[R-sig-ME] gamm4 model fit error

[Ben Bolker]

On 14-07-13 01:37 PM, Nick Lange wrote:
> Thanks, Ben,
> 
> Yes, a warning not and error, thanks. And also, yes, this is the
> first time I've used gamm4.
> 
> I tried coef(summary(fit$mer))
> 
> and it gave an improved regression summary table, but without
> p-values:
> 
> Estimate Std. Error     t value X(Intercept)    1310.68401   9.204793
> 142.3914670 Xs(AgeYears)Fx1   19.89873  25.072824   0.7936372
> 
> which I can't calculate without the df (but it's large).

  I'm not surprised that you don't get p-values: there is even an R FAQ
on this subject

http://cran.r-project.org/doc/FAQ/R-FAQ.html#Why-are-p_002dvalues-not-displayed-when-using-lmer_0028_0029_003f

(which simply refers to

https://stat.ethz.ch/pipermail/r-help/2006-May/094765.html

 -- you may also want to look at http://glmm.wikidot.com/faq)

> 
> My colleague has the same versions as I do, except she's running lme4
> 1.1-6 not lme4 1.1-7. Everything works fine for her.

  I'm not surprised she doesn't get the warning messages (those were
introduced in 1.1-6), but I *am* surprised if she gets p-values.  Since
you have large degrees of freedom, you could just compute two-tailed
p-values as:

cc  <- coef(summary(fit$mer))
2*pnorm(abs(cc[,"t value"]),lower.tail=FALSE)

or

2*pnorm(-abs(cc["t value"]))



> 
> I've been using R for decades (was at MIT when Ross Ihaka has writing
> it) but have never joined the sigs. How do I post this reply on the
> R-sig-ME thread?
> 
> Nick
>