[R] glht (multicomparisons) with a binomial response variable

Frank Harrell f.harrell at vanderbilt.edu
Mon Feb 6 13:55:25 CET 2012


Making sure you have the most recent version of rms, try
require(rms)
dd <- datadist(catroj, month); options(datadist='dd')
f <- lrm(catroj ~ month)
contrast(f, list(month=2:12), list(month=1), conf.type='simultaneous')

The actual coding will depend on how month is defined.  The contrast above
contrasts January with all other months, with simultaneous confidence
intervals provided by glht.
Frank

gaiarrido wrote
> 
> Hi,
> I,ve a run a model like this
> mcrm<-glm(catroj~month,binomial)
>       being catroj a binary response variable with two levels (infected
> and non infected) 
> 
>> anova(mcrm3,test="Chisq")
>      Df Deviance Resid. Df Resid. Dev P(>|Chi|)    
> NULL                   520     149.81              
> mes   3    16.86       517     132.94 0.0007551 ***
> 
> When I´m trying to do a post comparisons between factors i tried with glht
> of multcomp package by Torsten Hothorn I use "Tukey" contrast, but dídn't
> get diferencies
>  > summary(glht(mcrm3,linfct=mcp(month="Tukey")))
>                     Estimate Std. Error z value Pr(>|z|)  
> may - april == 0   -18.2470  1555.0564  -0.012   1.0000  
> june - april == 0   -1.9436     0.8307  -2.340   0.0671 .
> july - april == 0   -0.6253     0.5477  -1.142   0.6133  
> june - may == 0    16.3034  1555.0565   0.010   1.0000  
> july - may == 0    17.6216  1555.0563   0.011   1.0000  
> july - june == 0    1.3182     0.7899   1.669   0.2870  
> 
> I'm afraid "Tukey" it is not the correct contrast, but, What should i use
> for a binary response variable?
> Any advice, please?
> 


-----
Frank Harrell
Department of Biostatistics, Vanderbilt University
--
View this message in context: http://r.789695.n4.nabble.com/glht-multicomparisons-with-a-binomial-response-variable-tp4360898p4361171.html
Sent from the R help mailing list archive at Nabble.com.



More information about the R-help mailing list