Using survival estimates from MARK as GLM response variable

[simonk]

My first post here, so take it easy with me.
We have long-term mark-recapture data (>50 years) from a snake population in Sweden. We want to know if winter survival are affected by temperature and snow depth (and later on some additional covariates).

We analysed annual survival rates (and CI's) in MARK which varied from 0.3-1.0 across years. I have three questions:

1) What do you think of using these survival rates as a response variable for further analysis in R includning several explanatory variables, and use CI as weights in the GLM model?

e.g.
snake<-glm(MARKsurvival~ snow+meantemp, data=sdat3, weights= (upperCI-lowerCI)

2) Sometimes the upper CI from the survival rate estimates is much higher than 1.0. Is this a problem?

3) One article states that: "For adult survival, estimates were omitted because it was indistinguishable from a survival rate of 1 (i.e. not robustly estimated)". Is this really ok? I we do that our correlations with snow and temp are much more clear if we omit the survival rates of 1.0.

Thanks!

[cooch]

My first post here, so take it easy with me.

Not to worry. Welcome!

We have long-term mark-recapture data (>50 years) from a snake population in Sweden. We want to know if winter survival are affected by temperature and snow depth (and later on some additional covariates).

We analysed annual survival rates (and CI's) in MARK which varied from 0.3-1.0 across years. I have three questions:

1) What do you think of using these survival rates as a response variable for further analysis in R includning several explanatory variables, and use CI as weights in the GLM model?

e.g.
snake<-glm(MARKsurvival~ snow+meantemp, data=sdat3, weights= (upperCI-lowerCI)

No -- this is generally not valid. You should rarely (if ever) 'do statistics on statistics'. In fact, for the sort of problem(s) you seem interested in, you can (and should) do this entirely within a linear models framework in MARK. If you haven't already, I strongly suggest your read Chapter 6 in the 'Gentle Introduction to Program MARK'. I'm *guessing* that you have estimates of survival from a very basic model (or set of models). The next step is to evaluate the affect of covariates in a linear models framework.

If you're really new to MARK, I'd *strongly* advise you to slow down, get the MARK book, and work through chapters 1 -> 6.

2) Sometimes the upper CI from the survival rate estimates is much higher than 1.0. Is this a problem?

The problem is more likely that you're looking at the beta estimates, not the real probability estimates. MARK constrains the CI on the real probability scale to fall between [0,1]. This is covered in detail in Chapter 4 of the book referred to above (yet another clue that you might want to spend some time going through 1 -> 6 in the Book).

3) One article states that: "For adult survival, estimates were omitted because it was indistinguishable from a survival rate of 1 (i.e. not robustly estimated)". Is this really ok? I we do that our correlations with snow and temp are much more clear if we omit the survival rates of 1.0.

Thanks!

That depends. I'd not worry about that yet. Focus on increasing your skill set (chapters 1 -> 6), and go from there.

[simonk]

Thank you Evan for your effort and clarifications.