log_sel1(nages-2)=0;<br>    log_sel1(nages-1)=0;<br>
    log_sel1(nages)=0;<br><br>    log_sel2(nages-2)=0;<br>    log_sel2(nages-1)=0;<br>    log_sel2(nages)=0;<br>..................<br><br>From ".cor" file:<br><br> index   name               value<br>     1   log_sel1_coff    -2.2926e+00<br>
     2   log_sel1_coff    -1.3813e+00<br>     3   log_sel1_coff    -1.0400e+00<br>     4   log_sel1_coff    -1.1053e+00<br>     5   log_sel1_coff    -1.1191e+00<br>     6   log_sel1_coff    -1.1294e+00<br>     7   log_sel1_coff    -1.0320e+00<br>
     8   log_sel1_coff    -8.5347e-01<br>     9   log_sel1_coff    -8.0637e-01<br><b>    10   log_sel1_coff     1.0760e+01</b><br>    11   log_sel2_coff    -3.2898e+00<br>    12   log_sel2_coff    -1.6544e+00<br>    13   log_sel2_coff    -1.0402e+00<br>
    14   log_sel2_coff    -7.6604e-01<br>    15   log_sel2_coff    -8.2858e-01<br>    16   log_sel2_coff    -9.0862e-01<br>    17   log_sel2_coff    -9.5011e-01<br>    18   log_sel2_coff    -9.6055e-01<br>    19   log_sel2_coff    -8.8192e-01<br>
<b>    20   log_sel2_coff     1.1280e+01</b><br>..............................................<br><br>I have tried to shorten estimates for selection so that instead of estimating (nages-2) <br>I end up with (nages.-3) parameters:<br>
<br>PARAMETER_SECTION<br>  init_bounded_dev_vector log_sel1_coff(1,nages-3,-15.,15.,2)    <br>  init_bounded_dev_vector log_sel2_coff(1,nages-3,-15.,15.,2)    <br>  init_bounded_dev_vector log_sel3_coff(1,nages-3,-15.,15.,2)    <br>
.................................<br><br>for (i=1;i<=nyrs;i++)<br>  { <br>  if (i<=16)<br>  {  for (j=1;j<nages-2;j++)                                  <br>    {<br>    log_sel1(j)=log_sel1_coff(j);<br>    }<br>  }<br>
  else<br>  {<br>  if (i<=35)<br>  {  for (j=1;j<nages-2;j++)                                  <br>    {<br>    log_sel2(j)=log_sel2_coff(j);<br>    }<br>......................................<br><br>    log_sel1(nages-2)=0;<br>
    log_sel1(nages-1)=0;<br>    log_sel1(nages)=0;<br><br>    log_sel2(nages-2)=0;<br>    log_sel2(nages-1)=0;<br>    log_sel2(nages)=0;<br>..................<br><br>Estimates by this second approach are even more flawed:<br>
<br>index   name               value    <br>     1   log_sel1_coff    -5.3813e-01<br>     2   log_sel1_coff    -5.1953e-01<br>     3   log_sel1_coff    -3.2202e-01<br>     4   log_sel1_coff    -2.2866e-01<br>     5   log_sel1_coff    -1.1985e-01<br>
     6   log_sel1_coff    -1.4601e-02<br><b>     7   log_sel1_coff     2.0729e-01<br>     8   log_sel1_coff     5.7745e-01<br>     9   log_sel1_coff     9.5804e-01<br>    10   log_sel2_coff    -1.2548e+00<br></b>    11   log_sel2_coff    -8.8436e-01<br>
    12   log_sel2_coff    -3.8221e-01<br>    13   log_sel2_coff     1.0742e-01<br>    <b>14   log_sel2_coff     2.8319e-01<br>    15   log_sel2_coff     3.8595e-01<br>    16   log_sel2_coff     4.4048e-01<br>    17   log_sel2_coff     5.7487e-01<br>
    18   log_sel2_coff     7.2942e-01<br>....</b>.....................................................<br><br>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?<br>
On the other hand selection in such an old age group (10=age-12 ) is quite noisy and therefore difficult to estimate<br>(and probably not so bad to fixed it to 1.0)<br><br>Tips and comments are welcome,<br>Thanks in advance,<br>
Luis<br><br>