proc mixed with imputated (multiple) data

[mahmood]

Hello,

I have got a problem and wondering if someone can help me please. I am trying Proc Mixed with the random intercept: and getting a problem. I have done multiple imputation (so, there are 10, for example _imputation_ = 1 ...._imputation_ = 10). The proc mixed is as follows (variable names are just for example):

proc sort data = reg_fin_1 ;

by _imputation_ ID ;

run;

proc mixed data = reg_fin_1 noitprint PLOTS(MAXPOINTS= 20000000000) ;

class ID ;

by _imputation_ ;

model Y = x1 x2 x3

/solution ddfm = bw residual ;

random intercept / subject = NEW_PIN type = un ;

ods output solutionF=mxparms3 ;

run;

The problem is that I get the following message:

Estimated G matrix is not positive definite.

The above message was for the following BY group: _imputation_=1

Estimated G matrix is not positive definite.

The above message was for the following BY group: _imputation_=2

......

.......

However, I don't get this notice when I run proc mixed with the complete cases: I mean, excluded the missing values, hence no imputation, and unlike the multiple imputation I have one dataset.

Thanks to all.

Rahid

[mahmood]

Hello,

Thanks to everyone. This "error" has been solved: the problem was due to the organization the imputed data properly. Howeve, I have another question: First of all, this is a longitudinal/panel data. I have a random intercept model and the G matrix I am getting shows (only) row one for the first ID:

 Estimated G Matrix

                            Row    Effect       NEW_PIN         Col1

                              1    Intercept       10001      0.1448