Survival estimates with variable treatments

[nzwormgirl]

I am trying to estimate the survival of ducks in an age-structured (marked as young and adults) CJS with sex and a predator control treatment. Some birds were exposed to predator control for all 6 years of the study, while others had no predator control for the first three years and then predator control for the last three years.

I was wondering if someone could give me some advice on how to approach the variable treatment issue. Is it possible to code this as a covariate using a treatment*time type interaction in the design matrix or am I better to do two separate analyses; one for predator control and one without.

Thanks, Amy

[jlaake]

Sounds like what you need is a time varying covariate. Create 6 variables in your input data -- one for each year and then give them the value 0 for no predator control and 1 for predator control. Some will have all 1's and then others will have 3 0's and then 3 1's based on your description. Lets say you named them p1,p2...p6, In your design matrix for phi, in one column for the rows for phi1, phi2... add the variable p1,p2...

Hope this helps. --jeff

[gwhite]

Amy:
Jeff's solution is the hard way with individual covariates. You just need a temporal covariate that you enter directly in the design matrix. All you have to do is create a design matrix from the 2 groups and the appropriate time model. To estimate the predator control effect, you will have to assume something about either equal groups or equal times. Then, use a single column with 1's for years and groups to indicate predator control. This will create a single column that is a 1 df estimate of the effects of predator control, or more complex models that estimate the predator contol effect separately for each year, or each area, or both.

You definitely need to read the Gentle Introduction.

Gary

[cooch]

Amy:

You definitely need to read the Gentle Introduction.

Gary

I agree - Jeff needs to review the 'Gentle Guide'.


Gary is correct, of course - to some degree, your 'problem' is quite similar (structurally) to the Dipper flood example. Have a look at Chapters 4 & 6, especially.

[cschwarz@stat.sfu.ca]

Before going too far down the analysis route, you indicate that some birds have predator controls and some do not.

Exactly how was the experiment performed. For example, if different islands were used some with predator control and some without predator control, then the experiment unit is the island and NOT the bird and simple covariates are NOT appropriate!

It is hard to believe that each bird is individually ``treated'' with predator control or no predator control!

I'm currently working on a paper showing how to analyze designed experiments using MARK. One example that I have is a skink experiment where different sites were subject to different manipulations and the manipulations changed over time in a BACI design. For example, all sites in year 1 had no manipulations. In year 2, different manipulations were applied to sites. Your design may be similar with 3 years pre and 3 year post treatment.

Site is the experimental unit and skink is the observational unit.

I can send a (very rough) working draft showing how such designed experiments can be analyzed in MARK. I'll be presenting the results in St. Andrews at the July meeting.

[jlaake]

Okay your turkeys! Obvioulsy I can't answer to this list before I've had my coffee in the morning, Note they didn't say my solution wouldn't work. It would give the exact same results as splitting into groups and splitting the times for the group with both predator and non-predator time periods, However, the one drawback with what I suggested would be that it would be much slower to run with individual covariates.

--jeff