time intervals

[Claudio]

I'm performing survival analysis in the framework of the CJS model. I have costant time interval between sampling occasions (4 days). I found that, when I run the analysis, I got different results in model selection when I have set time intervals =1 days (the best model is _.t_, the second one is _t._, delta AICc 0,44) or = 4 (the best model is _t._, the second one is _.t_, delta AICc 1,7). This is strange for me, as I expected that just the estimations of phi would have been changed. Why does it happen? Which is the right time interval (I suspect 4)?
Thanks in advance.

[dhewitt]

If your time intervals are constant, just use all 1s in the interval specifications. Your estimates apply across 4-day periods.

[cooch]

I'm performing survival analysis in the framework of the CJS model. I have costant time interval between sampling occasions (4 days). I found that, when I run the analysis, I got different results in model selection when I have set time intervals =1 days (the best model is _.t_, the second one is _t._, delta AICc 0,44) or = 4 (the best model is _t._, the second one is _.t_, delta AICc 1,7). This is strange for me, as I expected that just the estimations of phi would have been changed. Why does it happen? Which is the right time interval (I suspect 4)?
Thanks in advance.

In case you haven't had a chance to read it, the -sidebar- beginning on p. 27 of Chapter 4 of the MARK book covers a lot of the basics concerning handling intervals in MARK.

[Claudio]

Thanks for both your replies.
Dear prof. Cooch,
did you mean the sidebar beginning at pag 4-13 of the 6th edition of the book, or is it a new release of the gentle introduction?
In the meanwhile, I found that the problem is the number of estimable parameters, that vary with the time intervals (the deviances are the same). Is it possible that handling the time intervals bias the model selction? Has one to adjust the number of parameters?

[cooch]

Thanks for both your replies.
Dear prof. Cooch,
did you mean the sidebar beginning at pag 4-13 of the 6th edition of the book, or is it a new release of the gentle introduction?
In the meanwhile, I found that the problem is the number of estimable parameters, that vary with the time intervals (the deviances are the same). Is it possible that handling the time intervals bias the model selction? Has one to adjust the number of parameters?

Current version - references to the book are generally made wrt to whatever is currently online.