Deviance = 0

[dglover]

I am trying to run a robust design with 3 primary sessions (with 6, 8, and 10 secondary sessions). Could anybody tell me why I would get a zero deviance on every model I run and/ or tell me where in the "Bible" I can find more info on this? I have not yet changed anything in the design matrix, but do suspect I have overdispersion. Could this be why?

Thanks,
Dave

[gwhite]

Dave:
The reason deviance = 0 has nothing to do with overdispersion in your data. Rather, you are running the robust design data type, where the closed capture models are part of the likelihood. For these models, I've not been able to compute a proper satured model likelihood, plus a negative constant that could be added to the likelihood is left out to speed up computation. Leaving out this constant often results in positive likelihood values, resulting in negative deviances (and even negative AICc values). The AIC values are okay, in that they still scale appropriately. However, the lack of an appropriate saturated model value plus negative deviance values means that the deviance is reported as zero. Its not a problem -- you can still use the -2log likelihood values out of the full output.

Gary

[dglover]

Gary,

Thanks for the fast reply!

Dave

[cooch]

<snip>.. and/ or tell me where in the "Bible" I can find more info on this?

Heh - not sure I'd refer to the book by that moniker (although the notion of a religious treatise typeset in a font called 'comic' is amusing perhaps) - but...no robust design chapter...yet.