Groups with different numbers of occassions

[cmb]

Is it possible to analyse data from two study sites with different numbers of trapping occassions? I have data from two longish-term studies, where one study has been running 30 years, the other 20. I can see that it would be possible to code a DM with this format (in a manner analagous to defining interactions between age classes, where the number of years this interaction is available is less than the total number of years), but I can't see how to do this in practice. Defining the .inp file seems to mean that certain combinations rows and columns within the DM are not available(?).

I could analyse the data separately for both sites, or I could analyse only the years where both studies were running in one analysis, but I see no real reason why I shoudn't be able to do both together with uneven numbers of occassions. I'm sure such data are regularly available and analysed, but I can't find any mention of a similar question in the archives. If anyone has any pointers to postings I've missed or how to define the DM necessary in practice I'd be very grateful.

Thanks!

[Bill Kendall]

I believe the best way would be to create the input file with the appropriate number of columns for the group with the longer study. For the other group, enter all zeros in the columns where there was no sampling effort. Then, set the p's for that group equal to 0 for those periods where there is no sampling effort. Of course some of the parameters of interest will not be estimable because of these data gaps (especially under a fully time and group dependent model). Which those are will depend on the model, and the trick will be to recognize them.

[Andrea]

I believe the best way would be to create the input file with the appropriate number of columns for the group with the longer study. For the other group, enter all zeros in the columns where there was no sampling effort. Then, set the p's for that group equal to 0 for those periods where there is no sampling effort. Of course some of the parameters of interest will not be estimable because of these data gaps (especially under a fully time and group dependent model). Which those are will depend on the model, and the trick will be to recognize them.

I have a similar issue with my capture-recapture data. It comes from 2 sites very similar to each other, one of them was not sampled in one occassion. If I follow Bill's suggestion, then I'll allways have to have site as a grouping factor even though I'm not interested in site differences. By adding this grouping factor to the models I feel that I'm expending some precious degrees of freedom that could be used for more interesting factors and interaction terms.
Is there an alternative to this approach?
Thanks,
Andrea

[Eric Janney]

I have a somewhat similar issue. I have a dataset of PIT tagged fish from 1995 - 2007. We changed to a new and improved PIT tag in 2005 but we do not retag recaptured fish that have the old tag type. This improved read range doesn't make any different for fish that we catch, but we also use fixed underwater PIT antennas to detect fish on spawning grounds. So, the fish tagged with the new tags could potentially have high p's than fish tagged with the old tags due to the improved read range. If you assign groups according to what type of tag they received then their are no data prior to 2005 for the new tags. Therefore, a bunch of unestimable parameters. It doesn't matter if you fix them to zero because MARK does not count them either way.

[Andrea]

In my case setting the p for that group in that particular occasion to zero works well. But I still don't like the idea of being obligated to set recapture probabilities as time and group dependent in order to fix those parameters. This increases the number of parameters to be estimated and I rather increase the model complexity by looking at other more interesting parameters instead. Does anybody know of an alternative way to get around the problem of unequal number of encounter occasions for different sites/groups?
Andrea

[jsaracco]

Andrea,

I don't understand how you would be using up df by following Bill's suggestion. You do need to distinguish 2 groups in the input file, but you only need to indicate one extra parameter (via the pims or dm), which is never really estimated because you set it to zero. If you set all of the other parameter indices or corresponding rows in the dm to be the same then you have not used up any extra dfs. Right?