Partial continuous variable for S estimation

[scha]

Hi all,

I am currently running models with Rmark to investigate senescence on birds. The discrete age model on survival (S(age)) give strong support of aging in my population. Nevertheless, to reduce the number of parameter estmates, I wish :
- let the the three first age being estimated independently (one survival estimate by age)
- create class to group age between 4 and 15 (with the command bins I suppose)
- to fit an affine curve only for the age between 15 and 35 years old

Create age classes is really easy with RMark but i don't know how can I mix discrete and continuous function for age variable to estimate survivais.


Thank in advance for the responses

SC


PS : For details, I use multi strata models with dead recoveries...

[jlaake]

Sure. You didn't say if they start at age 0 or 1 but I'll let you work out details. S is an interval parameter so the value of age depends on age at beginning of time interval. Assuming it starts at 1, then use 0,1,2,3,4,36, so you'll have 4 groups ((0-1],(1-2],(2-3],[4-35]). Pay attention to setting right=TRUE. Then create a 0/1 variable say p16 which is 1 if age 16 or older and 0 otherwise. Formula will be something like

~age +p16:I(Age-15)

It should have 5 parameters, Intercept, age2,age3,age4-35 and slope for age 16+. For example,

logit(S(1))=Intercept
logit(S(2))=Intercept+beta_age2
logit(S(3))=Intercept+beta_age3
logit(S(4))=Intercept+beta_age4-35
logit(S(15))=Intercept+beta_age4-35
logit(S(16))=Intercept+beta_age4-35+beta_p16
logit(S(20))=Intercept+beta_age4-35+5*beta_p16

--jeff