[R-sig-ME] lmer, no residual in the output - want REML not laplace
bbolker at gmail.com
Thu Jul 19 23:30:54 CEST 2012
Yolande Tra <yolande.tra at ...> writes:
> I run the following code.
> 1. REML fit was not tin the output
As a recent post on this list stated, the definition of REML
is somewhat unclear for GLMMs. (Dave Fournier has given a reasonable
definition in the past, but not all researchers in this area agree
with his definition.) If you can specify exactly what you want
the code to do in order to implement restricted ML for a GLMM,
that might spark some discussion.
> 2. There was no residual in the output
How about residuals(g) ?
> 3. I could not run anova(g)
> > (g=lmer(total_count ~ c+(1|c:b:a), d2, REML=TRUE, family = "poisson"))
> Generalized linear mixed model fit by the Laplace approximation
> Formula: total_count ~ c + (1 | c:b:a)
> Data: d2
> AIC BIC logLik deviance
> 661.9 673.1 -326.9 653.9
> Random effects:
> Groups Name Variance Std.Dev.
> c:b:a (Intercept) 1.8368 1.3553
> Number of obs: 122, groups: c:b:a, 18
> Fixed effects:
> Estimate Std. Error z value Pr(>|z|)
> (Intercept) 1.6383 0.5605 2.923 0.00347 **
> cLovers Point 0.2080 0.7924 0.262 0.79295
> cPoint Pinos -0.4282 0.7998 -0.535 0.59242
> Signif. codes: 0 *** 0.001 ** 0.01 * 0.05 . 0.1 1
> Correlation of Fixed Effects:
> (Intr) cLvrsP
> cLoversPont -0.707
> cPointPinos -0.701 0.496
> > anova(g)
> Error in anova(g) : single argument anova for GLMMs not yet implemented
> What might be wrong?
Maybe it's not implemented? What do you want it to do?
Perhaps try fitting a reduced model (g2 <- update(g2,.~.-c)
and go from there? Or try drop1() ?
More information about the R-sig-mixed-models