Best approach for group*time analysis

[tjk]

I've spent a fair amount of time searching through the forum postings and reading the "Gentle Introduction" book. Still having troubles so here it goes:

My data set consists of 2 groups of fish (river-released and reservoir-released) that were released during 3 years apiece (2008, 2009 and 2010 for river fish and 2007, 2009, and 2010 for reservoir fish). There are 3 capture locations including the release site, so the capture histories look like 100, 101, 111, etc. I'm interested in survival differences between the release site and the first monitoring location. I'm trying to figure out how to read the data in to test for interactions between the 2 groups and the years of the study.

Should I read the data in as follows with 6 groups (group*year):

101 1 0 0 0 0 0;
111 0 1 0 0 0 0;

or should year be used as a covariate?

Any advice would be appreciated.

Toby

[jlaake]

You can do it either way but I suggest that you use your first approach of 6 groups: 2 locations and 3 years. The pre-defined model Phi(g*t) is the full interaction model location*year*time. If you go with groups then you can use PIMS to construct the various models. For example location*time would be

Group1: Location 1 Year 1
1 2
2
Group2: Location 1 Year 2
1 2
2
Group3: Location 1 Year 3
1 2
2
Group4: Location 2 Year 1
3 4
4
Group5: Location 2 Year 2
3 4
4
Group6: Location 2 Year 3
3 4
4

Year*time would be

Group1: Location 1 Year 1
1 2
2
Group2: Location 1 Year 2
3 4
4
Group3: Location 1 Year 3
5 6
6
Group4: Location 2 Year 1
1 2
2
Group5: Location 2 Year 2
3 4
4
Group6: Location 2 Year 3
5 6
6

You could also construct additive models in which case you would use time pims that were different for each group and then construct a design matrix
for the additive effects.

[tjk]

That makes sense. Thanks for the assistance!

Toby