[R] a question on statistics (rather than R-specific)

[Prof Brian D Ripley]

On Thu, 20 Jun 2002, Boryeu Mao wrote:

> I have used plor() to model a rather large 3-category dataset (~1500 data
> points, ~15 independent variables); from the resulting model (with a
> deviance slightly below the residual degrees of freedom), the training data
> are placed in only the two extreme categories.  Though the result appears to
> indicate that there's only a relative 'narrow' bin for the medium group,
> [and when the dataset is re-organized into 2 categories, glm(family =
> binomial(link = logit) ...) gives a model with a deviance reduced by about
> half from the 3-category polr() result], I am questioning if this (the
> 'narrow-bin') interpretation is too simplistic (or entirely incorrect), and
> wondering about the meaning of the absence of (fitted) data points in the
> medium group.
>
> Thanks in advance for any hints/pointers!

Probably the POLR model is inappropriate: try multinom for a fair
comparison (you cannot compare likelihoods on grouped and ungrouped data).

See the examples in MASS (the book) which is where polr() comes from.

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
r-help mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-help-request at stat.math.ethz.ch
_._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._._