QAIC and number of parameters

[Alan Dextrase]

I am dealing with overdispersed data and consequently using QAICc for model selection. Burnham and Anderson (2002) state that for overdispersed data, K = the number of parameters plus one to account for the estimation of the overdispersion parameter c-hat. In Presence, QAIC's appear to be calcualted using the same value for K as AIC's. I am wondering if the value for K in calcualting QAIC should be (number of model parameters+1). Thanks for any help you can provide.

Alan Dextrase

[jhines]

Alan,

That sounds reasonable. By default, program MARK doesn't add 1 when computing QAICc, but it does have an option which can be set to do it. I had not thought of it before, but will add it to PRESENCE (as an option). In the meantime, you can copy the table of results in PRESENCE to the clipboard, paste in Excel, then compute QAICc using the modified number of parameters.

Jim

[Alan Dextrase]

Thank you Jim,

I have recalculated the QAICc's and although the relative ranking of the candidate models did not change, there are changes in the deltaAIC's and model weights, which in turn impacts model averaging results.

Alan