DM and PIM discrepancy

[vailama]

Dear all,
I have radiotracking data about male and females for 7 years (1997-2003). I want to estimate survival with known fate model. To obtain data on interactions among group, I prepared the inp file considering female from 1997 female from 1998 ….and so on until the last group which is male03, obtaining a total of 14 groups for 15 occasions. I run the full model using:
1. pre-defined model with PIM specifications
2. pre-defined model with DM specifications
3. full model constructed editing DM by myself
The link used is always the logit link.
All 3 models return the same values of AIC, n. of parameters and deviance but different parameters estimates. So what to do? Any help or suggestions would be greatly appreciated.

[cooch]

Dear all,
I have radiotracking data about male and females for 7 years (1997-2003). I want to estimate survival with known fate model. To obtain data on interactions among group, I prepared the inp file considering female from 1997 female from 1998 ….and so on until the last group which is male03, obtaining a total of 14 groups for 15 occasions. I run the full model using:
1. pre-defined model with PIM specifications
2. pre-defined model with DM specifications
3. full model constructed editing DM by myself
The link used is always the logit link.
All 3 models return the same values of AIC, n. of parameters and deviance but different parameters estimates. So what to do? Any help or suggestions would be greatly appreciated.

The classic DM that MARK defaults to uses a reference-cell coding where the final parameter is (typically) the reference against which other levels of the effect are tested. Sometimes, this poses a problem, IF the final parameter is a confounded (i.e., not separately identifiable) parameter (what is call a 'Beta' parameter in the Lebreton et al. monograph). This is described in the 'sidebar' begining on p. 60 of chapter 7 (sidebar titled 'design matrix coding and estimability').

So, quick suggestions:

1. stop using the pre-defined models. You shouldn't be using them at this point. They're intended only as short-cuts, and only if you know exactly what they are doing.

2. try changing the reference cell coding convention you use in your design matrix, and see if/how that my change the number of estimated parameters.