[R-sig-ME] Two random effects with identical conditional modes

[boomslang]

Hi Jarrod,

Thanks for your reply and the link you provided.  I tried to get lme4a to work on several installations of R (2.13, 2.15, 3.0) to follow Douglas Bates' tips. Unfortunately I keep getting the error message:

Loading required package: minqa

Loading required package: Rcpp
Loading required package: MatrixModels
Error in inDL(x, as.logical(local), as.logical(now), ...) : 
  function 'cholmod_l_start' not provided by package 'Matrix'
In addition: Warning messages:
1: package 'minqa' was built under R version 2.13.2 
2: package 'Rcpp' was built under R version 2.13.2 
3: package 'MatrixModels' was built under R version 2.13.2 
Error: package/namespace load failed for 'lme4a'

I use  Windows XP (32bits) and 7 (64 bits)...


Kind regards






----- Original Message -----
From: Jarrod Hadfield <j.hadfield at ed.ac.uk>
To: boomslang <boomslang_fibs at yahoo.co.uk>
Cc: "r-sig-mixed-models at r-project.org" <r-sig-mixed-models at r-project.org>
Sent: Monday, 22 July 2013, 18:40
Subject: Re: [R-sig-ME] Two random effects with identical conditional modes

Hi,

This is called a multimembership model, and there is information on  
how to fit them in this thread:

http://comments.gmane.org/gmane.comp.lang.r.lme4.devel/6264

There are now better ways to specify this type of model in MCMCglmm,  
but the lmer suggestions by Doug may still be up to date.

Cheers,

Jarrod


Quoting boomslang <boomslang_fibs at yahoo.co.uk> on Mon, 22 Jul 2013  
00:33:17 +0100 (BST):

>
>
> Hello all,
>
> I have data with one categorial variable 'p'.  A binary outcome  
> ('result') is the result of a process determined by exactly two  
> DIFFERENT levels of p. 
>
> I am interested in the contribution of each level of p to the  
> outcome of 'result'.
>
> The code below produces a small but similar version of the dataset:
>
>     set.seed( 1)
>     p <- letters[1:6]
>
>     p1 <- sample( p, 30, replace = T)
>     p2 <- sample( p, 30, replace = T)
>     result <- runif( n = 30) > 0.7
>
>     x <- data.frame( p1, p2, result)
>     x <- subset( x, p1 != p2)
>     print( x, row.names = F)
>
>
> If I do
>
>    glmer( result ~ 1 + (1|p1) + (1|p2), x, binomial)
>    ranef( fm)
>
> ... I get different conditional modes for the levels in p1 and p2,  
> but this is not what I want since the condional mode of 'a' in p1  
> should be the same as the conditional mode of 'a' in p2.
>
> If I look at the random effects model matrix fm at Zt I see:
>
>
> 12 x 27 sparse Matrix of class "dgCMatrix"
>
>  [1,] . . . . . . . . . 1 . . . . . . . . . . . 1 . . 1 . . #p1-a
>  [2,] 1 . . . 1 . . . . . 1 1 . . . . . . . 1 . . 1 . . . . #p1-b
>  [3,] . 1 . . . . . . . . . . 1 . 1 . . 1 . . . . . 1 . 1 .
>  [4,] . . 1 . . . . 1 1 . . . . . . . . . . . 1 . . . . . .
>  [5,] . . . . . . . . . . . . . 1 . 1 . . . . . . . . . . .
>  [6,] . . . 1 . 1 1 . . . . . . . . . 1 . 1 . . . . . . . 1 #p1-f
>  [7,] . . . . . . . 1 . . . . . . . 1 . . . . . . 1 1 . . . #p2-a
>  [8,] . . . 1 . . . . . . . . . . . . . . . . . 1 . . 1 . . #p2-b
>  [9,] 1 . 1 . . . . . . 1 . . . . . . 1 . 1 . 1 . . . . . .
> [10,] . 1 . . . . . . . . . 1 1 1 . . . . . . . . . . . 1 1
> [11,] . . . . 1 1 1 . 1 . 1 . . . 1 . . 1 . . . . . . . . .
> [12,] . . . . . . . . . . . . . . . . . . . 1 . . . . . . . #p2-f
>
>
> ... while I need Zt to be more like this:
>
>  [1,] . . . . . . . 1 . 1 . . . . . 1 . . . . . 1 1 1 1 . . # a
>  [2,] 1 . . 1 1 . . . . . 1 1 . . . . . . . 1 . 1 1 . 1 . . # b
>  [3,] 1 1 1 . . . . . . 1 . . 1 . 1 . 1 1 1 . 1 . . 1 . 1 . # c
>  [4,] . 1 1 . . . . 1 1 . . 1 1 1 . . . . . . 1 . . . . 1 1 # d
>  [5,] . . . . 1 1 1 . 1 . 1 . . 1 1 1 . 1 . . . . . . . . . # e
>  [6,] . . . 1 . 1 1 . . . . . . . . . 1 . 1 1 . . . . . . 1 # f
>
>
> i.e. a 6x27 with exactly two 1's in each column.
>
> Is there a simple way/trick to obtain this?  Is the Pinheiro package  
> perhaps better suited for my problem?
>
>
> Any help is appreciated. Thanks in advance.
>
>
> PS: The dataset is very large: it has about 2 million rows and 'p'  
> has about 10000 levels. (This is the reason why I use random effects  
> for p. Preferably I would use fixed  
> effects.)                                                            
>                                                            
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>


-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.