GOF testing - missing test 2 and 3

[wijw]

I have a dataset with encounter histories (3 occasions) from 42 individuals, 2 sexes, and two individual covariates.

I wanted to run a CJS model to estimate recapture and survival probabilities for the individuals and I attempted to run a general model phi(g*t)p(g*t). When I tested the GOF using Program RELEASE within MARK, it only provides me with the result from TEST 1. I cannot find TEST 2 and 3 in the notepad.

I cannot figure out what I am doing wrong. Any help is appreciated. Thank you.

[cooch]

I have a dataset with encounter histories (3 occasions) from 42 individuals, 2 sexes, and two individual covariates.

Meaning -- you have very little data. Sorry, but there is not much you're going to be able to infer from such a limited data set (MSc study?)

I wanted to run a CJS model to estimate recapture and survival probabilities for the individuals

With only 3 occasions, a fully time-dependent CJS model will give you only one estimate of survival, and one recapture estimate. If your objective (I'm guessing) is to compare males and females, you have almost no 'degrees of freedom' to say much of anything.

and I attempted to run a general model phi(g*t)p(g*t). When I tested the GOF using Program RELEASE within MARK, it only provides me with the result from TEST 1. I cannot find TEST 2 and 3 in the notepad.

You don't have enough occasions to generate those tests. In fact, you simply don't have enough occasions worth of data to do a lot. Full stop.

[wijw]

Thank you for the quick reply.

I believe this issue can be resolved by adding another survey year and increasing the number of recaptures through increased survey efforts. Is there a minimum sample size necessary to run Program MARK? (I don't think I came across a range of numbers/recommendations in the book.) I simply know that a bigger sample size is preferred, but how big?

Thank you.

[cooch]

In my opinion, unless you've done some manipulative experiement (which generally yields stronger inference than 'post hoc' story-telling based on time-series of encounters), you need at least 5 sampling occasions for open-population CJS models to come up with something reasonably useful (for a fully time-dependent CJS model, 5 occasions = 4 intervals, from which you could generate 3 estimates of survival and 3 for encounter probability). For robust design models (chapter 15), you can potentially get away with fewer (say, 3 primary sessions, at least 2 secondary smples per primary).

Largely, it depends on your question(s). For example, if you're doing a closed population abundance estimation in the presence of significant capture heterogeneity (Chapter 14), anything fewer than 6-7 sampling occasions is probably a waste of time. If you're using random effects to parse out process from sampling variation (Appendix D), you probably need at least 10 occasions. And so on. Much of the time, the data needs for many models outstrip what is possible given short-run (often, thesis-based) data sets.

The question(s) concerning adequate 'sample size' means 2 different things in the present context -- how many sampling occasions, and how many marked and released individuals per sample. To get a sense of what is appropriate for a given set of 'study questions', its worth simulating the problem -- Appendix A in the book.

[wijw]

This is very helpful! Thank you.