[R] polr (MASS): score test for proportional odds model

[Frank E Harrell Jr]

Michael Friendly wrote:
> For the following model,
> 
> library(vcd)
> arth.polr <- polr(Improved ~ Sex + Treatment + Age, data=Arthritis)
> summary(arth.polr)
> 
> where Improved is an ordered, 3-level response I'm looking for a 
> *simple* way to test
> the validity of the proportional odds assumption, typically done via a 
> score test
> for equal slopes/effects over the predictors.

That test is anti-conservative.  I like graphical methods using partial 
residual plots as in my book.  See the following reference.  -Frank

@ARTICLE{pet90par,
   author = {Peterson, Bercedis and Harrell, Frank E.},
   year = 1990,
   title = {Partial proportional odds models for ordinal response 
variables},
   journal = Applied Statistics,
   volume = 39,
   pages = {205-217}
}

> 
> I do find a po.test= option in the repolr package for repeated measures, 
> but for my
> (tutorial) purposes this is too complex.  Did I miss something simpler?
> 


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