Strange State Assignment estimates and EM(20) not working

[simone77]

Hi Josefa,

This morning I have been checking again your first post and I have realized something that perhaps might be important.
I don’t know if you are using the inp format or the headed format (this last one, once you have understood how it works, allows much more flexibility), anyway, you can try to create three groups according to the certainty/uncertainty of the reproductive status. Like this (in inp format):
10002000…110034 1 0 0;
02002100…110000 0 1 0;
00004330…111002 0 0 1;

Individuals of the first group (1 0 0;) will be all the individuals firstly captured as breeder, individuals of the second (0 1 0;) will be all those firstly captured as non-breeders and individuals of the third (0 0 1;) will be all those firstly captured as unknown breeder status.

Once you have prepared your data set this way, you might repeat your analyses using the same GEPAT structure and almost the same GEMACO syntax, you should just modify the sentence for the Initial State:

IS = g(1)+g(2)+g(3).t

and in the IVFV you should fix the first two parameters [corresponding to group 1 (sure first capture as breeder) and to group 2 (sure first capture as non-breeder)] respectively to “1” and “0”.

A second point, always good to check before starting an analysis: have a look at the m-array, you can do it in very simple way, by opening your data in U-Care and somewhere in “tools” you have that option.

[jbleu]

Hi,
Thanks for your responses, I did not have the time to test your ideas yet but I am not sure why I should create groups with the state of first capture?
Also I am going to a workshop on E-surge next week and thus I hope to understand the problem and find a solution there... I'll post here the answers or suggestions after the workshop...

[simone77]

Hi Josefa,

Enjoy the workshop, it is very useful, I was there last year.
Regarding my suggestion, I believe you can reduce the extent of your uncertainty in the analysis. That's why I suggested to create three groups according to their reproductive status first time they were captured. You really have not uncertainty about lot of individuals with respect to the initial state probability and this way you could make work easier for E-SURGE.

Anyway, I believe there are several other things you have to take into account (for instance: are observations from spring and summer pooled?). There at the workshop you have the possibility to speak with very skilled people and to show your data and analyses in detail: not doubt you'll find a solution.

By the way it will be useful and interesting to see how you solved this.

[jbleu]

Hi,
I completely forgot to post on the forum after the workshop. It was very useful, and I learn how to write a correct model for trap-dependence. However it did not solve my problem on state assignment, which is probably due to the quality of the data...
I think I will work with a similar probability of state assignment for breeder and non breeder, since it seems I can't test whether this probability is different or not. Probably I will check for the best models how the estimates are affected if I change the state assignment probability.