Dear S+ users,
on Mon, 19 Apr 1999, I posted the question under this message to the
group. Unfortunately, I didn't get any responses, thus I am unable to
summarize any ideas or comments towards this topic.
If anyone has an idea or suggestions, I would still be very glad to hear
it and of course summarize them for the gruop.
Many thanks in advance.
All the best!
wolfgang
--------------
original question
-------------
Dear S+ users,
I try to classify two groups of patients with different outcomes after
psychotherapy - the successful vs. the not so successful. There are about
80% patients having good vs. 20% having an unsatisfying outcome in my
sample. Identifying those 20% doesn't really work out without applying
weights or costs to the smaller group.
The resulting models, after applying costs, seem very plausible and
usable. There is one problem though. Applying costs to the smaller group
results in many more patients with good outcomes misclassified and I don't
know in how far I can use the estimates from the v-fold crossvalidation,
since the error of course gets bigger.
The only way out I can think of, is to use another criterion to
estimate the goodness of the model (and following to decide on model
size), eg. the number of unsuccessful patients correctly classified. I'm
not sure, if this is the proper way, an not exactly correct, but usable
way or if I'm making a big mistake here.
I would be very grateful for comments or tips on other ways out.
I'm running S+ 3.3 under Win95 and use Thernau & Atkinson's Rpart library.
Thanks a lot in advance.
wolfgang
*********************************************
Wolfgang Hannoever
Forschungsstelle fuer Psychotherapie
Christian-Belser-Str. 79a
70597 Stuttgart
Tel.: 0711 / 6781-407 Fax:0711 / 6876902
e-mail:hann@psyres-stuttgart.de
*********************************************
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
|