## GEMACO sentence about an unestimable parameter

questions concerning analysis/theory using programs M-SURGE, E-SURGE and U-CARE

### GEMACO sentence about an unestimable parameter

Hi,

This is something I forgot to ask during the past Montpellier workshop, so here I am (again!).
I am dealing now with an analysis where individuals were sexed since the 3rd occasion (total=14 occasions).
These are the states:
1=female
2=male
and these are the events:
0=not seen
1=captured and identified as male
2=captured and identified as female
3=captured and sex unidentified
The INITIAL STATE pattern matrix would be:
p*
and the Event step two (SEX IDENTIFICATION) would be:
*---
-p-*
--p*
As the INITIAL STATE is not estimable for the two first parameters (there are not sexed individuals), I would use the sentence "t(3_14)" and for the SEX IDENTIFICATION I would use f.t(3_14).

I have tried to use as alternative GEMACO sentence in SEX IDENTIFCATION "f.t(3_14)+t(1 2)" (instead of "f.t(3_14)" alone) and set the second part (t(1 2))=0 in the IVFV. I get different results in terms of deviance and AIC (the ranking is not affected). This is a bit confounding to me because in these cases if you don't specify nothing about these somehow nonsensical parameters (in this case the first two occasions) in GEMACO, ESURGE will try to estimate these parameters because they are not structurally unestimable (I mean it is not as the case of age2 in t1 for instance). In some way there are not "good" data to do that, i.e. in this case there are not sexed individuals but yes there are captures.

Question:
So, in general terms what is the correct way to deal with this? writing down a sentence like "f.t(3_14)" or using a sentence like "f.t(3_14)+t(1 2)" and setting the second part equal to zero in the IVFV?

Hope you are doing well,
Thanks

Simone
simone77

Posts: 148
Joined: Mon Aug 10, 2009 2:52 pm

### Re: GEMACO sentence about an unestimable parameter

Hello,
The first sentence "f.t(3_14)" leads to parameters for time 1 and 2 unmodelized.
In that case, E-SURGE sets them by default to 0.5 (in our case).
This is the reason why your results are different.
Unmodelized parameters are not recommended at all.
Sincerely,
Rémi
CHOQUET

Posts: 211
Joined: Thu Nov 24, 2005 4:58 am
Location: CEFE, Montpellier, FRANCE.

### Re: GEMACO sentence about an unestimable parameter

Hi Rémi,

OK, it gives an answer to why they get different results.
That means I should explicitly define in GEMACO all the (structurally estimable) parameters.

And how should I set parameters that have some unestimable unknown value?
For instance, in the previous example, I could easily set to zero the second part of the sentence "f.t(3_14)+t(1 2)" as I know that the p that the gender of a captured individual is ascertained in the first two occasions is zero.
But what about the values of INITIAL STATE parameters for the first two occasions?
If I used the sentence "t(3_14)+t(1 2)" I would not know which values assign to that second part (t(1 2)), because the p a firstly captured individual is male or female is not zero, is not estimable with my data.
simone77

Posts: 148
Joined: Mon Aug 10, 2009 2:52 pm

### Re: GEMACO sentence about an unestimable parameter

If first initial state probabiities are not separately estimable then
use a constraint of equality like t(1:3,4_14) which set equal time 1 to 3.

Rémi
CHOQUET

Posts: 211
Joined: Thu Nov 24, 2005 4:58 am
Location: CEFE, Montpellier, FRANCE.

### Re: GEMACO sentence about an unestimable parameter

CHOQUET wrote:If first initial state probabiities are not separately estimable then
use a constraint of equality like t(1:3,4_14) which set equal time 1 to 3.

Rémi

OK. I didn't know it.
Thank you so much.
simone77

Posts: 148
Joined: Mon Aug 10, 2009 2:52 pm