Missing survival estimates with time covariate models

[scha]

Hi all

Sorry to bother you again. I have a problem of missing estimates when I look the output of models using time-covariate in age models.

To give you a rapid description of my models. I have three age-classes (young, mature, old):

vaut.ddl.age$S$a0=0
vaut.ddl.age$S$a0[vaut.ddl.age$S$Age==0]=1
vaut.ddl.age$S$a4_27=0
vaut.ddl.age$S$a4_27[vaut.ddl.age$S$age_covar=="[4,28)"]=1
vaut.ddl.age$S$a28_37=0
vaut.ddl.age$S$a28_37[vaut.ddl.age$S$age_covar=="[28,37]"]=1

If I do time-covariate models with 3 different intercepts and 3 different slopes on survival as :

S(~-1 + a0 + a1_27 + a28_37 + a0:Tmax + a1_27:Tmax + a28_37:Tmax)

as expected, I have 6 beta parameters which are estimated (3 intercepts and 3 slopes) but when I look real parameter estimates, I see that some years (normally 31 years) are not estimated. It is the same years for all age-classes. For old age class, I cannot have estimates before 2003 but it is normal. But for example the survival estimates of 2009 are missing for all age-classes and I do not understand why.

When I use other time-covariates, some years were also missing (but not the same)

I do not have problems when I fit temporal trend (with ~Time)

Has someone already encountered this problem ? What am I doing wrong?

Why Mark cannot estimate these parameters?

Thank in advance for treply

stephane

[jlaake]

Did you check your design data to make sure it is all correctly specified? If it is and you are still having a problem you'll have to send the code and data to me offlist. Hard to tell otherwise. --jeff

[scha]

Hi jeff

thank for the reply

I don't see any problem in the design matrix

ok I will contact you offlist

stephane

[scha]

Hi all

I have discussed with Jeff offlist and he found the reason of my missing real estimates

the answer

When you have individual covariates the real values it provides are only for a specific set of values for those covariates. A simple example that is more obvious than your example, if you had a single covariate say mass that was a predictor for survival then there are many different real survival possible survival values but the output from MARK only provides the estimates for a single value of mass. To compute the real values for the various combinations of covariate values you need to use the function covariate.predictions."

In my data, I use temperature as covariate and for some year I have the same temperature, so the mark output only provided real estimates for one of these years.

I hope this could help someone

Stephane