confusing error running robust models with rmark

[rmark4salamanders]

I am running a Robust model on a capture-mark-recapture dataset with animal size class as a time varying individual covariate and precipitation as a time varying covariate. I have three precipitation datasets because I am comparing 3 ways of summarizing precipitation data for each primary sampling period. most of my models run just fine, but when I make survival dependent on size+pt, that is, size and the pt precipitation covariate, I get the following error:

Code: Select all

STOP ERROR
MARK did not run properly.  If error message was not shown, re-run MARK with invisible=FALSE

Model output is not available

Error in wtable[i, cols] = real[pim[i, cols]] :
  number of items to replace is not a multiple of replacement length


I have tried re-running with invisible=FALSE, and I get the same error message.

In the out file there are a number of warnings and errors:

Code: Select all

Number of function evaluations was 97 for 38 parameters.
 Time for numerical optimization was 136.36 seconds.
 -2logL { S(~siz + pt)Gamma''(~1)Gamma'(~1)p(~1)c(~1)f0(~session) } = -0.0000000   
 Penalty { S(~siz + pt)Gamma''(~1)Gamma'(~1)p(~1)c(~1)f0(~session) } = -0.0000000   

   * *  WARNING  * *   Numerical overflow occurred during optimization of this model.

   * *  WARNING  * *   Numerical underflow occurred during optimization of this model.
     IEEE flag status at end of optimization:
     overflow       T
     divide by zero F
     invalid        T
     underflow      T
     inexact        T

 Gradient { S(~siz + pt)Gamma''(~1)Gamma'(~1)p(~1)c(~1)f0(~session) }:
   5.207684      1.653186     0.6246231      1.487248      1.469445   
   101.0596     -1.591131      3.244535     0.2256693    -0.1612746   
 -0.1852874    -0.1620151     -1.231431    -0.9543933    -0.8142962   
  0.8527453    -0.1948382     0.1503171    -0.3617251    -0.2747045   
 -0.3235642    -0.3200927    -0.1186596    -0.5451706E-01 0.2435025   
   2.827649     0.3434547E-01-0.3309865    -0.7997107     0.5755665   
  0.3605221E-01-0.2039876    -0.1653346E-01-0.5212007     -2.875855   
   3.754608      1.259797     0.4192307   
 ERROR -- Numerical convergence never reached, with maximum G =   101.0596   
 Final Parameter Values:
   0.985968-304  0.985968-304  0.985968-304  0.985968-304  0.985968-304
   0.985968-304  0.985968-304  0.985968-304  0.985968-304  0.985968-304
   0.985968-304  0.985968-304  0.985968-304  0.985968-304  0.985968-304
   0.985968-304  0.985968-304  0.985968-304  0.985968-304  0.985968-304
   0.985968-304  0.985968-304  0.985968-304  0.985968-304  0.985968-304
   0.985968-304  0.985968-304  0.985968-304  0.985968-304  0.985968-304
   0.985968-304  0.985968-304  0.985968-304  0.985968-304  0.101423+305
   0.101423+305  0.101423+305  0.101423+305  0.101423+305  0.101423+305
   0.101423+305  0.101423+305  0.101423+305  0.101423+305  0.101423+305
   0.101423+305  0.101423+305  0.101423+305  0.101423+305  0.101423+305
   0.101423+305  0.101423+305  0.101423+305  0.101423+305  0.101423+305
   0.101423+305  0.101423+305  0.101423+305  0.101423+305  0.101423+305
   0.101423+305  0.101423+305  0.101423+305  0.101423+305  0.101423+305


  INPUT --- proc stop;

 ERROR -- Numerical convergence never reached.


what really bemuses me is that this model which returns the error message is almost identical to other models that will run properly, the only difference is in the precipitation dataset being used. that is, i have other models which have survival dependent on size+po and size+pf where pf and po are my other precipitation datasets and these models run perfectly. I have built each precipitation dataset exactly the same way, each has the same number of rows and columns, the only difference between them is in the prefix of the column headings and of course the precipitation values.

Additionally, the pt precipitation dataset which is being used in this model doesn't cause problems in other simpler models such as when i make survival dependent on body size alone and other parameters dependent on pt alone.

I am running R version 3.2.3, Rmark version 2.1.14, and Mark version 8.1

Any help deciphering these errors and solving this problem would be much appreciated, thank you in advance!

[jlaake]

As you can see in the output file, the optimization code in MARK went awry based on the final parameter values it gave. Not sure how you are using animal class size as a time varying covariate because it seems like it would not be known when the animal was not captured. Anyhow, that is not your question. The best way you can help out the optimization code is to provide initial values. The easiest way is to specify the argument initial=model where model is the name of a model that is nearly the same and converged. It will pull off relevant starting values from it and set any other parameters to (eg pt) to 0.

[rmark4salamanders]

Thanks for your help Jeff, that fix worked really well!