setting up input file for multiple years, multiple groups

[jclaisse]

We are trying to get daily survival rates of juvenile fish(live recaptures only model). We sampled every few days for ~2 months each summer for 3 years at 3 sites. There were 10 to 15 sampling occations each summer(# of days between sampling events varied, but all sites were done on the same day when sampling occured). Fish are only at this stage for a couple months so there is no chance an individual from one year would show up the next year. I would like to look at the year effect and site effect on survival seperately and together and I'm tying to figure out the best way to make my input file. So far I have made my encounter history set up for each individual, with all sampling events for all three years in a row and just filled in with 0's for the other 2 years not sampled for any particular individual.

Something like this:
year 1_______year 2______year 3___coding for year and site
101010101010000000000000000000 1 0 0 1 0 0
000000000001010101010100000000 0 1 0 1 0 0
000000000000000000000011010101 0 0 1 1 0 0

The first one being for a fish sampled during year 1(site 1), second for a fish during year 2(site 1) and third for a fish from year 3(site 1).

Please let me know if I am on the right track and below are some specific questions I have:

1) I've read that the extra string of 0's in front of those sampled in year 2 or 3 is not a problem becuase they are not included in the liklihood, but the extra ones at the end (after a year 1 or year 2 sampled fish) could be? Should I use "." instead of 0's for those or deal with this in some other manner?

2) I plan to deal with the varying time between sampling events within each year the typical way by entering the number of days between sampling events, but how do I deal with the intervals between year 1 & 2 and year 2 & 3? (or does this even matter?)

Thanks so much for your help.

[jlaake]

It sounds like the number of occasions and intervals between occasions varied between years so what you chose is the best way to go. I don't think you need the year coding for grouos because of the way you structured the ch but obviously do need the site coding for groups. What I would do is to fix the Phis to zero for the time intervals between years. The time intervals between years will be meaningless because Phi=0 and because you don't expect to see them across years anyhow. There will be 2 Phis to set to zero (between year 1 and year 2 and between year 2 and year 3) per group.

1) I've read that the extra string of 0's in front of those sampled in year 2 or 3 is not a problem becuase they are not included in the liklihood, but the extra ones at the end (after a year 1 or year 2 sampled fish) could be? Should I use "." instead of 0's for those or deal with this in some other manner?

THe CJS model is conditional on first capture (release) so yes only the 1's and 0's matter after the first 1. You won't have any 1's past the year they were sampled and if you set Phi to zero for the intervals between years then it expects all 0's from that point forward.

2) I plan to deal with the varying time between sampling events within each year the typical way by entering the number of days between sampling events, but how do I deal with the intervals between year 1 & 2 and year 2 & 3? (or does this even matter?)

It won't matter if you set Phi=0.