Season Covariate from Full-Year Data

posts related to the RMark library, which may not be of general interest to users of 'classic' MARK

Season Covariate from Full-Year Data

Postby MercyMelo » Wed Apr 05, 2023 1:55 pm

Hello everyone,

I tracked transmittered birds weekly throughout the year with the goal of using known fate models to estimate survival. My data therefore consists of an encounter history of 52 occasions and several individual covariates such as sex, age, and study site to build comparative models. I would also like to build a model that uses "season" as a covariate to see if survival is different between seasons (breeding, migration, overwintering). I have considered the following approaches:

1) Create separate rows for each individual in each season (ex. Bird1 Breeding, Bird1 Fall Migration, Bird1 Winter all as different rows), but this causes nonindependence of samples and differing numbers of occasions in the encounter history.

2) Add time-dependent covariates to specify which occasions are part of which season: Columns "Season1" through "Season52" would have either 0 (Winter), 1 (Spring Migration), 2 (Breeding), or 3 (Fall Migration). Unfortunately, I believe this would treat the values as continuous (as numbers 0-3) instead of categorical. From reading about time-dependent covariates, I have not found an instance where a categorical covariate was used in a time-dependent manner.

3) Add time-dependent covariates with dummy variables to compare only 2 seasons to each other: Columns "Winter1" through "Winter52" with either a 0 or 1 value; columns "Spring1" through "Spring52" with either a 0 or 1 value; etc.. We could then test if spring (or any season's) survival is different from all other seasons, but wouldn't be able to use "season" in our model comparison.

Could anyone guide me on a way to move forward with separating our full-year data into seasons in a way that would allow me to use season in our model selection approach? Does this require multi-state models?

Thank you very much in advance,

-Mercy
MercyMelo
 
Posts: 3
Joined: Fri Dec 02, 2022 3:02 pm

Re: Season Covariate from Full-Year Data

Postby jhines » Wed Apr 05, 2023 2:01 pm

I think #2 is what you want, but change the codes, 0,1,2,3 to "W'", "S","B", "F". RMark will treat them as categorical covariates, giving you an estimate for each season.
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Re: Season Covariate from Full-Year Data

Postby jlaake » Wed Apr 05, 2023 2:05 pm

Definitely not #1. You want to do this with design dat rather than time varying covariates. First question is whether they were all tagged at the same time or was it spread across several weeks to months? If they were all done at same time, it is dead easy because you just assign each time to a season. If spread over time you can use groups or as a last resort use dummy variables 0/1 for levels of season as a time varying covariate.

Jim's idea is fine if used in design data but not time varyingindividualcovariate which have to be numeric..
jlaake
 
Posts: 1417
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA


Return to RMark

Who is online

Users browsing this forum: No registered users and 10 guests