NA in likelihood calculations

[peter_s]

Hello!

I’ve got a multi-year, multi-site northern flying squirrel trapping data set. It includes four sessions; two seasons of trapping at one site and one season each at two sites. Each site included 8 to 12 trapping grids.

For starters, I’m running simple, single-session SCR models to estimate density for each session and trapping grid (using a trap covariate). The models run fine for three of my sessions, but for the fourth, I get this output:

> secr.LM11_glsa <-secr.fit(LM11_glsaCH, start=c(0.1,-4.5,3.95))
Checking data
Preparing detection design matrices
Preparing density design matrix
Maximizing likelihood...
Eval Loglik D g0 sigma
1 NA 0.1000 -4.5000 3.9500
2 NA 0.1000 -4.5000 3.9500
3 NA 0.1000 -4.5000 3.9500
4 NA 0.1000 -4.5000 3.9500
5 NA 0.1000 -4.5000 3.9500
6 NA 0.1001 -4.5000 3.9500
7 NA 0.1000 -4.4999 3.9500
8 NA 0.1000 -4.5000 3.9504
9 NA 0.1002 -4.5000 3.9500
10 NA 0.1001 -4.4999 3.9500
11 NA 0.1001 -4.5000 3.9504
12 NA 0.1000 -4.4998 3.9500
13 NA 0.1000 -4.4999 3.9504
14 NA 0.1000 -4.5000 3.9508
Completed in 0.73 seconds at 09:47:27 23 Mar 2015
Warning messages:
1: In secr.fit(LM11_glsaCH, start = c(0.1, -4.5, 3.95)) :
multi-catch likelihood used for single-catch traps
2: In secr.fit(LM11_glsaCH, start = c(0.1, -4.5, 3.95)) :
using default buffer width 100 m
> print(secr.LM11_glsa)

secr.fit(capthist = LM11_glsaCH, start = c(0.1, -4.5, 3.95))
secr 2.9.3, 09:47:26 23 Mar 2015

Detector type single
Detector number 572
Average spacing 209.6113 m
x-range 507404 608467 m
y-range 4848766 4851070 m
N animals : 40
N detections : 52
N occasions : 8
Mask area : 250.3466 ha

Etc.


Why is the log-likelihood calculation not working on this? I tried providing fixed and starting parameter values based on output from the other sessions, but I get the same result. My data is formatted consistently across all of my sessions, and I get the “No errors found ” message when I run read.caphist(). So… I’m kind of stuck…

Any suggestions would definitely be appreciated! Thanks!

Peter

[murray.efford]

I'm clutching at straws, but... you don't specify a mask so one is built automatically using the default buffer width (100 m), dividing the x-dimension by 64 and throwing out pixels not close to a trap. When your trap layout is fragmented (lots of empty space between grids) that can result in a mask that is far too coarse relative to the movements of animals about each grid. To get around this, construct the mask specifying a suitable spacing (1/4 trap spacing?) directly, and pass that as an argument to secr.fit.

I hesitate to say this in case Evan is listening, but I mentioned the same issue yesterday on the secrgroup list!

Of course the problem may be something else altogether, but I'm sure its solvable. You do have only a modest number of recaptures in the session, so it's possible one extreme movement is throwing it out. If you really can't solve it then post the data or send to me offline.
Murray

[peter_s]

Thanks very much Murray!

Yes - it turns out that there was a spatial error in my data. Your comment about extreme movement pointed me in the right direction for chasing that down.

Thanks again - I feel a little sheepish for having bugged you for something so trivial, but it had me stumped for several days...

Peter