Skip to content

negative AIC and BIC values in gls

7 messages · Gary Dong, Jeremy Miles, David Winsemius +2 more

#
It's fine. Just interpret them as you would any other (lower is better).
On 22 August 2012 16:43, Gary Dong <pdxgary163 at gmail.com> wrote:
#
On Aug 22, 2012, at 5:17 PM, Jeremy Miles wrote:

            
And it is the printed logLik that is out of step here. log-likelihoods  
_should_ be negative.
#
Inline.
On Wed, Aug 22, 2012 at 5:17 PM, Jeremy Miles <jeremy.miles at gmail.com> wrote:
I don't think so. I believe all 3 values are the negative of what they
should be. AIC is defined as -2*log(L) + k*{degrees of freedom for
model) . BIC is similar (different k's). This should be positive and,
as you said, the smaller the better. So if the signs are reversed,
models with lower absolute AIC and BIC -- bigger negative AIC and BIC
-- is what you want.

However, this really needs comment by an expert, which I ain't. I also
suggest you post to r-sig-mixed-models, since gls is an nlme function.

Cheers,
Bert

  
    
#
O course! And for the same reason, my stupid comment should be ignored.

-- Bert
On Thu, Aug 23, 2012 at 12:02 AM, Ingmar Visser <i.visser at uva.nl> wrote: