Program Aborting when running Bootstrap GOF

[Ryan Fisher]

Hi there,
I recently upgraded to the newest version of MARK (v. 4.1) and have been trying to run some Bootstrap GOF testing on a couple datasets that I had constructed on an older version of mark. As soon as I begin to run the BootStrap procedure (i.e. after it asks how many simulations, etc) the program stops running and aborts. I have been trying it on just basic models (phi(t), p(t)) in a recapture only model.

I have however been able to run a BootStrap GOF procedure on a dataset (again, just recaptures only, and phi(t)p(t) model) that I just constructed on the new version of MARK and it seems to run fine. Is this just a bug?

Thanks,
Ryan

[cooch]

Hi there,
I recently upgraded to the newest version of MARK (v. 4.1) and have been trying to run some Bootstrap GOF testing on a couple datasets that I had constructed on an older version of mark. As soon as I begin to run the BootStrap procedure (i.e. after it asks how many simulations, etc) the program stops running and aborts. I have been trying it on just basic models (phi(t), p(t)) in a recapture only model.

There are a number of small incompatabilities between newer versions of MARK, and some older MARK files (i.e., the .fpt and .dbf files). This often shows up when you read in the models in an old browser, and try to retrive and run one of them and have MARK abort. I'll leave it to Gary to explain the underlying reasons (my guess is it has to do with changes on how the deviance of the saturated model is calculated between older versions of MARK, and the newest version).

A solution which seems to work more often than not is to re-run all the models without retrieving them. Simply bring up the models in the browser, then select 'Run' -> 'Re-run model(s)' ->
Select All'.

Re-run all the models, and then try the bootstrap GOF (actually, you shouldconsider not using the bootstrap GOF test - see the discussion in the newest version of the GOF chapter).

[gwhite]

Ryan:
Did you completely uninstall the old version of MARK before installing the new version? Because of changes in the DLL files that accompany MARK, you have to get rid of the old version in order to have the new version work properly.
Gary

[Ryan Fisher]

Thanks for the quick reply. I tried re-running all of the models and no luck. Every other test in the "Tests" list seems to work fine except the bootstrap procedure and the median c hat procedure - both of which cause MARK to abort when I try to run them.
Also, I did uninstall the previous version of MARK that I had on my computer before installing this latest version. However, I remember some windows that popped up when I uninstalled the old version referring to the *.dll files, but can't remember what exactly they contained. Is there anyway that I can check this to make sure everything from the old program was deleted? Should I just try uninstalling and then re-installing this latest version again?


Thanks for all the help

[gwhite]

Ryan:
The messages that pop up when you uninstalled and re-installed MARK are probably not an issue as long as the graphics module works okay.

Does your analysis have individual covariates included in the data file? If so, that is probably the reason for the abort.

Gary

[Ryan Fisher]

Yes my data file does contain individual covariates. So that must be the problem. Should one then just run the GOF procedures on a data file without the individual covariates included?

Thanks for the help,
Ryan

[gwhite]

Ryan:
Unless I get ambitious and change the MARK code this weekend, your easiest work around is to re-create the MARK files without any covariates. You can use the same input file, but don't specify any individual covariates. Then, create the global model that you want to use.

However, individual covariates remove overdispersion, so the c-hat you estimate with the above global model is not necessarily a good estimate for your models that include individual covariates.

I'll get this bug fixed shortly -- I've known about the problem for some time, but keep forgetting to fix it when I do an update.

Gary

[apotticary]

Hello-

I have recently tried to run a bootstrap GOF test on my dataset, for which I am using a multi-state capture recapture model. The test will run (albeit very slowly, as expected), but when it finishes and I try to open the result dbfs, MARK says there is no information in any of the output files and then aborts. I don't have any individual covariates in my data, I've only downloaded one version of MARK and I'm running out of ideas of how to figure this out. Any suggestions?

[cooch]

Hello-

I have recently tried to run a bootstrap GOF test on my dataset, for which I am using a multi-state capture recapture model. The test will run (albeit very slowly, as expected), but when it finishes and I try to open the result dbfs, MARK says there is no information in any of the output files and then aborts. I don't have any individual covariates in my data, I've only downloaded one version of MARK and I'm running out of ideas of how to figure this out. Any suggestions?

Couple of things:

1\ this should be in the 'software problems' subforum - I'll move it there shortly.

2\ try the median c-hat approach. The bootstrap doesn't work very well, at a number of levels. Also, for a MS analysis, have a look at U-CARE. All of this is discussed in Chapter 5.