Proc mianalyze WARNING "Between-imputation variance is zero for variable"

[Clemence]

Hello,

I use the PROC MIANALYZE to obtain summary statistics but I have a warning message "Between-imputation variance is zero for variable". So, I don't have confidence interval. I would like to know if the standard error of the mean was correct or not ? 

Here my code :

proc univariate data=mimpute noprint;
var LB_BIFIDO_OLDnumconvert;
output out=mimpute_uni mean=LB_BIFIDO_OLDnumconvert
stderr=SLB_BIFIDO_OLDnumconvert;
by _Imputation_ ;
run;

proc mianalyze data=mimpute_uni ;
modeleffects LB_BIFIDO_OLDnumconvert;
stderr SLB_BIFIDO_OLDnumconvert;
run;

And I obtain : 

Parameter Estimate Std Error 95% Confidence Limits DF Minimum Maximum Theta0 t for H0 Pr > |t|

8906410753

1049291539

.

.

.

8906410552

8906410954

0

.

.

Thanks for your anwser                                                                                                                                                                              

Clémence                                                                                                                                                                                                    

[SAS_Rob]

When the between imputation variance is zero then the number of Degrees of Freedom is undefined so you cannot get a confidence interval or p-values. 
Unfortunately, there is not a good approach to take in this case. This is one of the limitations of multiple imputation in general. I have not seen any suggestions in the multiple imputation literature about this issue. 

The standard error that is reported is correct however.

[Clemence]

Ok, thank you for your answer. 

Clémence 

[kan007]

Hi Clemence,

did you resolve your issue? If so, Please share how did you handle this, I am having the same issue when I execute below program.

proc mianalyze data=abc;
modeleffects logrr;
stderr selogrr;
run;

WARNING: Between-imputation variance is zero for variable logrr.

[SAS_Rob]

As I mentioned above there isn't necessarily a fix for this since all of the estimates across the imputed data sets are indentical.  That being said, it can sometimes be an indication that the imputation model itself is wrong or omitted a key variable.  I would go back and make sure that any related variables in the analyst's model is included in the imputation model.

[kan007]

I get your point. thanks Rob for quick reply.