Reasonable pop estimates, but crazy CIs

[GNealeMT]

I searched the forum and did not find this topic covered. We are running relatively simple encounter history data with only 4 entries per animal. We have broken down the input files to only include the encounter history and group ID column, like this:

1011 1;

We have run 4 models on every parameter:

1) p(t)c(t) not equal
2) p ( c ) not equal
3) p.c. equal
4) p.c(t) not equal

Most often MARK likes #1 the least, but this may be due to the run generating incorrect CIs for that model, although the estimates themselves seem reasonable. Occasionally some of the other models have incorrect CIs too; they might be very large, negative values, or do not bracket the population estimate at all. Any thoughts on changes in input, or others experience with these models and crazy output?

[cooch]

I searched the forum and did not find this topic covered. We are running relatively simple encounter history data with only 4 entries per animal. We have broken down the input files to only include the encounter history and group ID column, like this:

1011 1;

We have run 4 models on every parameter:

1) p(t)c(t) not equal
2) p ( c ) not equal
3) p.c. equal
4) p.c(t) not equal

Most often MARK likes #1 the least, but this may be due to the run generating incorrect CIs for that model, although the estimates themselves seem reasonable. Occasionally some of the other models have incorrect CIs too; they might be very large, negative values, or do not bracket the population estimate at all. Any thoughts on changes in input, or others experience with these models and crazy output?

Encounter history structure is (now) correct. As for why the fully time-dependent model (your model 1) is not behaving, this is a known problem (with the model, not with MARK). See Chapter 15 in the book - specifically, sections having to do with 'constraints' on time-dependent models and 'identifiability'.

Searching the forum is useful, but - for many problems - especially for *new* MARK users - search 'the book' is the recommended first step.

[GNealeMT]

Thanks for the information ~ we'll have another look at the book. We've found that the model works for some analyses (same type of input formatting), and not others, so hopefully Ch. 15 will shed light on that as well. gn