[R] Missing p-values using lmer()

[Doran, Harold]

Yes, you're exactly right. You can use the mcmcsamp() function to sample from the posterior of an lmer object. This returns an object of mcmc class and you can do all of your diagnostics using the coda package.

-----Original Message-----
From: Spencer Graves [mailto:spencer.graves at pdf.com] 
Sent: Thursday, April 20, 2006 2:23 PM
To: Doran, Harold
Cc: Amelie LESCROEL; r-help at stat.math.ethz.ch
Subject: Re: [R] Missing p-values using lmer()

Hi, Harold:

	  Am I correct that the tool currently preferred for estimating p-values for lmer is "mcmcsamp"?

Amélie:  My favorite tool for exploring the archives is 'RSiteSearch'. 
You can also get to it via www.r-project.org, but the last time I tried to copy a web address to paste in an email, the address I got from 'RSiteSearch' worked but the one I got from the web page directly did not.

	  hope this helps,
	  spencer graves

Doran, Harold wrote:

> You didn't do anything wrong, lmer doesn't give them. 
And, for good reason. I've been a bit indoctrinated by D. Bates, so let me share what I've learned.
> 
> With simple analysis of variance models with simple
error structures, it is known that the ratio of the variances follow and F distribution. However, with more complex error structures, the null distribution is unknown. Most other multilevel programs accept by analogy that the ratio of the variances do follow an F distribution. That is, it works well for the simple case, therefore it probably is also true for the more complex case.
> 
> In SAS, one can choose ddf options, such as Kenward-
Roger, which hopes that after assuming the ratio of variances follow an F distribution, the only remaining challenge is to properly estimate the denominator degrees of freedom. These kinds of options do not currently exist in lmer and after many discussions on this list Doug Bates decided to remove the p-values for now.
> 
> This topic has been discussed often on this list and
you can see other discussions on the archive which may be more insightful.
> 
> Harold
> 
> 
> -----Original Message-----
> From: r-help-bounces at stat.math.ethz.ch 
> [mailto:r-help-bounces at stat.math.ethz.ch] On Behalf Of Amelie LESCROEL
> Sent: Thursday, April 20, 2006 1:41 PM
> To: r-help at stat.math.ethz.ch
> Subject: [R] Missing p-values using lmer()
> 
> Hello,
> 
>  
> 
> I'm trying to perform a REML analysis using the lmer() function (lme4 package). Well, it seems to work well, except that I'm not getting any p-value (see example below). Can someone tell me what I did wrong?
> 
>  
> 
> Thanks for your help,
> 
>  
> 
> Amélie
> 
>  
> 
> 
>>library(gdata)
> 
> 
>>dive <- read.xls("C:/Documents and Settings/Amelie/My 
>>Documents/Postdoc/CE
> 
> 2005-2006/divebydive.xls", perl="C:/perl/bin/perl.exe")
> 
> 
>>library(lme4)
> 
> 
> Loading required package: Matrix
> 
> Loading required package: lattice
> 
> 
>>reml.res <- lmer(UNDS~SUCCESSMN+(1|BIRD), dive)
> 
> 
>>summary(reml.res)
> 
> 
> Linear mixed-effects model fit by REML
> 
> Formula: UNDS ~ SUCCESSMN + (1 | BIRD)
> 
>           Data: dive
> 
>       AIC     BIC    logLik MLdeviance REMLdeviance
> 
>  60032.37 60053.8 -30013.19    60031.9     60026.37
> 
> Random effects:
> 
>  Groups   Name        Variance Std.Dev.
> 
>  BIRD     (Intercept)  4.4504  2.1096  
> 
>  Residual             36.4240  6.0352  
> 
> number of obs: 9324, groups: BIRD, 12
> 
>  
> 
> Fixed effects:
> 
>             Estimate Std. Error t value
> 
> (Intercept) 13.39764    0.63887 20.9707
> 
> SUCCESSMN    4.22197    4.11527  1.0259
> 
>  
> 
> Correlation of Fixed Effects:
> 
>           (Intr)
> 
> SUCCESSMN -0.276
> 
> 
>>anova(reml.res)
> 
> 
> Analysis of Variance Table
> 
>           Df Sum Sq Mean Sq
> 
> SUCCESSMN  1 38.337  38.337
> 
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html