I am a physician-researcher in Los Angeles. I recently sent an email to
Yang Yuan at the SAS institute regarding the new SAS procedures dealing with
imputation. He referred me to your group.
I have a technical question regarding the PROC MI
procedure that I'm wondering if you may be able to assist me with. I am
trying
to impute values for certain variables within a weighted (i.e. complex
sampling design with probability weights for each observation) database
entitled the National Accident Sampling System (NASS). Because PROC
MI cannot take a "weight" command, I have been trying to find a technique
to get around this problem. The following details the technique which I
plan to use to produce a valid database with imputed values through SAS.
Write code such that every non-integer probability weight (e.g. 0.10) would
be used as a probability of a given observation being included in the final
database. Following this, 10 databases would be created from the
original database, each with a slightly different make-up of observations
based on the probability of the cases with non-integer weights being
included. This technique would allow the sample size to remain constant,
with integer-based observations, without the need for weight adjustment.
For each of these 10 databases, we would run PROC MI (nimput=10), producing
an additional 10 databases for each one of the 10 datasets we had created.
The _imputation_ variable would be coded such that the final 100 databases
could be recognized by _imputation_ number 1-100. PROC MIANALYZE would then
be used to combine the analysis from each of the 100 datasets using
generalized estimating equations (PROC GENMOD).
Does this technique seem plausible and valid? Also, if the databases
differ slightly in total observation number, will PROC MIANALYZE have
difficulty combining results? Do you have any further suggestions?
I appreciate any assistance you may be able to provide. Thank you.
Craig
Craig D. Newgard, MD,MPH
Research Fellow
Department of Emergency Medicine
Harbor-UCLA Medical Center
1000 West Carson Street, Box 21
Torrance, CA 90509
(310)222-3666 (Office)
(310)782-1763 (Fax)
[EMAIL PROTECTED]
