R help - May 2008 - rpart for survival fits

Hello Gurus:
 
When I plot a survival fit using rpart for the classification tree, for each
node, there is a decimal based number above the event/total. I tried to see if
it's the exponential ratio or logrithmics, neither seem to be the case.
I'm wondering if anyone knows what they are.
 
Thanks,
Karen
_________________________________________________________________
Find hidden words, unscramble celebrity names, or try the ultimate
crosswor[[elided Hotmail spam]]

	[[alternative HTML version deleted]]

On Mon, 5 May 2008, Chang Liu wrote:
>
> Hello Gurus:
>
> When I plot a survival fit using rpart for the classification tree,
I don't think that is a 'classification tree': no classification is
going
on.  I will assume you called rpart() with a survival response.
> for each node, there is a decimal based number above the event/total. I 
> tried to see if it's the exponential ratio or logrithmics, neither seem
> to be the case. I'm wondering if anyone knows what they are.
Yes, I am sure the author knows.  But to be precise, plotting the fit does 
not output those numbers, the text() method does.  What it adds is the 
frame$yval component of the fit, the 'fitted value' according to 
?rpart.object (which is printed when you print the fit).

You really do need to read the documentation to understand what the fitted 
values are in this case.  E.g. http://www.mayo.edu/hsr/techrpt/61.pdf
> Thanks,
> Karen
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>
-- 
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 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

> When I plot a survival fit using rpart for the classification tree, for
each
> node, there is a decimal based number above the event/total. I tried to see
if
> it's the exponential ratio or logrithmics, neither seem to be the case.
I'm
> wondering if anyone knows what they are.
---------
 
  It is an estimate of the event rate, using a Baysian shrinkage argument
  	(#events +a) / (n +b)  *c
 
To understand how the constants a and b are chosen, you need to read the 
detailed documentation (as pointed out already by Brian R).  The constant c is 
chosen to make the printed "rate" of the top node equal to 1.  This
last is done
just to make it easier to scan the tree -- one can easily see that the rate is, 
say, 20% lower in some particular node than for the data set as a whole.


	Terry Therneau