20 Dec
2002
20 Dec
'02
5:45 p.m.
On Fri, 20 Dec 2002, Kevin Timothy Arceneaux wrote:
Prof. King,
My understanding is that interactions, which will be used in the analysis,
should be included in the imputation data set. Say I have missing values
in X1 and X2, which produce missing values in X1*X2. Amelia will impute
values for X1, X2, and X1*X2, but it will not necessarily be the case that
the imputed values for X1*X2 will equal the multiplication of X1 and X2.
Is there a way to set this sort of constraint in Amelia? (I am using the
Windows version).
If not, would it be appropriate to set the missing values in X1*X2 to some
arbitrary number (e.g., -9) and then recomputed the interaction with the
imputed values for each of the resulting imputed data sets? Or should I
allow Amelia to impute missing values in X1*X2, and then correct these
values in the resulting imputed data sets?
Good question. I would alow Amelia to impute x1*x2 and then recompute
them from the contituent (imputed) variables. that allows the interaction
to have an effect and it produces consistent estimates.
One issue in Amelia (and all methods of imputation) is that if x1 is
missing for a variable but x2 is observed, x1*x2 will be coded as missing
even though it is partially observed. nothing you can do about this tho.
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 :
Thanks for your help,
Kevin.
Kevin Arceneaux
Ph.D. Candidate
Department of Political Science
Rice University
6100 Main St., MS 24
Houston, TX 77005
(713) 348-2107