The idea of the convex hull goes back to interpolation versus extrapolation.
From our observed data, we can calculate a convex hull. Then we can have
some counterfactuals, which may or may not be in the convex hull. If they
are not in the convex hull, then any prediction about the expected value of
Y will be heavily model dependent, in the sense that if we slightly tweak
the model, we may get a very different prediction of E(Y). If the
counterfactuals are in the convex hull, then our predictions of E(Y) will be
less model dependent in the sense that different models will probably give
similar predictions of E(Y) for our counterfactual(s).
On Thu, Apr 9, 2009 at 11:56 AM, Olena Ageyeva <eageeva at hotmail.com> wrote:
Still can not get clear the idea of convex hull. We
create it based on the
data we observed and if we ask model a question that is distant from the
observed data the answer either fall inside the convex hull or fall outside.
If it's outside that means that the model can not answer this question,
right? Or it means that the result is too far from a model to be certain
about it? That is somewhat not understandable how can we estimate how right
our predicted results are if we have just one model and nothing else. What
we compare the model with? From my understanding any result we get from the
same model should be in it's convex hull. Where am I wrong?
Sincerely,
Olena Ageyeva
------------------------------
Rediscover Hotmail?: Get quick friend updates right in your inbox. Check
it
out.<http://windowslive.com/RediscoverHotmail?ocid=TXT_TAGLM_WL_HM_Redis…
_______________________________________________
gov2001-l mailing list
gov2001-l at
lists.fas.harvard.edu
http://lists.fas.harvard.edu/mailman/listinfo/gov2001-l
--
Patrick Lam
Department of Government and Institute for Quantitative Social Science,
Harvard University
http://www.people.fas.harvard.edu/~plam