different AIC in browser than in model output

[Diego.Pavon]

Hi all,

I am running CJS models in MARK and I have notice a "weird" thin (to my mind).
After running fixed effect models and before adding the results to the browser, I am presented with a small pop-up window with the results (i.e. name, AIC, num par., deviance...) and of course, after clicking OK, the results are added in the browser (same values for AICc, deviance, etc..). However, after running Random Effects models, the information that appears in this small results window does not match with the AIC and deviance that, after clicking OK, is added to the browser. For instance, a Random effect model which according to the results wondow (AIC=1109) would be ranked as the "best" model, then when I click OK, is ranked far below with an AIC=1152.

I was just wondering if you guys see anything wrong there or I just have to trust what is added in the browser and forget about the information in the small "results window" that pops-up after running the models.

Thank you very much for your time

Diego

[cooch]

Hi all,

I am running CJS models in MARK and I have notice a "weird" thin (to my mind).
After running fixed effect models and before adding the results to the browser, I am presented with a small pop-up window with the results (i.e. name, AIC, num par., deviance...) and of course, after clicking OK, the results are added in the browser (same values for AICc, deviance, etc..). However, after running Random Effects models, the information that appears in this small results window does not match with the AIC and deviance that, after clicking OK, is added to the browser. For instance, a Random effect model which according to the results wondow (AIC=1109) would be ranked as the "best" model, then when I click OK, is ranked far below with an AIC=1152.

I was just wondering if you guys see anything wrong there or I just have to trust what is added in the browser and forget about the information in the small "results window" that pops-up after running the models.

Thank you very much for your time

Diego

As discussed in appendix 4, the AIC reported for a random effects model is based on number of parameters K, where



where G is a function of the mapping of the ML estimates and the shrinkage estimates, and is the number of free parameters not being modeled as a random effect.

What you se in the little popup window (that you click to add the model results to the browser) is the AIC before adjustment for . So, everything is fine -- all MARK is doing in the popup window is showing you the 'pieces', which are ultimately combined into the correct RE AIC that is shown in the browser.