Time Since Marking DM

[simone77]

Hi all,

I have a dataset with 6 groups, 15 occasions. I have first modeled by PIMs the effect of age*time on apparent survival of each group (TSM for phi) and the effect of time on probability of capture of each group (CJS for p). So the model is: {phi(group*age2-t/t) p(group*t) PIM}.
I have fixed at zero all the p values corresponding to two occasions with no capture. Run the model with logit link function.

After I got the identity design matrix of that (PIM) model and set up the same model in DM. Something is wrong because the deviances are very similar but not identical.
Both of them have the same p values fixed at zero and logit function.

Some details on the way I set the DM model up:

(i) The number of phi columns should correspond to this sum:
Logit(phi)= 1 beta intercept + 5 beta group + 1 beta age + 13 beta time + 13 beta group1*time + 13 beta group2*time + 13 beta group3*time + 13 beta group4*time + 13 beta group5*time + 13 beta age*time + 5 beta groups*age + 13*5 beta group*age*time (group1*age*t, group2*age*time, ecc.) = 168 columns

(ii) The number of “P” columns should correspond to this sum:
Logit(p)= 1 beta intercept + 5 beta group + 13 beta time + 13*5 beta group*time (group1*time, group2*time,ecc.) = 84 columns

So the overall number of columns should be 168+84= 252 that is more columns with respect to the 246 columns corresponding to the PIM model.
I believe that I could be doing something wrong in the structure more than in the way I draw the 01 values but can’t find the error, any hint?

Thanks for any response

[simone77]

I answer myself this question because I have been doing many different trials by changing the way I coded the DM and I believe to have found the reason. It was depending on which reference interval/occasion I was choosing to be the reference one. I have just confirmed that by reading another recent post ( viewtopic.php?f=1&t=1729&p=5097&hilit=age#p5102 ).

The only difference with respect to what said in the other post is that, in my case, the same deviance (between PIM and DM) was obtained when the first interval was chosen like the reference one (but using the final interval, as I initially did, gave a slightly different deviance).