[ADMB Users] Another sdreport mystery

[dave fournier]

this appears to be a rare example where looking at the eigenvalues of 
the Hessian at
the putative solution is informative.  Check the file *eva.  You will 
see that you have
a minimum eigenvalue around 1.e-9.  If you use the -eigvec option to run 
the model
you will also get the corresponding eigenvectors.  Unfortunately the 
format is difficult
to deal with but you can see in the second (or whatever) line the 
eigenvalues in the
same order as the eigenvectors. So I think you want the 10'th eigenvector
which looks like

  [1] -2.886751e-01 -2.886751e-01  2.886751e-01  2.886751e-01 2.886751e-01
  [6]  2.886751e-01  2.886751e-01  2.886751e-01  2.886751e-01 2.886751e-01
[11]  2.886751e-01  2.886752e-01 -1.362096e-10 -6.788726e-10 2.781558e-10
[16]  6.526706e-10  3.567118e-10 -1.684850e-09 -2.326444e-09 1.962649e-09
[21]  9.708643e-11 -2.305247e-10 -3.908958e-10  5.738751e-08 2.368356e-10
[26] -1.728703e-10 -2.676910e-09 -1.892710e-09  3.860461e-10 1.389939e-09
[31] -4.481481e-09 -4.928454e-10  8.555867e-10 -1.641832e-09 -4.746479e-09
[36] -2.533100e-09  1.399407e-09 -1.617055e-10 -4.510348e-09 -7.339353e-10
[41]  0.000000e+00  0.000000e+00  3.687877e-10  0.000000e+00

rescaling it looks like

  [1] -1.000000e+00 -1.000000e+00  1.000000e+00  1.000000e+00 1.000000e+00
  [6]  1.000000e+00  1.000000e+00  1.000000e+00  1.000000e+00 1.000000e+00
[11]  1.000000e+00  1.000000e+00 -4.718439e-10 -2.351684e-09 9.635600e-10
[16]  2.260917e-09  1.235686e-09 -5.836491e-09 -8.059040e-09 6.798814e-09
[21]  3.363172e-10 -7.985610e-10 -1.354103e-09  1.987962e-07 8.204225e-10
[26] -5.988403e-10 -9.273087e-09 -6.556540e-09  1.337303e-09 4.814890e-09
[31] -1.552431e-08 -1.707267e-09  2.963839e-09 -5.687474e-09 -1.644228e-08
[36] -8.774917e-09  4.847688e-09 -5.601642e-10 -1.562430e-08 -2.542426e-09
[41]  0.000000e+00  0.000000e+00  1.277518e-09  0.000000e+00

so it is telling yoiu that if  x is your vector of parameters then
the model is almost completely insensitive to

    x(1)+x(2) -x(3) - ... -x(12)

that should tell you something about the model

Also parameter 41 corresponse to the next small eigenvalue that is the 
first logits value

# logits:
  -28.3793844393 -8.79090361669 -2.88304855852

you get trouble with logits when the value is near -infinity or +infinity
which -28.3  is.

Be nice if someone made the -eigvec output easier to use.