Unbalanced design

[megpetrie]

Hello,

I have a theoretical question regarding my MARK data analysis.

I've been analysing a data set looking at survival and
recapture rates of steelhead from various populations and three
rearing types over five years. I do not have a balanced design,
meaning that I have different populations and rearing types for
different years (though some populations I have data for every year).
I know that with ANOVA one is not permitted to legitimately test for
effects between blocks (in my case years) if the treatments do not
overlap. Is this true in MARK as well for linear models? For example,
I'd like to test for rearing type effects, but I do not have paired
treatments within years, just between years. If I use year as a
factor, could I also legitimately include rearing type as a factor in
a candidate model?

Thanks so much, Megan

[dhewitt]

Since no one else replied...

Your question isn't one about MARK (capture-recapture in general really) vs. ANOVA, but rather one about proper design of field studies. Since you are interested in estimation rather than testing silly null hypotheses as in ANOVA (congrats!), the issues about what you can "test" are subjective. If arguments about why you think it is OK to compare, e.g., popn B under rearing treatment A in year 2 with popn A under rearing treatment B in year 1 can pass the straight face test, then there is a way to compare them. You could ignore year entirely and do it in a model selection framework (I think, without knowing more about the available data).

As to "If I use year as a factor, could I also legitimately include rearing type as a factor in a candidate model?," the same issues arise as in any linear model context.