[R-sig-ME] Model comparison with anova and AIC: p=0 and different AIC-values

[Ben Bolker]

On 13-11-16 01:12 PM, Stefan Th. Gries wrote:
> Darn, I made a typo:
> 
>> m.01a.reml <- lmer(OVERLAParcsine ~ USED*poly(SAMPLE, 2) + (USED*SAMPLE|NAME), REML=TRUE)
>> m.01b.reml <- lmer(OVERLAParcsine ~ USED*poly(SAMPLE, 2) + (USED+SAMPLE|NAME), REML=TRUE)
> 
> m.01a.reml is the more complex one: both have the same fixed-effects
> structure, but the a model has the more complex random-effects
> structure. (Reading rest now ...)
> 

  OK, sorry about the confusion.  It looks like something is wrong with
the optimization of m.01a.reml, since it should always have a higher
likelihood/lower negative log-likelihood than m.01b.reml.  Can you (1) try

  control=lmerControl(optCtrl=list(maxfun=1e5))

and (2) consider sending or posting data?

  Ben