topic Warning: within-imputation StdErr is zero for Variable Risk in Statistical Procedures

Warning: within-imputation StdErr is zero for Variable Risk

yw2757 — Tue, 11 May 2021 19:04:01 GMT

Multiple imputation: using proc freq procedure to summarize an imputation dataset (number of imputation=5). The statistics need to find are "proportion of responders" and "95%CI of proportion".

PROC FREQ DATA= Impdsin;
By NIMPUTE;
Table TRTN*depvar / RiskDiff(Equal CL=(Wald));
Ods Output RiskDiffCol2 = Imp_propci01 (Where=(Index(Row, 'Row')));
Ods Output RiskDiffCol2 = Imp_diffci01 (Where=(Index(Row, 'Difference')));
RUN;

PROC MIANALYZE DATA = Imp_propci01;
By TRTN;
Modeleffects Risk ;
Stderr ASE ;
ODS Output
ParameterEstimates= Imp_propci_Miana;
RUN;

Warning Appear because

Anyway to solve this issue?

Thank you!

Re: Warning: within-imputation StdErr is zero for Variable Risk

ballardw — Tue, 11 May 2021 21:27:41 GMT

When there is no variability for a variable within a group the SdtErr is 0. Can't be "fixed" as that comes from you data.

Example:

data example;
  input x y;
datalines;
1   23
1   22
1   24
2   18
2   18
2   18
2   18
;

proc means data=example min max mean stderr lclm uclm;
  by x;
  var y;
run;

Note that for X=2 the STDERR is 0 because all of the values of Y are the same for that value of X.

Likely that means for one or more values of your variable TRTN your dependent variable has the same values, i.e. all of them imputed(possibly) to the same value, or before your imputation no values were missing, all the same and none imputed.