Unequal Time Intervals Clarification Needed

[Terri Donovan]

Hi.
We are working on a robust design model in which the primary periods have unequal intervals. We've worked through the interval.inp file that accompanies the sidebar in Chapter 4 (which demonstrates the unequal time interval property in Mark). The example is very straight-forward (thanks Evan!). There are 6 primary periods but they differ in length (5 or 7 months), and by using the "set time intervals" button you can estimate constant monthly survival well. Question: Do you need to set the time intervals for each model run (how did you get a model with time intervals 5, 7, 5, 7, 5, 7 and a model with time intervals 1,1,1,1,1 in the same results table? Pg 4-30 top)?
Many thanks for your help.
Isaac and Terri

[cooch]

Hi.
We are working on a robust design model in which the primary periods have unequal intervals.

Not sure what happened, but the latter half of the original question seems to have been truncated. I'll 'cut-and=paste' here:

(In the example - interval.inp) there are 6 primary periods but they differ in length (5 or 7
months), and by using the "set time intervals" button you can estimate
constant monthly survival well. If you take this example one step further
and try an analysis where, say, interval 4 is uniquely estimated (and the
other intervals have the same survival rate), the output provides two
survival estimates. Interval 4 (which is 5 months long) is estimated at
0.8935148 and the other intervals are 0.9476936 (close to the simulated
monthly survival value). Question 1: Why is 0.89 so far off the simulated
value of 0.95?

In the simulated data (interval.inp), the 'true' intervals are 7, 5, 7, 5, 7, 5. It sounds as if the model you're fitting would have a PIM structure that looks like



But, I don't get estimates anywhere near what you report for parameters 1 or 2. So, you need to clarify what model you're in fact talking about (the phrase 'identified uniquely' isn't particularly informative - when in doubt, specific the PIM structure in some way). Posting details of the PIM for survival, and the time intervals specified, would help.

Question 2: Does 0.8935148 represent monthly survival
estimate for period 4 or does this represent the 5 month survival rate
(which translates to a .978 monthly estimate).

Again, I have no idea where the estimate of 0.8935 comes from...and as such, can't answer the question.

Question 3: Do you need to
set the time intervals for each model run (how did you get a model with
time intervals 5, 7, 5, 7, 5, 7 and a model with time intervals 1,1,1,1,1
in the same results table? Pg 4-30 top)?

You can't reset the intervals for each model run (not directly, anyway - there is a way you can 'trick' MARK to do this, but not worth elaborating here - Hint: save the models, and manually edit the intervals, then run the saved models). More to the point, the models shown in the browser at the top of p. 4 - 30 were both run using 5/7/5/7/5/7 -- this was noted in several places where it tells you to 're-run the analysis' - after changing the intervals (which is implicit). In other words, start over, re-set the intervals to 5/7/5/7/5/7, and re-run both models.

[Terri Donovan]

Hi Evan,
Actually, sorry for the post. Unfortunately you responded to a question that was edited. Just wanted to clarify that when you specify unequal time intervals the output is automatically adjusted for the time interval -- which is clear now. Thanks for the info about modifying the fpt file and re-running the models with the corrected time interval. That was very helpful.

[cooch]

Hi Evan,
Actually, sorry for the post. Unfortunately you responded to a question that was edited. Just wanted to clarify that when you specify unequal time intervals the output is automatically adjusted for the time interval -- which is clear now. Thanks for the info about modifying the fpt file and re-running the models with the corrected time interval. That was very helpful.

Makes sense now - the original post generates the email. I've flipped a few switches so the edits don't generate a new email.

Yes - output automatically accounts for the intervals. Glad you have it sussed out.