Nest Survival Variance Components Error

[DTI]

Hopefully this question hasn't been asked already (though after RTFM and searching the forums, I haven't come across a mention of this problem yet).

I'm trying to fit random effects models to nest survival data in MARK. Data was collected over 23 years, and there are 76 encounter occasions in the dataset (thus, the longest "field season" of the 23 years spanned 76 days). Each year of study was classified as one of two types: "Pre-exodus" (given a value of 0), or "Post-exodus" (given a value of 1).

I'm interested in fitting a random effects model in which the intercept differs depending on the type of year. Thus, I have an individual covariate called “exodus” that is coded 0 for nests in the pre-exodus years, and 1 for nests in the post-exodus years.

To ensure that my .inp file was working correctly, I started by running a fixed-effect full time model S(t), a fixed-effect intercept model S(.), and a fixed-effect exodus model S(exodus). All three returned with no errors (though there were some non-convergence issues in several time intervals for the S(t) model).

Next, I ran an intercept-only random effects model and a linear trend random effects model, both of which returned with no errors. Finally, I tried to run a random-effects model where the intercept differs as a function of “exodus.” I retrieved S(t), opened the Variance Components/Real Parameters menu, and selected User-Specified Random Effects. After parameterizing the Design Matrix (1 column for intercept, 1 column for “exodus”), I clicked “OK” and got the following error message:

forrtl: severe (59): list-directed I/O syntax error, unit4, file [file path... mrk5549z.tmp] Image PC Routine Line Source DFORRT.DLL 00354BD9 Unknown Unknown Unknown

I have the most recent version of MARK, and tried running it on a separate computer. I tried shortening my .inp file to see if it would work with less data in the file, and also tried running it with/without constraining the non-converging time intervals. Nothing seems to work, and I keep running into this error.

Any suggestions would be greatly appreciated.

[DTI]

In case it helps:

I'm running MARK version 6.1
Operating System: Windows Vista Home Premium Service Pack 2
Time stamp for MARK_INT.EXE (17/12/2011)
Time stamp for MARK.EXE (03/12/2011)

[cooch]

Hopefully this question hasn't been asked already (though after RTFM and searching the forums, I haven't come across a mention of this problem yet).

Good on you for trying.

I'm trying to fit random effects models to nest survival data in MARK. Data was collected over 23 years, and there are 76 encounter occasions in the dataset (thus, the longest "field season" of the 23 years spanned 76 days). Each year of study was classified as one of two types: "Pre-exodus" (given a value of 0), or "Post-exodus" (given a value of 1).

That is easily enough 'time steps' for random effects models to give meaningful estimates of the latent parameter(s).

I'm interested in fitting a random effects model in which the intercept differs depending on the type of year. Thus, I have an individual covariate called “exodus” that is coded 0 for nests in the pre-exodus years, and 1 for nests in the post-exodus years.

I'm curious how you coded this -- form the rest of your note, it sounds as though you're using an intercept-based offset model, wherein the beta coefficient for your 'exodus' covariate (a 0 or 1), is simply the offset of a given effect from the common intercept. While there are ways to generate models with separate intercepts, I'm not entirely sure why you'd want to (but, I'm not familiar with the aims and intents of your study).

To ensure that my .inp file was working correctly, I started by running a fixed-effect full time model S(t), a fixed-effect intercept model S(.), and a fixed-effect exodus model S(exodus). All three returned with no errors (though there were some non-convergence issues in several time intervals for the S(t) model).

This is not uncommon, especially if you have a lot of occasions relative to the time horizon, such that interval-specific estimates of survival are -> 1.0. Having a slew of estimates 'o the boundary' can cause convergence issues.

Next, I ran an intercept-only random effects model and a linear trend random effects model, both of which returned with no errors. Finally, I tried to run a random-effects model where the intercept differs as a function of “exodus.” I retrieved S(t), opened the Variance Components/Real Parameters menu, and selected User-Specified Random Effects. After parameterizing the Design Matrix (1 column for intercept, 1 column for “exodus”), I clicked “OK” and got the following error message:

forrtl: severe (59): list-directed I/O syntax error, unit4, file [file path... mrk5549z.tmp] Image PC Routine Line Source DFORRT.DLL 00354BD9 Unknown Unknown Unknown

I have the most recent version of MARK, and tried running it on a separate computer. I tried shortening my .inp file to see if it would work with less data in the file, and also tried running it with/without constraining the non-converging time intervals. Nothing seems to work, and I keep running into this error.

Any suggestions would be greatly appreciated.

Trying it on another computer suggests it isn't a corrupted .tmp file problem (as is often the case). One silly thing to try -- turn off you virus software temporarily, and try again. Also, try running MARK using administrator privileges. If that doesn't ork, I might ask you to send me some files 'off-list'.

[cooch]

In case it helps:

I'm running MARK version 6.1
Operating System: Windows Vista Home Premium Service Pack 2
Time stamp for MARK_INT.EXE (17/12/2011)
Time stamp for MARK.EXE (03/12/2011)

Your version of MARK is an iteration or two out of date. I doubt it will 'solve your problem', but as a general strategy, not a bad idea to uninstall your current version, and installing the latest and greatest version (January 2, 2012 time stamp).

When uninstalling, using a formal uninstall procedure (don't simply delete the MARK directory).

[DTI]

Thank you for the pointers and the very quick reply!

As you suggested, I uninstalled (formally) and re-downloaded/re-installed MARK. Unfortunately, while running MARK as an administrator with my firewall turned off, I still received the error message.

I'm curious how you coded this -- form the rest of your note, it sounds as though you're using an intercept-based offset model, wherein the beta coefficient for your 'exodus' covariate (a 0 or 1), is simply the offset of a given effect from the common intercept. While there are ways to generate models with separate intercepts, I'm not entirely sure why you'd want to (but, I'm not familiar with the aims and intents of your study).

Yep, that's the idea - my hypothesis is that μ(pre exodus) =/= μ(post exodus). If I understand correctly, we can't directly test that σ^2(pre-exodus) =/= σ^2(post-exodus) using the variance components approach in MARK. The end goal is to generate separate beta distributions for annual nest success in the pre- and post-exodus time periods to feed into a population model.

[DTI]

Perhaps this has something to do with trying to use a "named" individual covariate in the random effects model. In Appendix D (section D.4.3), the environmental covariate used in the example was not a named individual covariate.

To get around using a named individual covariate in my random effects model, I think I can group my nests based on their pre-/post-exodus status (group 1 = pre-exodus, group 2 = post exodus). Thus, my design matrix will now have 75*2 = 150 rows, and I can use 1's and 0's to parametrize an intercept-offset model rather than having a second column for an individual covariate.