Issues with RELEASE/GOF testing

[maralions]

Hello everyone,

I am having trouble running a GOF test in MARK. I have been using RMark up to this point because of my complex models and have exported a model using export.MARK for GOF testing. My model has individual covariates and therefore I understand that RELEASE GOF is my only/best (?) option. I am actually revisiting this data set from last year to prepare it for publication (at the minor revision stage) and did manage a GOF test previously, but I am wanting to revisit this because of some questions regarding my use of QAICc. My previous GOF test indicated that c-hat was <1 and therefore I left this unadjusted, but I am now dubious about this result.

When I try the RELEASE test I get a message "Dots were found in the encounter history, and will be replaced with zeros". The resulting text file doesn't contain any of the tests.

Here are the first two lines of my encounter histories including some individual covariates:

ch AgeClass Pride Sex ID Count_ind Density_ind MB_ind
11111111000 C1 1 F MLF11 0 0 0
11111101000 C1 1 F MLF13 0 0 0
00000011000 C1 1 U MLU283 0 0 0

The rest continues as above and I cannot see any dots.

Does anyone know what I am doing wrong here? I really am at a loss! I am very reluctant to restart the whole process in MARK as I made many modifications in RMark and the thought of recreating PIMs fills me with dread.

Thanks!
Sara

[cooch]

Hello everyone,

I am having trouble running a GOF test in MARK. I have been using RMark up to this point because of my complex models and have exported a model using export.MARK for GOF testing. My model has individual covariates and therefore I understand that RELEASE GOF is my only/best (?) option....

Not quite. For models with individual covariates, there are no GOF tests that apply. Do the GOF testing on your most general model without the individual covariates.

See section 11.9 in Chapter 11 of the MARK book:

http://www.phidot.org/software/mark/doc ... chap11.pdf

[maralions]

Thanks very much for that. I've managed to do a RELEASE GOF test and this hasn't come out as significant so I think I am OK...!

Sara

[jlaake]

I worked with Sara offlist. The problem was that the line length for the input file was greater than allowed. I had her cut back on the group structure including the most important variables of sex and age. This reduced the line length for Release and it reported a result which was not significant and thus she is assuming c=1. Had she been able to include her Pride variable as well, the chi-square would have been further reduced. Thus even though she could not use her exact global model, one with some of the important structure suggested no issue with over-dispersion.

regards --jeff