Modeling sex with unknown assignments

questions concerning analysis/theory using program MARK

Modeling sex with unknown assignments

Postby nperlut » Wed Jul 03, 2019 7:56 am

Hi all- I need some advice on how to best set up a sex covariate for CJS models. I know the sex for about 60% of the data set. I built in three sex covariates - male (0,1), female (1,0) unknown (0,0). A second option could be to model sex as a single covariate (male =1, female =2, unknown = 3). I lean towards the first method, but would very much appreciate hearing how others have dealt with this. Thanks.
nperlut
 
Posts: 14
Joined: Tue Feb 07, 2006 9:28 am

Re: Modeling sex with unknown assignments

Postby cooch » Wed Jul 03, 2019 8:19 am

nperlut wrote:Hi all- I need some advice on how to best set up a sex covariate for CJS models. I know the sex for about 60% of the data set. I built in three sex covariates - male (0,1), female (1,0) unknown (0,0). A second option could be to model sex as a single covariate (male =1, female =2, unknown = 3). I lean towards the first method, but would very much appreciate hearing how others have dealt with this. Thanks.


There are two issues here:

1\ how to code for 'group' effects --> this issue (and relative merits of the 2 approaches you describe) are discussed in section 11.7 of Chapter 11.

2\ the 'unknown' sex group is, in fact, a mixture of individuals that are either male or female. There are any number of approaches to handling this that have been discussed in the literature, which you're urged to consult. You might start with Conn & Cooch (2009), which discusses one approach, and which also collates some of the other literature. A key factor to consider is if the 'unknown sex' individuals are always unknown, on each encounter, or if they're only unknown 'some of the time'. It makes a difference in terms of how you might approach the problem.
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: Modeling sex with unknown assignments

Postby nperlut » Wed Jul 03, 2019 11:08 am

Thanks!
nperlut
 
Posts: 14
Joined: Tue Feb 07, 2006 9:28 am

Re: Modeling sex with unknown assignments

Postby simone77 » Fri Jul 05, 2019 5:07 am

A framework like this is especially suitable to be modeled using multi-event modelling in E-SURGE as E. Cooch mentioned. This is the section of analysis help in MARK and I hope not to go off-topic in my answer, this is just to help you to figure out a bit more how it would work in E-SURGE. The main advantage would be that you do not loose information by having to consider a third group made of unknown sex individuals.

In your case, if you have no mis-classification (female wrongly assigned to male and vice versa) you would have three states (female alive, male alive, dead) and four events (0, not detected, 1 detected assessed as female, 2 detected assessed as male, 3 detected without sex assessment).

You will have to define your probabilistic matrices in GEPAT (see*). In the GEPAT syntax a letter, any, means the parameter/probability you want to get estimated, an "*" is the probability of the complementary event (i.e. 1-p), and a "_" means you know that probability is zero. In your case you would have:

INITIAL STATE (one single step)
Initial State (two columns: fem, mal)
* p

TRANSITION (one single step)
Survival (3 rows and 3 columns: fem, mal, dead)
s _ *
_ s *
_ _ *

EVENT (two steps)
Encounter (3 rows: fem, mal, dead; 3 columns: non-detected, female detected, male detected)
* p _
* _ p
* _ _

Sex Assessment (3 rows: non-detected, female detected, male detected; 4 columns: 0, 1, 2, 3)
* _ _ _
_ d _ *
_ _ d *

Note you can put any letter in the matrix (its position determines the parameter you want to be estimated), it is arbitrary.

There are good references (e.g. **;***) with good supplementary material you may want to see if you decide to use this approach. In that case, you may find extra help in the E-SURGE Phidot section.

* https://www.cefe.cnrs.fr/images/stories ... MANUAL.pdf
** Pradel, R., Maurin‐Bernier, L., Gimenez, O., Genovart, M., Choquet, R., & Oro, D. (2008). Estimation of sex‐specific survival with uncertainty in sex assessment. Canadian Journal of Statistics, 36(1), 29-42.
*** Genovart, M., Pradel, R., & Oro, D. (2012). Exploiting uncertain ecological fieldwork data with multi‐event capture–recapture modelling: an example with bird sex assignment. Journal of animal ecology, 81(5), 970-977.
simone77
 
Posts: 197
Joined: Mon Aug 10, 2009 2:52 pm

Re: Modeling sex with unknown assignments

Postby cooch » Sat Jul 06, 2019 1:43 pm

simone77 wrote:A framework like this is especially suitable to be modeled using multi-event modelling in E-SURGE as E. Cooch mentioned. This is the section of analysis help in MARK and I hope not to go off-topic in my answer, this is just to help you to figure out a bit more how it would work in E-SURGE.


Very helpful. For Martkov problems with unobservable states, or states with uncertainty, E-SURGE is quite elegant (which is why Paul Conn and I used it in our 2009 paper).

It will be of interest to know that many of the E-SURGE capabilities are now in MARK, just not documented (book author being somewhat too distracted to write it up).
cooch
 
Posts: 1628
Joined: Thu May 15, 2003 4:11 pm
Location: Cornell University

Re: Modeling sex with unknown assignments

Postby Bill Kendall » Mon Jul 22, 2019 4:50 pm

I agree with both Evan and Simone. The multievent structure is a good general framework for tackling this problem. LOLASURVIV, on the Patuxent website, was developed Jim Hines for this particular problem. One point from this work, discussed in Nichols et al. (2004, Ecology), is that it's important not to simply deal with unknown sex through covariates. This will produced sex-specific survival rates that are biased high, because an individual has to live long enough for its sex to be identified.

This all assumes the case where individuals are first detected with sex unknown, and then their sex is determined with some probability later. If each individual is either of known sex or unknown sex when first marked, and stay that way for the entire study, then your original plan of using covariates in a CJS model would be fine.
Bill Kendall
 
Posts: 96
Joined: Wed Jun 04, 2003 8:58 am


Return to analysis help

Who is online

Users browsing this forum: No registered users and 10 guests