Number of parameters and QAIC

[egmartins]

I have been studying the book by Burnham & Anderson on model selection and I have read that when calculating QAIC, the number of parameters (K) must include one for the estimation of c-hat. My questions are:

1) Does MARK automatically include one parameter for the estimation of c-hat when calculating QAIC? I am not sure about it because in the browser output window only the number of estimable parameters in the models are being reported for my analysis.

2) In case MARK does not include one parameter for the estimation of c-hat when calculating QAIC, should I add one more parameter for all models being compared or just for the model used to estimate c-hat (I am using the bootstrap procedure to estimate c-hat)?

Thanks a lot and I am sorry if these are stupid questions.

[bmitchel]

The default in MARK is to NOT include an extra parameter for QAIC. There is an option in File... Preferences to "Add 1 to K when computing QAICc with c-hat > 1". However, be careful if you use this; there is a bug that causes the number of parameters to continue to increase if you switch back and forth between c-hat = 1 and c-hat > 1, which could potentially foul up the information in your results browser.

Brian R. Mitchell

[prldb]

Has the bug refered to by Brain Mitchell been fixed? Why is Burnham and Anderson's recommendation to increase the model parameter count by one when computing QAIC not the default in MARK?