Parameter estimates using covariates

[Alex Rasiulis]

Dear forum,

My data set has two age groups, yearlings and adults. However the adults have been monitored from 1993 to 2011 while the yearlings have only been monitored since 2008 (to 2011). If I include all my individuals in the same model I obtain 19 estimates (1993-2011) for both my yearlings and adults.

My question is, when I start adding covariates (such as average temperature or precipitation) the estimates for the years with no data (for the yearlings) start to make sense (biologically). Is this because MARK uses the covariates to obtain the estimates even without any CMR data?

Thanks you very much!

[bacollier]

Dear forum,

My data set has two age groups, yearlings and adults. However the adults have been monitored from 1993 to 2011 while the yearlings have only been monitored since 2008 (to 2011). If I include all my individuals in the same model I obtain 19 estimates (1993-2011) for both my yearlings and adults.

My question is, when I start adding covariates (such as average temperature or precipitation) the estimates for the years with no data (for the yearlings) start to make sense (biologically). Is this because MARK uses the covariates to obtain the estimates even without any CMR data?

Thanks you very much!

Alex,
Couple of questions:

1/ what model are you using, and are you setting up your enc histories correctly (this is typically the usual suspect)? Since you did not catch any yearlings except in last few years, I assume you are using the '.' notation for all those periods, such as

111111 Adults
...111 Juveniles

2/ are you fixing the parameter value for the period when you have no data. You should not be getting estimates for periods when their is no data (are you sure you are getting time-specific estimates, or just a single estimate repeated for each time period)

3/ I suspect that when you add covariate what you are seeing is a estimate based on the data you have, not on time-specific, covariate specific predictions, but without more info that is a guess.

bret

[Alex Rasiulis]

1) I am using a known fate model because this is radio collar data. And no, I am not using the "." notation, I was not actually aware of this. I am simply using "00" for time intervals before capture.

2) I have not been fixing the parameter (I only have survival) for the time periods before the collar data for the yearlings. Without the covariates, the estimates are all identical for time intervals without collar data, however when I add the covariates, the estimates become time specific (and realistic). This is actually what led me to believe that these estimates are calculated as a result of the effect that the covariates have on the survival.

Thank you very much
Alex