I am having issues with what seems like relatively simple analyses of quite simple data.
I am estimating survival for a migrant bird from data I collected from colour-ringed birds on the wintering grounds in Africa. I have data over 3 winters and focussing on one site for this analysis.
I want to estimate both overwinter and summer survival. Over winter survival is when they are on the wintering grounds (Sep to April) and summer survival is when they are in Europe breeding (May - Aug). I only see them over winter. From prior analyses I am confident that if a bird returned it was detected - all birds across the study returned to exactly the same territory and the ability to detect local dispersal was high. Over the entire study, no individuals have been found to move territories between years.
Birds were ringed in the second half of winter 1 (n = 28) and throughout winter 2, but mainly before december (n = 25). I resighted throughout winters 1, 2 and 3. 15 birds from winter 1 returned in winter 2, and 7 of these 15 returned also in winter 3. 15 of the 25 birds ringed in winter 2 returned in winter 3. This gives return rates of 0.56/0.57.
Each bird was recorded whether to be present (1) or absent (0) during the start and end of each winter (start = 1 September to November, end = March to May) from their capture until the end of winter 3. I can therefore estimate survival over 4 intervals: the summer following winter 1 (end of winter 1 to start of winter 2); the middle of winter 2; the summer proceeding winter 2 (end of winter 2 to start of winter 3); and the middle of winter 3.
I also have whether a bird is a first winter or adult bird at capture (and of course returning birds from winter 1 are adults when they return). I am interested in looking whether there are differences in survival between adult and first winter birds.
My main goal is therefore overwinter survival, as I have good estimates of summer survival (i.e. return rates). My assumptions are that if a bird returns between years, it will be found; however in some cases a bird was not resighted at both the beginning and end of winter.
I have run the analysis in MARK on a number of models, starting with both phi and p to be time-dependent. I have looked at how phi varies with age (I don't have a biological explanation as to why p would vary with age), how phi and p vary with age at capture, and how phi and p vary with season (summer or winter).
My final model is phi(season) p(.). This makes perfect biological sense and supports my other findings: overwinter survival doesn't vary with age but differs between summer and winter, and recapture (resight in my case) probability does not vary with time. I have high overwinter survival (which I expect) and lower summer survival (again, expected).
My problems are these:
- firstly the estimate for summer survival is 0.30, but I know it to be at least 0.56. Why such an underestimate? All I can think is that the model is not 'excluding' winter 1 birds that did not return in winter 2 from return rates for winter 3 (this cannot happen - if a bird did not return in year 2, we assume it is dead)
- the upper and lower bounds for overwinter survival (phi 2) are 0.001 to 0.99!! The estimate of 0.94 is good - I expect it to be high, but why such huge SE and CI?
Any help would be really appreciated. Unfortunately I've had to pick this up quickly and from scratch without help and troubleshooting the MARKBOOK is not bringing me any joy. I don't know where to go from here as this is the limit of my understanding.
Thanks in advance
Emma