23 yr tern study diffs in adult_juvenile_ cohort survival

[Lachie Mc]

I would like to expand on some issues that Dr Laake provided me with advice for last yr. I have just returned to this analysis and further advice would be greatly appreciated

I have recently completed a 4 yr mark recapture study for a tern population. All individuals were marked as chicks (batch released) between 1985 and now, but adult recaptures were undertaken in 4 yrs btw 2004 and 2008 ie 4 occasions on betw 19-23 cohorts and estimates of age are available for all individs. between 450 and 500 adults were captured on each occasion.

Calculating adult survival seems intuitive enough. Obviously the first interval between chick release and 1st recap incorporates juvenile history and may vary as a function of time and age (and cohort) and may not be comparable (but see below). The second and third intervals btw 2004-2005 and 2005-2006 may reflect real enough estimates of adult survival (obviously 2006-2007 cant be estimated) and can be modelled as a fn of sex, age etc.

My question re this anlysis of adult survival is: if my data are too sparse to analyse for age diffs in surv, should I pool the data into 5yr classes (1-5y) (5-10y) (10-15y) (15-20y) and apply the yr classes as covariates (1, 2, 3, 4) or as groups ? given that I would like to also include sex as a group ie which should be set up as group first sex or age classes ?

Secondly survival of seabird chicks is highly variable. Presumably prey or environmental conditions play a big role in influencing chick survival. I would like to investigate whether differences in cohort survival between release and 1st recapture in 2004, 2005 or 2006 reflect the conditions experienced as chicks ie covariates of SST, prey biomass. If I find diffs in survival btw cohorts as a fn of such covariates, do you think that survival estimates btween cohorts are comparable given that the time interval between initial release and first recapture in 2004 varies as a function of age? The age structure, and anecdotal evidence suggests chicks in 1995 and 1998 suffered higher mortality rates because of decreased sardine abundance but I am concerened I am making too many assumptions.

ALso should the input file use . (dots) for years when adults werent recaptured ie 1985-2004) or 0's or do 0's go into the PIMs. ie what exactly do you do for input files and PIMS for parameters that cant be estimated?

Any help or advice much appreciated

regards

Lachie

[abreton]

You may get a more efficient response if you refine some of your questions, but I'll try to respond to issues that you highlighted and I could interpret.

"if my data are too sparse to analyse for age diffs in surv, should I pool the data into 5yr classes": For this "adult survival analysis" you say you have ~450-500 adults marked per occasion. Assuming your recapture probabilities are above ~0.6 your sample size should be more than adequate to detect a moderate to large age effect. However, I wonder how you aged terns caught as adults? Perhaps by age you mean time since initial capture? If the latter is true, you need to be careful here not to confuse readers - see general suggestions regarding TSM (time since marking) models in Chapter 8 Section 8.1.3 in a Gentle Intro to MARK. Regarding the value of true age versus TSM or some other proxy of 'age' obviously the best metric for age is the actual age of the individual. Each incremental step away from this 'reality' including your age classes is increasing less likely to account for variation in the data that are a function of age.

"apply the yr classes as covariates...or as groups": It is possible to fit them as [i]individual[/i] covariates or groups. One factor to consider is how many groups you'll have if you fit classes and sex this way - 2 x 4 classes = 8 groups. This is manageable, but as groups approach ~10 or more I would start considering fitting binomial grouping factors as covariates - e.g., sex (male/female = 0/1 dummy variable). I would also consider a priori whether I thought my binomial grouping factors, e.g., sex, might have contributed important variation to the data [i]and[/i] whether or not my sample was large enough to detect it. If I suspected a weak effect and/or a sample too small to detect the effect, I would likely proceed to fit (e.g.) sex as an individual covariate. This way, it could easily be dropped leaving (in your case) four groups rather than eight to manage for the remainder of the analysis. Another factor to consider is goodness-of-fit. You may want to assess GOF for all 8 of your groups which would not be possible if you coded sex as an individual covariate. See additional suggestions in previous posts and a Gentle Intro.

"given that I would like to also include sex as a group ie which should be set up as group first sex or age classes": See Chapters 6 and 7 in a Gentle Intro. If you group-on sex and class, then you'll have 2x4 = 8 groups. If you group-on class and fit sex an an individual covariate (see Chapter 12 in Gentle Intro), then you'll have 4 groups.

Regarding "chicks" that were "batched marked" you wrote "I would like to investigate whether differences in cohort survival": Assuming these were batched marked, i.e., chicks were marked with a batch mark rather than individually identifiable marks, then you're analysis options will be extremely limited. Capture-mark-recapture models generally require units are individually marked, but see http://www.stat.sfu.ca/~cschwarz/Peters ... /paper.pdf. Note that the author offers this as "a work in progress [which] could change in the future".

" (dots) for years when adults werent recaptured ie 1985-2004) or 0's"or do 0's go into the PIMs": Assuming you use a single-strata model type, 0 = not encountered, 1 = encountered, (.) = no attempt was made to encounter marked individuals (e.g., recapture effort = 0 on that occasion). Asking whether or not 0's go into the PIMs suggests that you haven't read the first 7 Chapters in a Gentle Intro - this is required reading for success using MARK.

"what exactly do you do for input files and PIMs for parameters that cant be estimated?": Once you're comfortable with material in the first 7 chapters, try searching the forum archives for additional insights. If that fails, post specific issues that you're experiencing. FYI - Input files are discussed in Chapter 2 and PIMs are introduced in Chapter 3. Issues of estimability and convergence are discussed throughout Chapters 1-7 and elsewhere in the manual. I suspect you'll gain much in confidence by carefully reading these material.