www.phidot.org

Hi everyone,
Sorry if this is a bit basic. I'm am working with 6 single-capture trapping grids. When I estimate density for each individual grid, site learned response (g0 ~ 1, sigma ~ k) most commonly comes out as the top model. I would like to estimate the average density for all trapping grids. I set up each replicate as a session in a single capthist object and pooled:

secr01a<- secr.fit(myCH01,model = list(D ~ 1, sigma ~ k),buffer = 300, detectfn = 2, method = "BFGS", trace = TRUE).

The response was:
Checking data
Preparing detection design matrices
Error in temp[1:dimx[1], 1:dimx[2]] <- x :
incorrect number of subscripts on matrix
In addition: Warning message:
In secr.fit(myCH01, model = list(D ~ 1, g0 ~ K), buffer = 300, detectfn = 2, :
multi-catch likelihood used for single-catch traps

Site learned (k) and site transient (K) are the only detection parameters that give the message and it's not clear to me why. Learned (b), and animal x site learned (bk) run fine. Is there a way around this or another alternative? Any assistance would be greatly appreciated.

Thanks,
G

Looks like a bug in this model with multi-session data. I get:

secr.fit(ovenCH, buffer=300, model=g0~K)
Checking data
Preparing detection design matrices
Error in temp[1:dimx[1], 1:dimx[2]] <- x :
incorrect number of subscripts on matrix

I'll get onto it in the next day or so.
Murray

Hi Murray,
Thanks for this. I'm was worried it was something more fundamental that I was missing. It's much appreciated!

No, it was entirely my doing! I've fixed the bug in my working copy of the release due in the next few weeks (secr 2.3.3 ). If you're using Windows, and send me your address offline, I can send a link to a zip file that will install the beta version via the R menu option 'Packages | Install package(s) from local zip files'.

Obviously, these models haven't been much used - they're slow to fit, and there's seldom a biological reason to expect sites to 'behave' like this - so proceed with caution. On the other hand, it would be interesting to see an example where a 'learned' site response really made a difference, and that could not be explained e.g., by inherent differences among sites.
Murray