naomi (01/10/83)
#N:ubc-medgen:1300006:000:1667
ubc-medgen!naomi Jan 10 12:54:00 1983
S is great but... whenever I need something its not there! I know
you people out there in net.Sland are busily churning out S macros and
interface routines. How about letting us know what you've written.
(Even if it is only a few lines, you'll probably save me and some other
people hours of think and debug time.)
Here is what I have:
1. forward step-wise regression with nice output
(yes,leaps is better, but some people are attached to
stepwise regression, at least for forcing the ordered
entry of variables.)
Actually, this program is incomplete since it does only
FORWARDS stepping, and does not check tolerance, etc.
But...the algorithm does permit *SIMPLE* adaptation,
or so I understand from my source.
2. a rewrite of leaps and a print macro so that the output
looks like
varnames size Cp r2 adjr2
listed in order of subset size and selection criterion
3. a little routine to produce regression residuals, given y, x,
and beta.
Here is what I am about to write (probably today)
4. a little macro (or perhaps a function) to produce empirical
percentiles.
Note: once this BMDP interface is working, I will probably abandon all
this, but...I had some (more than enough) experience converting
BMDP to MTS (which is supposed to be IBM compatible) and so I
am not too optimistic about the time required to convert BMDP to
UNIX. On the other hand, the BMDP folks had so many complaints
from the MTS community that they have made a sincere effort to
improve their consistency and debugging.
Naomi Altman
!decvax!microsoft!uw-beaver!ubc-vision!ucb-medgen!naomi