[R] proportional odds model

[Frank E Harrell Jr]

Ramon Martínez Coscollà wrote:
> Hi all!!
> 
> I am using a proportinal odds model to study some ordered categorical
> data. I am trying to predict one ordered categorical variable taking
> into account only another categorical variable.
> 
> I am using polr from the R MASS library. It seems to work ok, but I'm
> still getting familiar and I don't know how to assess goodness of fit.
> I have this output, when using response ~ independent variable:
> 
> Residual Deviance: 327.0956
> AIC: 333.0956
>> polr.out$df.residual
> [1] 278
>> polr.out$edf
> [1] 3
> 
> When taking out every variable... (i.e., making formula: response ~ 1), I have:
> 
> Residual Deviance: 368.2387
> AIC: 372.2387
> 
> How can I test if the model fits well? How can I check that the
> independent variable effectively explains the model? Is there any
> test?
> 
> Moreover, sendig summary(polr.out) I get this error:
> 
> 
> Error in optim(start, fmin, gmin, method = "BFGS", hessian = Hess, ...) :
>         initial value in 'vmmin' is not finite
> 
> Something to do with the optimitation procedure... but, how can I fix
> it? Any help would be greatly appreciated.
> 
> Thanks.

You might also look at lrm and residuals.lrm in the Design package, 
which provides partial and score residual plots to check PO model 
assumptions.

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