Pradel Lambda models do not converge

[Tom.B]

Hi all,

I've just started working with MARK and seem to have walked immediately into a wall. I promise I have attempted to figure things out myself by (continuing to) RTFM and searching the forums, but it looks like I've missed something pretty crucial.

My dataset consists of Robust Design style mark recapture data of a single mouse species across 3 sites and 12 primary sessions (each with 4 or 5 secondary sessions), totaling ~4k captures of ~700 individuals over 3 years.

I'm primarily interested in examining differences in survival between sexes and grids, and differences in fecundity between grids.

I got as far as getting my data into RMark (file here, if anyone wants a play - https://www.dropbox.com/s/xyeaxwzkcmyo7ng/mouseb.txt) and tested it using a null 'Robust' model, which seems to work;

Code: Select all

mouseRMb=import.chdata("mouseb.txt", header=TRUE)

mouse.process=process.data(mouseRMb, model="Robust", begin.time=1, groups=c("grid","sex","agemarked."),
   time.intervals=c(0,0,0,123,0,0,0,0,74,0,0,0,0,83,0,0,0,0,68,0,0,0,0,125,0,0,0,0,76,0,0,0,0,82,0,0,0,0,72,0,0,0,0,125,0,0,0,0,69,0,0,0,0,94,0,0,0,0))

mouse.dll=make.design.data(mouse.process)

test1=mark(mouse.process, mouse.dll)

However, any of the Pradel models I've tried all seem to fail, with an "ERROR-- Numerical convergence never reached" message from Mark.

eg.

Code: Select all

mouse2.process=process.data(mouseRMb, model="RDPdfClosed", begin.time=1, groups=c("grid","sex","agemarked."),
   time.intervals=c(0,0,0,123,0,0,0,0,74,0,0,0,0,83,0,0,0,0,68,0,0,0,0,125,0,0,0,0,76,0,0,0,0,82,0,0,0,0,72,0,0,0,0,125,0,0,0,0,69,0,0,0,0,94,0,0,0,0))

mouse2.dll=make.design.data(mouse2.process)

test2=mark(mouse2.process, mouse2.dll)

"RDPdGHuggins" will run, but all the beta estimates are identical with 0 error.

I had seen a suggestion to use initial values from simpler models, but given that the null model won't run I'm a bit stumped and suspect there's a deeper issue that I'm failing to appreciate.

I'm probably falling down simply because I'm not sufficiently up to speed on the theory, which I am trying to correct, but if anyone could spare a minute to have a look and perhaps point me in the right direction I'd be extremely grateful.

Cheers,
Tom

[jlaake]

Tom-

Well done providing everything I needed to figure this out. I was able to get it to converge by dividing your time.interval values by 365. You can use any unit you want but when Phi is close to 1 (or any parameter near a boundary) convergence problems can arise. With days as unit it is a daily survival rate which will be close to 1. By dividing by 365 I changed it to a yearly survival rate. You can use any scale value that makes sense for your species.

regards --jeff

[Tom.B]

Jeff that makes perfect sense - seems obvious now you've said it!

Converted time units to months and everything seems to be running well for now, thank you.