Visual Fortran Run-time Error

[Miina Kovanen]

Hi!

I'd need to get the profile likelihood estimates for some beta-parameters that seem to be at the boundary (large beta estimate value, SE=0), in my analysis using Nest Survival Models in "rugged" telemetry data. However I get an error message below - could somebody please comment how to find out what's wrong? I am using the MARK version 5.1, build 2600.

Visual Fortran Run-time error
Forrtl: severe (161): Program Exception – array bounds exceeded
Image PC Routine Line Source
MARK.EXE 004E1E3B ESTMAT 1124 estmat.for
MARK.EXE 005A5D9C MARK 280 mark.for
MARK.EXE 0066F369 Unknown Unknown Unknown
MARK.EXE 0065830B Unknown Unknown Unknown
Kernel32.dll 7C80B729 Unknown Unknown Unknown

My data set consists of 253 adult Black Grouse females that were followed by radio telemetry, troughout the seasons, on monthly basis. Study period is 3 years - we aim at detecting possible seasonal variation in survival of females from two age classes. The data is analysed according to the Nest Survival Model using coding for weekly survival rates, since the study area & field conditions in coniferous forests prevent more frequent & regular observation schedule.

I would very much appriciate your comments!

[gwhite]

Miina:
You can't get profile likelihood confidence intervals on beta parameters. If you send me you MARK files, I see what is causing this error. Thanks.

Gary

[cooch]

Hi!
....I am using the MARK version 5.1, build 2600.

Quick addendum - it is always helpful when reporting these problems to report something about which version of MARK you are using. Thanks for trying to do so.

However, what you reported is not the MARK version number, but information about the OS build. What you want to report is

1. the operating system in full (e.g., Windows XP SP 3)

2. time stamp for MARK_INT.EXE (currently 11/07/2009)

3. time stamp for MARK.EXE (currently 10/30/2009)

[Miina Kovanen]

Thank you for your reply!

Gary, yes, sorry for mixing things up. The real estimates for survival seem to be at the boundary during some seasons of interest in my study, and certain coefficients corresponding for these seasons get large values and SE=0. Thank you for offering your help, I’ll e-mail you. I’m also nervous if it might be just that too much sparseness in the data during some seasons causes troubles.

Cooch, thanks, here’s the OS build I was using:
1. Windows XP professional (build2600.xpsp_sp3_gdr.090804-1435: Service Pack 3)
2. MARK_INT.EXE 19.07.2009
3. MARK.EXE 27.7.2009

Generally, thank you for keeping this forum going. It is very helpful during many steps (and missteps ) of learning to use MARK.

[dhewitt]

Although the capture-recapture gurus are certainly aware of this, I have the sense that many practitioners do not appreciate the problems caused by sparse data and boundary estimates (worst when both are at issue). Inference becomes troublesome. Is the boundary estimate reasonable or is it caused by sparse data? How does this affect other estimates from the likelihood, as in a time-varying model? Model selection does not always resolve these issues, as models with boundary estimates can be highly favored. Any optimization algorithm cannot always generate the answer you hope is in the data. And, as Gary notes in his Help files and I have found through experience, profile likelihood intervals do not help with boundary estimates.

Sometimes a Bayesian approach with an "uninformative prior" will generate a nicer estimate by yanking the estimate off of the boundary (with decent precision at that), but a comparison of the two outputs forces one to make a philosophical decision.

Ultimately you return to the realities of the density of your data, which may not be "enough" for what you had hoped. I don't think there is enough guidance on this issue "out there".