[Impute] proc MI syntax

[newgardc]

To the SAS PROC MI/MIANALYZE experts,
 
I am struggling with the correct syntax for PROC MIANALYZE when using a 
logistic regression model with multiply imputed data.  The model actually uses 
PROC SURVEYLOGISTIC, but I believe the MIANALYZE syntax is the same as for PROC 
LOGISTIC.  I am unable to create a "covb" dataset from the model output, but 
the standard errors are output to the same dataset as the parameter estimates.  
When I try to combine results using MIANALYZE,I get an error message if "mult" 
is included as an option (needed for combining multivariate inferences).  If 
the parameter estimates are combined without including code for the standard 
errors (STDERR), I'm concerned that the resulting final standard errors are too 
small.  See below.  Any input is greatly appreciated.  Thanks.  Craig-
 
proc surveylogistic data=f.final_sas;
cluster psu;
model y = x1 x2 x3 x4 x5;
strata psustrat;
weight ratwgt;
by _imputation_;
ods output Parameterestimates=gmparms; 
run;
 
proc mianalyze parms=gmparms mult;
modeleffects x1 x2 x3 x4 x5;
run; 


-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
http://lists.utsouthwestern.edu/pipermail/impute/attachments/20060701/3e0aad2e/attachment.htm
From allison <@t> soc.upenn.edu  Sat Jul  1 07:04:48 2006
From: allison <@t> soc.upenn.edu (Paul Allison)
Date: Sat Jul  1 07:04:55 2006
Subject: [Impute] proc MI syntax
In-Reply-To: <s4a5f025....@ohsu.edu>
Message-ID: <200607011204.k61c4nbc027...@facstaff.sas.upenn.edu>

Craig:

With your current code, the standard errors are correct but you can't do
multivariate inference (i.e., use the MULT option or TEST statements).  You
need to request and output the covariance matrix for the parameter
estimates, then read it into mianalyze.  

This should do it:

proc surveylogistic data=f.final_sas;
cluster psu;
model y = x1 x2 x3 x4 x5 / COVB;
strata psustrat;
weight ratwgt;
by _imputation_;
ods output Parameterestimates=gmparms COVB=COVMAT; 
run;
 
proc mianalyze parms=gmparms COVB=COVMAT mult;
modeleffects x1 x2 x3 x4 x5;
run;


-----------------------------------------------------------------
Paul D. Allison, Professor and Chair
Department of Sociology
University of Pennsylvania
3718 Locust Walk
Philadelphia, PA  19104-6299
215-898-6712, 215-898-6717
215-573-2081 (fax)
http://www.ssc.upenn.edu/~allison

________________________________________
From: impute-boun...@lists.utsouthwestern.edu
[mailto:impute-boun...@lists.utsouthwestern.edu] On Behalf Of Craig Newgard
Sent: Saturday, July 01, 2006 6:47 AM
To: imp...@lists.utsouthwestern.edu
Subject: [Impute] proc MI syntax

To the SAS PROC MI/MIANALYZE experts,
?
I am struggling with the correct syntax for PROC MIANALYZE when using a
logistic regression model with multiply imputed data.? The model actually
uses PROC SURVEYLOGISTIC, but I believe the MIANALYZE syntax is the same as
for PROC LOGISTIC.? I am unable to create a "covb" dataset from the model
output, but the standard errors are output to the same dataset as the
parameter estimates.? When I try to combine results using MIANALYZE,I get an
error message if "mult" is included as an option (needed for combining
multivariate inferences).? If the parameter estimates are combined without
including code for the standard errors (STDERR), I'm concerned that the
resulting final standard errors are too small.? See below.? Any input is
greatly appreciated.? Thanks.? Craig-
?
proc surveylogistic data=f.final_sas;
cluster psu;
model y = x1 x2 x3 x4 x5;
strata psustrat;
weight ratwgt;
by _imputation_;
ods output Parameterestimates=gmparms; 
run;
?
proc mianalyze parms=gmparms mult;
modeleffects x1 x2 x3 x4 x5;
run;