Sample size of groups in multistate model (simulation?)

questions concerning analysis/theory using program MARK

Sample size of groups in multistate model (simulation?)

Postby quinn.c » Mon May 09, 2022 12:51 pm

I am interested in looking at sex-specific survival and transition probabilities for our seabird species. We monitor five breeding sites and so I am planning to use a multistate model.

From earlier research, juvenile survival is approximately 68% and adult survival 84%. Dispersal probability (natal site to new breeding site) is variable (0-20% of each cohort) with frequent pre-breeding transitions observed prior to age seven. Transitions are extremely rare after age seven. Survival, recapture, and transition probabilities differ between breeding sites. It is unknown if these parameters are time-specific.

Our species can’t be sexed visually, so we will be using genetic techniques to determine an individual’s sex. We have genetic samples from a subset of years from our study (9 of 20 years) and are planning to pick six cohorts to sex. Due to limitations of samples (100-300 per year) and budget, we are trying to optimize the sample size of sexed birds. First, is there a rule-of-thumb that can be used as a jumping off point? I am also aware of the simulation capabilities of MARK, but had a few questions regarding the set up for our particular study system:

Can I run a simplified simulation and still get a realistic estimate of our precision with different sample sizes? I.e., if I use two sites and fewer occasions. If not, this gets too unwieldy to run in MARK itself and I’d appreciate any pointers for multistate simulations in R (I plan to use RMark for my analysis).

In the PIMs, if I set the parameters to be the same number among groups (i.e., sexes), would that eliminate group differences? This is important for years where we won’t know the sex for any individuals, so group differences aren’t relevant.

I’ve included photos of my proposed PIM structure for the simulation (below). This is a reduced example (only 2 islands and 8 years). Logic behind my PIM structure:

1. Three groups: male, female, unknown-sex birds
1a. Note: most genetic samples were collected as chicks; so, in most cases, I don’t have to worry about longer-lived/more encountered birds being known-sex compared to birds captured once

2. Two states (breeding sites)

3. Individuals sexed in 1997 and 1998 cohorts
3a. Other cohorts function as single group (unknown sex); will also have unknown sex birds within known-sex cohorts

4. Recapture probability varies between sites but constant among sexes
4a. Possible this varies with time (this is unknown)

5. Survival varies by island, sex, age, and time(?)
5a. In this example, age/stage effect: 0-1, 2-4, and 4+
5b. This stage-class design may be its own rabbit hole but reflects a mostly unobservable juvenile state, a prospecting state (not settled to breed, but observed at the sites), and then a settled, breeding-state

6. Transition varies by island, sex, and age (0-2), fixed to 0 after age 2
6a. This represents a reduced example of the very rare transitions observed after age 7 in our study system

The design matrix is also a bit confusing because there aren’t age or group effects interactions in all years. Additionally, there are many potentially relevant interactions among time, age, sex, and site.

I would appreciate any help!

ImagePIMs_S_Ex by Quinn C, on Flickr

ImagePIMs_Psi_Ex by Quinn C, on Flickr

ImagePIMs_p_Ex by Quinn C, on Flickr
quinn.c
 
Posts: 2
Joined: Thu May 05, 2022 8:58 am

Return to analysis help

Who is online

Users browsing this forum: Google [Bot] and 14 guests