19 Feb
2003
19 Feb
'03
9:56 p.m.
yes, that's what I'd do. As you can see, Amelia is not fully (logically)
consistent in the presence of interactions, but the approach you're taking
should work fine in most situations.
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 :
On Wed, 19 Feb 2003, Matthew Vile wrote:
Suppose the following analysis model:
y = a + b1x1 + b2x2 + b3x3 +b4x1x2 + b5x1x3 + b6x1x2x3 + u
Suppose x1 is a multi item measure (factor scale) consisting of z1 + z2
+ z3 + z4
Suppose that all variables contain some degree of MAR data.
Based upon previous listserve responses I suppose the proper imputation
model would contain the following:
y, z1,z2,z3,z4,x1(built from z
variables),x2,x3,(x1*x2),(x1*x3),(x1*x2*x3) + any related imputation
model variables (which probably should be included in the analysis model
as exogenous terms, but that's another story).
Imputation would fill all missing values in all these variables.
However, to regain the information lost in the constructed variables
(i.e. x1 and interactions) when they are created with variable
containing some (but not all) missing data, (after imputation) you
should delete and rebuild x1 and the interaction variables.
Does this sound correct?
Matthew Vile,
University of New Orleans