that's right, you could get different results when you assume different
models, which you're doing with the 2 examples you propose. In one you're
assuming that the (>= 5) parameters of the (truncated bivariate normal)
model are constant across congressional districts and in the other you're
assuming that they vary. If you have sufficient observations within each
CD (and if you have precincts you probably do in most states), then you
could run them separately since you'll be able to estimate those in each
CD without much ncertainty, conditional on the model. you would still
have uncertainty due to the model of course, but you would find that in
some CDs you'd get fairly certain estimates due to narrow bounds, which
might be informative. in others, you might not have much info at all
aside from the model-dependent inferences. running them separately would
give you the chance to sort out the inferences in these and other
categories. on the other hand, if you don't have many observations within
each CD, then you will need some kind of pooling or you'd be sacrificing
too much variance to reduce some bias. i.e., its the same issue as with
any pooling problem.
Gary
: Gary King, King(a)Harvard.Edu
http://GKing.Harvard.Edu :
: Center for Basic Research Direct (617) 495-2027 :
: in the Social Sciences Assistant (617) 495-9271 :
: 34 Kirkland Street, Rm. 2 HU-MIT DC (617) 495-4734 :
: Harvard U, Cambridge, MA 02138 eFax (617) 812-8581 :
On Tue, 18 May 2004, Gregory Pettis wrote:
Imagine I am sitting down to have EI estimate turnout for whites and
blacks in two states. My hypothesis is that black candidates increase
white and black turnout in the Congressional districts in which they run.
I have precinct-level turnout for each precinct in the states, as well as
precinct-level racial composition. I have a choice as to two ways to
estimate the turnout rates. I can estimate
them Congressional district by Congressional district, or I can estimate
all of the turnout rates for whites and blacks for the entire state at one
time.
If I estimate these rates all at once for the state, then I am borrowing
the most strength from all of the turnout in all of the state. The
downside, however, might be that if I estimate turnout for each district
separately then I am allowing those districts to have turnout rates that
are more different from each other than would be the case if I estimated
the white and black turnout rates for each state at the same time. In
other words, we might expect EI to produce different answers depending on
how we run the models in the way I have described.
My question is, which way is preferable? And my second question, how
different would we expect the turnout rates to be depending upon how I run
EI?
Thanks,
Greg
Gregory A. Pettis
Political Science
Elon University
CB # 2203
Elon, N.C. 27244
-
EI mailing list served by Harvard-MIT Data Center
Subscribe/Unsubscribe:
http://lists.hmdc.harvard.edu/?info=ei
-
EI mailing list served by Harvard-MIT Data Center
Subscribe/Unsubscribe:
http://lists.hmdc.harvard.edu/?info=ei