fixing parameters

[ghorton]

I am attempting to analyze 3 years of 3 cohorts of tagged salmon. Although there is some overlap in time among cohorts occupying the study site (and therefore capture) before they migrate to sea, there is a significant period of time when I sampled in the absence of a younger cohort while older cohorts were present. For example: sampling in 2000 occurred in the absence of the 2001 cohort, but in 2001 the two cohorts overlapped.

In order to be as efficient as I can in the analysis, I would like to input all data into MARK one time (e.g., one input file) and only estimate parameters for the cohort when I know they were present and not before. I set up the input file to have cohorts as groups. My thought was to fix parameters for those time intervals in a group when that group was absent. First, before I tried fixing, I ran a model and MARK generated "reasonable-looking" parameter estimates (cv=10-20%) even for intervals before the cohort had recruited. It confused me as to why anything close to "reasonable-looking" should be estimated. Then, when I tried fixing phis to either 0 , 0.5, or 1 I got very different results and all had much higher AIC values than not fixing. I understand that in fixing parameters, the real parameters and not the betas are being fixed so there is still a cost to estimating parameters. However, I am unsure why the cost (AIC) should be different for different fixed values (0 vs. 0.5 vs. 1). Now I do not know what the best set of parameters to use are or even if my approach of keeping all cohorts together in the input file and fixing parameters is O.K.

Thanks in advance for the advice!