Constrain parameter values of a group

[simone77]

Hi,

I have three groups according to their gender:
1 0 0; (males)
0 1 0; (females)
0 0 1, (sex unknown)

I am using a CJS and would like to set the parameters values (apparent survival and capture rates) of group three to be intermediate between same parameters of males and females (as they are a mix of both genders).
Is there a way to do it?

Thanks in advance for any response,

Simone

[cooch]

Hi,

I have three groups according to their gender:
1 0 0; (males)
0 1 0; (females)
0 0 1, (sex unknown)

I am using a CJS and would like to set the parameters values (apparent survival and capture rates) of group three to be intermediate between same parameters of males and females (as they are a mix of both genders).
Is there a way to do it?

Thanks in advance for any response,

Simone

TO some degree, yes, using a cumulative logit link, and some thought.

But, what you're proposing is unnecessary, and not the preferred approach, since (in your question) you're already assuming something about the 'make-up' of the unknown group.The expectation that the survival for the unknown group being intermediate amongst the two sexes is based on implicit assumptions about the relative proportions of males and females in the unknown group.

There are other, more robust approaches. Basically, you're considering a problem where the 'state' (sex) of an individual is unknown (or, equivalently, unobservable). What you do with that depends on whether this state is 'permanent' (i.e., if the sex for the individual is never known), or transient (i.e., if there is a no-zero probability that at some point you might know the sex). There are any number of papers on the subject -- see work by Pradel, Nichols, Kendall -- search for the word 'unknown' and/or 'sex' in the title.

[simone77]

In my case the sex is genetically determined on a portion of chicks when they are captured first time and marked. The recapture are resightings and a bird whose sex is unknown (because it has not been bled by chick) will remain always unknown. About one third of all the marked individuals have been "sexed".

For this reason multievent approach seems not to be adequate (there are no transitions among states and neither among events - seen as unknown and after seen as male/female).
I wanted to constrain these parameters to have intermediate values in the third group (unknown sex) just to "save" a parameter and gain something in robustness and precision of estimates.


1. Do you believe it is worth it?
2. In case it might be, how should I do it? any example with cumulative Logit link?

Thank you for your help,

Simone

[cooch]

In my case the sex is genetically determined on a portion of chicks when they are captured first time and marked. The recapture are resightings and a bird whose sex is unknown (because it has not been bled by chick) will remain always unknown. About one third of all the marked individuals have been "sexed".

For this reason multievent approach seems not to be adequate (there are no transitions among states and neither among events - seen as unknown and after seen as male/female).
I wanted to constrain these parameters to have intermediate values in the third group (unknown sex) just to "save" a parameter and gain something in robustness and precision of estimates.


1. Do you believe it is worth it?
2. In case it might be, how should I do it? any example with cumulative Logit link?

Thank you for your help,

Simone

You're still missing the point about constraining to be 'intermediate'. What if 100% of the unknown sex individuals were in fact male? An extreme perhaps, but it makes the point. The unkown group is an unknown mixture. And, should be treated as such.

Have a look at 'Estimation of sex-specific survival with uncertainty in sex assessment
by Roger Pradel, Lory Maurin-Bernier, Olivier Gimenez, Meritxell Genovart, Rémi Choquet, Daniel Oro. 2008. Canadian Journal of Statistics.

Also, Roger discusses a multi-event version of the unknown sex problem in his paper 'The Stakes of capture-recapture models with state uncertainty', in the volume co-edited by Thomson, Cooch & Conroy.

[simone77]

Thanks for the references. I would have used a multi-event approach in case I had uncertainty in sex determination but in my case there is no uncertainty, there is just a certain determination of gender on just a portion of individuals (and as said above no transitions neither on states neither on events).

Perhaps I could approximate the sex ratio of unknown individuals for each cohort as I have a portion of individuals genetically sexed for each marking session, and perhaps I might estimate sex ratio of post-chicks individuals attending the colony (both marking and resightings sessions are done in the colony).
Anyway there would be uncertainty in both the sex ratio of chicks (cohort level) and of after-chicks resighted individuals and I would not know if it is possible to treat that issue and how to do it.

I will go on just by letting that parameters (of the unknown sex group) to be unconstrained and expecting to get intermediate estimates of both apparent survival and recapture rates.

All the best,

Simone