At 02:00 PM 1/30/2009, sdm1 wrote:
Hi Maarten,
The model is trying to explain variations in annual hospital costs at the
individual patient level. We have various individual level variables such
as age, gender, and previous hospital use, but we also have a number of
variables that reflect the socio-economic characteristics of the area in
which the person lives. The addition of these attributed
need-for-health-care variables boosts the model's adjusted R-squared a
little but the boost is almost the same (at least to 4 decimal places)
whether all of them are included or just 12 are chosen on the basis of
researcher experience. Rather than use researcher experience to identify
the relevant variables, it would be good to have a slightly more formal
procedure.
Personally, I think using researcher experience might be a more
formal procedure! You are using a little theory or at least insight
with those 12. If those 12 vars perform as well as all 200 vars
combined, I'd say that would be a good case for the 12. So if it was
me, I would present the model with the 12, and then show that it does
as well as any model that is far less parsimonious.
It may be useful to see whether personal insights and theory produce
the same model as mindless atheoretical stepwise does. If stepwise
suggests a very plausible alternative that you had not considered,
you might pursue it further. But otherwise I would feel much better
with the 12 you picked yourself than a bunch of vars picked only
through pure empiricism.
-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
HOME: (574)289-5227
EMAIL: [email protected]
WWW: http://www.nd.edu/~rwilliam
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
