On Sat, 28 Apr 2001, Arsad Tanjung wrote:
> When I fit a model using glm (e.g. low birth weight dataset, V&R pp
> 234-235), I found the results when stating this option:
>
> options(contrasts=c("contr.treatment", "contr.poly"))
>
> exactly the same as in the book:
>
> Coefficients:
> Value Std. Error t value
> (Intercept) 0.82301258 1.244073209 0.6615467
> age -0.03723398 0.038677748 -0.9626718
> lwt -0.01565297 0.007075944 -2.2121383
> raceblack 1.19240920 0.535745758 2.2256998
> raceother 0.74068107 0.461460949 1.6050785
> smoke 0.75552501 0.424764466 1.7786916
> ptd 1.34376082 0.480444518 2.7969115
> ht 1.91316170 0.720434406 2.6555668
> ui 0.68019467 0.464215580 1.4652560
> ftv1 -0.43637856 0.479161054 -0.9107138
> ftv2+ 0.17900705 0.456209017 0.3923795
>
> BUT... when I do not state this option, the results are:
>
> Coefficients:
> Value Std. Error t value
> (Intercept) 3.72790660 1.344892247 2.7718998
> age -0.03723398 0.038677748 -0.9626718
> lwt -0.01565297 0.007075944 -2.2121383
> race1 0.59620460 0.267872879 2.2256998
> race2 0.04815882 0.144406014 0.3334960
> smoke 0.37776250 0.212382233 1.7786916
> ptd 0.67188041 0.240222259 2.7969115
> ht 0.95658085 0.360217203 2.6555668
> ui 0.34009734 0.232107790 1.4652560
> ftv1 -0.21818928 0.239580527 -0.9107138
> ftv2 0.13239878 0.148931932 0.8889885
>
> QUestion: Why do we need to state the option in order to get the correct
> result?
Both results are correct! They mean different things, as the book
*does* explain.
And see page 233, BTW. You can't jump into the middle of an example
and expect to reproduce the results.
--
Brian D. Ripley, ripley@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
|