Skip to content

R-sig-mixed-models Digest, Vol 32, Issue 24

2 messages · David Evans, David Duffy

Original
David Evans
# 
Thanks for the help so far.  I'll look at mcmcglmm--a good idea, even  
though I was hoping to use R for my "quick" exploratory models before  
moving to WinBUGS.  "Quickly exploring" with mcmc makes me nervous.

David, would it be possible to help me with some references on using  
Gaussian errors for binary data which you mention below.  I've read  
the Paper by Cheung, A modified least-squares regression approach to  
the estimation of risk difference, AJE 2007, 166:1337-1344 but was  
hoping to find some more on this.  In particular, any references which  
say Gaussian errors are acceptable in mixed models would be most  
appreciated (I have an outcome prevalence in the 0.2's).

Thanks again.

David.

P.S. Judging by the responses I've received, I can see Australia is  
not on holidays at the moment, unlike the Northern Hemisphere!

Le 20 ao?t 09 ? 12:00, r-sig-mixed-models-request at r-project.org a  
?crit :
2 days later
David Duffy
#
On Sun, 23 Aug 2009, David Evans wrote:

            

      
      
      
    

        
A couple of reviews defending the general idea ;):

Harvey WR (1982). Least-squares analysis of discrete data. J Anim Sci 
54:1067-1071

http://jas.fass.org/cgi/reprint/54/5/1067

And also

Ch 17 in Gianola and Hammond Advances in Statistical Methods for
Genetic Improvement of Livestock.

The latter suggests:

V. Guiard, G. Herrend ??rfer, A. Tuchscherer (1985).  Variance Component
Estimation for Dichotomous Characters and Its Use for Estimating
Heritability. Biometric J 27: 653-658.

The heritability is the proportion of variance due to genetic random
effects.

A couple of genetics example I am aware of, where simulation finds that
the linear model gives similar answers to a GLMM or similar:

Visscher PM, Haley CS, Knott SA (1996). Mapping QTLs for binary traits
in backcross and F2 populations. Genetical Research 68(1):55-63.

http://genepi.qimr.edu.au/contents/p/staff/CVPV020.pdf

Zeegers MP, Rice JP, Rijsdijk FV, Abecasis GR and Sham PC (2003).
Regression-based sib pair linkage analysis for binary traits.
Hum Hered 55:125-31


There are many other applications of the approach in the animal 
breeding literature for heritability estimation.

But I'm afraid these merely support the approach in general - they
haven't usually applied a standard program like lme()/lmer(), but I
think one can argue that the same analysis of variance machinery is
being used.  The literature on binary intraclass/interclass correlations
in cluster sampling etc (Landis and Koch onwards) is also relevant.

Maybe also:

Lunney GH (2005).  Using analysis of variance with a dichotomous
dependent variable: an empirical study. J Educ Meas 7:263-269.

Cheers, David Duffy.