MARK's POPAN chase.inp example 12.4.1 not working?

[geofgivens]

See my last post. It is from June 2014 and says 8.0.

If you are using RMark, make sure there are no old mark32.exe, mark64.exe or mark.exe laying around that are being picked up in the path.

Ok, I'll do that, but the problem occurs when I use MARK as a standalone, starting it from the start menu or the icon that it installs. Whether RMark is confused about which installation to use is a different question.

[cooch]

See my last post. It is from June 2014 and says 8.0.

If you are using RMark, make sure there are no old mark32.exe, mark64.exe or mark.exe laying around that are being picked up in the path.

I have a July, 2014 build I can make available. Send me a note offline.

[geofgivens]

So in the process of following Jeff's suggestion to look for old installations, I found an installation of 8.0 on my machine and tested the example with it. Same failure. To summarize, I think we have:

Fail:
Geof 8.1, 8.0, 7.2 (and RMark with 8.1)
Bill version unspecified

Works:
Jeff 8.0 via RMark

I am getting worried that I may be doing something stupid, but if so, then Bill is doing the same thing. It's a pretty straightforward example.

[jlaake]

I sent you an email with links to my exe versions.

[cooch]

So in the process of following Jeff's suggestion to look for old installations, I found an installation of 8.0 on my machine and tested the example with it. Same failure. To summarize, I think we have:

Fail:
Geof 8.1, 8.0, 7.2 (and RMark with 8.1)
Bill version unspecified

Works:
Jeff 8.0 via RMark

I am getting worried that I may be doing something stupid, but if so, then Bill is doing the same thing. It's a pretty straightforward example.

Actually, I can''t replicate any of the results in Chapter 12, using any version of MARK back to 7.2. So, most likely explanation is that the data files Carl used are different than the ones I have associated with his chapter.

Not much I can do about that, I'm afraid. I'll email Carl.

[cooch]

I spent a fair bit of time sussing out the issues, and am fairly sure I know what the problems are (and what are not problems).

1\ there were some issues related to use of chase.inp. Scrap it. Re-download a fresh copy of markdata.zip (from the book download site), and use either chase_both.inp (for models with both adults and jacks), and chase_adults.inp (for analysis of adults only). Both data sets work fine with 'new MARK' (meaning, I can replicate the results shown in Chapter 12 without errors), but there are a fair number of things you might want to be aware of...(below)

2\ with 'new MARK', results for some data sets-models are annoyingly dependent on choice of the link function for phi and p parameters. In Carl's chapter, for example, when using the adults-only salmon data, Carl (in theory) reported estimates from 'old MARK' using the sin link on phi and p. However, if you try the sin link using 'new MARK', you get all sorts of convergence problems, and the estimates aren't even particularly close to what Carl reports in Chapter 2. But, a simple switch to the logit link on phi and p, and....pretty much the precise estimates Carl reports. So, you may need to 'play around' somewhat concerning the link function, wrt to the phi and p parms. pent is still MLogit, and N is still log.

2\ its also pretty clear that various things in 'new MARK' have changed substantially -- for example, virtually all of the model deviances Carl reports are positive, whereas in 'new MARK', same models invariably have negative deviances. The AIC values are more or less correct (but see point 4), but it will mean going through the chapter and updating all of the screen captures of results browsers, since as they stand, they appear quite different to what people will see using 'new MARK'. Its on my 'to do list'...

3\ I was 'impressed' (disappointed) at how poorly simulated annealing does in general with POPAN models. In almost every instance, got underflow + convergence errors. For the moment, I don't recommend it until the underlying issues get explored more fully.

4\ In some cases, there were some real differences between 'old MARK' and 'new MARK' in the number of estimated parameters -- even if/when the deviances were the same. This lead to differences in reported AIC, obviously. Again, under study.

5\ people were getting tripped up by Carl's use of 'adults only', when in fact the Chase data set everyone was using had both age classes. So, you might have all sorts of errors -- especially using RMark. To solve this problem in a 'blunt force' way, I simply created versions of the Chase data set: one with both age classes, and one with adults only (see pint 1, above).

[geofgivens]

Wow! Thanks for all your work on this. I don't think it addresses the original question, though.

This arose because I had some code that was working, then updated MARK, then the same code produced the "not enough time intervals" error. In the process of investigating the issue, I tried to run the chase.inp example and got the same "not enough time intervals" error. With the new datasets you provided now, I still get the "not enough time intervals" error. And with the simple dataset enclosed, same error. This is all for POPAN. Other models (e.g. Huggins) run successfully with the same data file.

I wonder if I'm doing something stupid. However, one of the earlier posters also got the same error. Aside from that replication, another piece of evidence is that taking Jeff Laake's RMark code and running it verbatim works for him but not for me. Same error.

I appreciate your effort to help diagnose this.

Geof

PS. I tried a variety of link functions with your new chase datasets, but the "not enough time intervals" error remained.
PPS. With my simple data, I also tried using 5 occasions (4 intervals) to try to produce a different error. However, I still got "not enough time intervals".

Code: Select all

001  359 ;
010  373 ;
011   37 ;
100  387 ;
101   39 ;
110   36 ;
111    4 ;


[jlaake]

Geof-

Did you get my email with the links to my version of the mark exes? If you copy those over your current versions it should work.

[cooch]

PPS. With my simple data, I also tried using 5 occasions (4 intervals) to try to produce a different error. However, I still got "not enough time intervals".

Code: Select all

001  359 ;
010  373 ;
011   37 ;
100  387 ;
101   39 ;
110   36 ;
111    4 ;


If the above are the simple data you're referring to, then no wonder -- these data show 3 occasions....not 5. [And 3 wouldn't be sufficient to run a POPAN model anyway...]

More to the point, if you take chase_adult.inp, run POPAN, with 8 occasions, 1 group, with interval lengths of 1.5,1,1,1,1,1,1.5, then you shouldn't get any errors. I tried this with 6-7 different versions of MARK, including the most recent one, and they all work perfectly. Don't use RMark to test - use classic MARK (only as a step in ruling out an RMark problem). I can assure you that chase_adult.inp works fine (I even had 2-3 students in my undergraduate class try, and it worked fine). But, make sure you specify 8 occasions. The error you're getting makes me suspicious you're simply entering the wrong number of occasions (if it helps, the number of occasions is simply the string length of the encounter history).

[geofgivens]

I completely uninstalled all MARK software, downloaded and re-installed 8.1, and the problem has vanished. My R code works. The chase example works (although I didn't compare the output to the book). My simple dataset works.

I have no idea why the POPAN model was susceptible to some sort of installation file issue (?) that didn't affect other models, but there it is.

Thanks to all of you for your help.

PS. Yes I was using 8 occasions for chase and 3 for my simple data. My point in the previous footnote was that I had intentionally screwed up and yet got the same message.