Missing parameters in results output?

[cebert]

Hello!
I have calculated Population estimates using the data type 'full closed captures with heterogeneity and misidentification'. Now I noticed in some cases large differences concerning the number of parameters (given in the results output) between models which did not differ at all when being build (in the PIM). The only difference between the two respective models was that in one I restricted alpha to 0.95 and in the other to 0.97. Both models had 44 parameters in the PIM (unfortunalety, all the more parsimonious models with less than 20 parameters were not supported at all...) , but in the model results, one had 44 and the other 34 parameters (and a delta AIC of almost eight). What I don't understand is where this difference comes from - where have the 10 parameters gone in the second model? Can someone give me a hint?
All the best,
cebert

[cebert]

Hello, I have an addition to my post last Friday. In the last days, it occurred several times that the parameter number of models I have build using the PIM differs from that given later in the results window. E.g. I just ran a model with 27 parameters, and in the results window it is said in the "number of paramaters" column for the [b]same[/b] model 22 parameters. It would be great if you could help me by unraveling why there is this difference in parameter numbers between he models when I construct them and the results output!!
Cheers,
Cornelia

[dhewitt]

It seems you are running into estimation problems and MARK is not counting parameters it had 'trouble' with. For example, estimates on boundaries (0 or 1). There is plenty of information about this problem in the MARK book (Evan can tell you where exactly) and also on this forum. Main problem: data too sparse to estimate all that stuff!

[cebert]

Thanks a lot for your reply!
I had already searched in the MARK handbook, but it seems that I did not search in the right places... (well, it is sort of comprehensive...). But now I found something on this topic in Chapter 6 (6.4. Running the model: details of the output).
What might the best advice be when parameters are 'lost' in some models? Should I just delete those models and use models with less parameters (even if they fit less well to the data)? I am already using the sin link function (because using data type 'full closed captures with heterogeneity and misidentification')...
Thanks again,
Cornelia

[dhewitt]

Cornelia,
You need to do some searching on your own on this forum. Here is one example of some discussion that comes up about this:

viewtopic.php?f=1&t=1484

No one can tell you what to do because it depends on what is really causing the problem, what is realistic to do with the data you have, and the system/critter you're studying relative to the models you're fitting. Simply deleting those models may not be the best solution.

- Dave