On Wed, 5 Feb 2003, Akunda, Eric wrote:
Dear Prof King,
I have a question for you regarding inclusion of variables in the
imputation. If you have variables for which values are missing, and
think you will use their higher order (non-linear) values (e.g. squares,
cubes), should you impute the squares or just impute their linear values
and square these after imputation. In other words, does it make sense to
create uncertainty in them when you know that if you have a value of 3
for x, you should have 9 for x-squared and not 3 and 17 respectively?
you're right that it would be better to have a method that recognized this
constraint, but Amelia-for-windows doesn't easily do this. its designed
as a really good 2nd best method that applies almost universally. that
means that it is often possible to beat it in any specific application,
but often at great expense or effort. However, there is actually a
solution within Amelia-for-gauss by applying constraints. See
http://gking.harvard.edu/amelia/node21.html
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 (928) 832-7022 :