[R] using rcorr.cens for Goodman Kruskal gamma

[Frank E Harrell Jr]

Colin Robertson wrote:
> Dear List,
> 
>  
> 
> I would like to calculate the Goodman-Kruskal gamma for the predicted
> classes obtained from an ordinal regression model using lrm in the Design
> package. I couldn't find a way to get gamma for predicted values in Design
> so have found previous positings suggesting to use :
> 
>  
> 
> Rcorr.cens(x, S outx = TRUE)  in the Hmisc package
> 
>  
> 
> My question is, will this work for predicted vs observed factors?  I.e. x =
> predicted class and S = observed class? Or is there a better way to obtain
> this? I used the maximum individual probability for each observation to
> determine the predicted class. 

Rank correlation measures are for correlating a continuous or ordinal 
prediction with a response (continuous, ordinal, or binary).  So you 
should be able to do something like rcorr.cens(predict(fit), 
as.numeric(Y), outx=TRUE).  Note that rcorr is all lower case.  This 
assumes that the levels of Y are in order, as does lrm.

Note that the new version of lrm has a method for getting predicted mean 
scores from an ordinal lrm.

Frank

> 
>  
> 
> Any help appreciated, 
> 
>  
> 
> Thanks
> 
>  
> 
> Colin
> 
>  
> 
>  
> 
>  
> 
> Colin Robertson
> 
> Dept of Geography
> 
> University of Victoria


-- 
Frank E Harrell Jr   Professor and Chair           School of Medicine
                      Department of Biostatistics   Vanderbilt University