## Robust Design and Ages

Forum for discussion of general questions related to study design and/or analysis of existing data - software neutral.

### Robust Design and Ages

Hello All,
I've searched for a similar question but I couldn't seem to find any so here it goes....
For my data, an analysis of small mammal populations, I have been using a robust design model in RMark. I have 6 primary periods with 3 secondary periods each. I had the covariates sex, species, and treatment levels. My adviser now wants to add age as a covariate. At first, I thought this would be straight forward but because small mammals mature quickly, I have individuals that start as a juvenile and are adults by a different primary period. Is there an easy( lol nothing easy but logical) way to add this to the robust design model? I saw in chapter 7 of a gentle introduction that it is normally done with a cjs model a specific order of the PIMS but the way my data was collected was for a robust design model. I would greatly appreciate any tips or ideas, thank you in advance.
jln1234

Posts: 5
Joined: Tue Sep 01, 2020 2:34 pm

### Re: Robust Design and Ages

This is a solvable problem, but it depends on what age is meant to predict. If it's survival or temporary emigration, then you would do it the same way that you saw with the CJS model (2 groups: first captured as juvenile and first captured as adults). The first diagonal of the PIM for the young would be unique, but the rest of the PIM would match with the PIM for the adults (see chapter 7 of the book). The first diagonal applies to the first survival (primary period) interval after initial capture. After that initial interval the individual is now an adult, and therefore parameters for that group match with the group for released as adult.

If you want to model detection probability as a function of age, then I believe you could do it with a time-varying individual covariate on p/c (e.g., a 1 for the primary period when they are juveniles to offset them from adults, and a 0 when they are adults). Alternatively, you could do the entire analysis using the multistate robust design model, defining a juvenile state and an adult state. Those that start off as juveniles transition to adults with probability 1.0. If you also have temporary emigration (unobservable states), then your state structure would be more complicated but still doable.
Bill Kendall

Posts: 96
Joined: Wed Jun 04, 2003 8:58 am

### Re: Robust Design and Ages

Thanks for getting back to me!
I have a few follow-up questions and aspects I would like clarified just to make sure I understand. The age would be to look at survival. So going back and reading through chapter 7, I would organize my encounter history with individuals who were first marked as juveniles or first marked as adults. So the encounter history would look like
Code: Select all
`100100001111111111 A Bla B Perm. F`
or
Code: Select all
`100100001111111111 J Bla B Perm. F`

or with A representing adult and J representing juvenile. Or would it be best to have that in 1s and 0s? My next question is how do I actually incorporate that into the robust design in RMark? Looking at the Crmark section in the gentle introduction it looks like I would make it a design covariate then use add.design.data? The example uses bins? Would I then create bin 1 and 2 ( one for adult and the other for juveniles). If that is the way to do that in Rmark how does that tie into creating the PIM tables from chapter 7. I apologize for the bombardment of questions and clarifications.
jln1234

Posts: 5
Joined: Tue Sep 01, 2020 2:34 pm

### Re: Robust Design and Ages

At this point, this is (rapidly) morphing into an RMark question, so I'm going to split this thread, and move the last post, and those following, to the RMark subforum.
egc

Posts: 201
Joined: Thu May 15, 2003 3:25 pm

### Re: Robust Design and Ages

Hello,

I have another question about how to create an Age variable for p and c using the make.design.data and process.data functions, model = RDHuggins.

I'm looking at Time-dependent capture and survival probability, as well as interaction effects between Time and two Age groups. Gammas are fixed, no temporary migration. Constant p and c between secondary sessions within each primary session.

***The issue: But for p and c, I'm missing the Age variable. Is there a way to include Age in the ddl for p and c with functions in process.data and make.design.data?

I got the correct ddl structure for S, g' and g''. Where all different cohorts where defined and within all cohorts different combinations of Age and Time. A new age group variable (eggs, chicks) was created directly in the ddl: (Age =< 25 eggs, > 25 chicks).

All capture histories are registered first time at the egg stage, age = 0. There is not an option to be registered as adult at Time = 0-24 in Cohort = 0.

I tried to use the pim.type argument. But maybe there is functions in the process.data function I need to use. However, I could not use the group argument, since my data only consists of capture histories.

Code: Select all
`proc = process.data(Kaffeskjaer, model = "RDHuggins", time.intervals = time.intervals) ddl = make.design.data(proc)# later tried:ddl = make.design.data(proc, parameters = list(p=list(pim.type="all")),c=list(pim.type="all")))`

You have previously suggested the following solution around the topic Robust design and Ages: include a time-varying individual covariate. Where I use value 1 and 0 to indicate for each capture history if they are in eggs or chicks phase in every session? Would you recommend it in this situation if this problem could not be solved?

Best regards,
Sine
sinedh

Posts: 1
Joined: Wed Apr 05, 2023 12:43 pm

### Re: Robust Design and Ages

Even though this is somewhat (largely?) an RMark-coding question, I'll leave it here for now since it does follow the initial thread.
egc

Posts: 201
Joined: Thu May 15, 2003 3:25 pm

### Re: Robust Design and Ages

With the robust design, the secondary occasions are assumed to be closed and occur over a short period of time, so animals don't age which is why I didn't add an age variable by default. That said, upon reading this thread (which I missed before), I can see it would be useful to have an age variable that was the same for all secondary occasions but say split them by juveniles vs adults for each session.

So I naively started an example thinking that I could simply create an age variable in the design data. But it became apparent very quickly that wasn't possible because the PIMS are square for p and c and there is no way to know the initial session at which the critter was first captured. Thus there is no way to modify their age through the sessions as they get older. Which is probably the actual reason I didn't include age in p and c. It can't be done.

But Bill's suggestion in his earlier post is the solution. You need to include a time varying individual covariate to the data. It should be a numeric value although if you want to be age categories you can use dummy values for the age categories. In your data you need a variable named something like age.1,age.2,...age.k where you have k sessions. The value is the age of the critter for the kth session. Values of age prior to first capture can be set to zero. Then you can use the variable age in the model for p/c and RMark will put "age.1" for session 1 and "age.2" for session 2 etc in the design matrix (DM) and when MARK fits the model it will plug in the age value for each critter at each session into the DM when it computes the p/c values.
jlaake

Posts: 1409
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA