Design matrix vs. PIM dead recoveries

[iwvining]

I am dong a series of dead recoveries analyses and I thought I had the design matrix right, but each time I run the design matrix and compare it to the PIM specified model, the AIC and associated weight and such are different. In the MARK book, Chapter 7, the design matrix looks to me to be the same as a full design matrix even though the diagonal is different in the PIM, however when I pull the PIM from the model I've specified in the design matrix, I get the correct PIM (with diagonal different than the upper right section of PIM). What am I missing?

[cooch]

I am dong a series of dead recoveries analyses and I thought I had the design matrix right, but each time I run the design matrix and compare it to the PIM specified model, the AIC and associated weight and such are different. In the MARK book, Chapter 7, the design matrix looks to me to be the same as a full design matrix even though the diagonal is different in the PIM, however when I pull the PIM from the model I've specified in the design matrix, I get the correct PIM (with diagonal different than the upper right section of PIM). What am I missing?

First, the DM doesn't change the structure of the PIMs. The DM is 'applied' (constrains) the parameter structure specified by the PIM's. Nothing you do in the DM will change the structure of the PIMs.

Second, have you had a long look at Chapter 8 in the 'MARK book'? Specifically, section 8.4 (which relates to analyzing individuals marked both as young and adults -- if you have only marked as young, you're out of luck anyway -- see section on this in the same chapter). Seber parameterization, or Brownie-Anderson?

[iwvining]

I am using the Brownie et al. group. I have looked at Chapter 8, which as you probably know, is missing anything with a DM in it (which is why I referenced Chapter 7 that shows the DM for multiple age models). I am doing an two age (adult and juvenile) type dead recovery analyses.

I have now run a few more of the models, and I have had a handful where the PIM and DM match in AICc,Delta AICc, AICc Weight, etc. when both model methods are run on the same design. From this and your comments, I thinking I am specifying things correctly, but I guess I don't understand why the PIM and DM estimates are not the same.

[cooch]

I am using the Brownie et al. group. I have looked at Chapter 8, which as you probably know, is missing anything with a DM in it (which is why I referenced Chapter 7 that shows the DM for multiple age models). I am doing an two age (adult and juvenile) type dead recovery analyses.

Indeed, because the assumption is by chapter 8 you don't need any more hand-holding on the DM.

I have now run a few more of the models, and I have had a handful where the PIM and DM match in AICc,Delta AICc, AICc Weight, etc. when both model methods are run on the same design. From this and your comments, I thinking I am specifying things correctly, but I guess I don't understand why the PIM and DM estimates are not the same.

Usually, when someone finds the 'PIM approach' and the 'DM' approach yield different estimates, its because of differences in the the link function. (sin by default for PIM, logit by default for DM-based models). But, that usually leads to differences in K, and thus AIC as well. If the model deviances (and or -2 log(L)) are the same between the two approaches, then I'm puzzled. If that is indeed the case, send me the .dbf and .fpt files offline, and I'll have a look.