Negative residual variance with ESEM
Message/Author
 Tom Hess posted on Tuesday, June 20, 2017 - 10:42 am
I am attempting to perform MULTIPLE-GROUP ESEM WITH CONTINUOUS FACTOR INDICATORS FOLLOWING THE DIRECTIONS ON PAGE 99 OF THE MANUAL (SECTION 5.27). USING THAT SAME CODE, ALL WORKS WELL EXCEPT FOR A SMALL NEGATIVE RESIDUAL VARIANCE ON ONE INDICATOR FOR ONE OF THE THREE GROUPS. CAN I FIX THIS TO 0 JUST FOR THAT ONE GROUP IN ORDER TO COMPLETE THE MULTI-GROUP ANALYSIS? IF SO, HOW.

HERE IS MY CODE, WITH C3 THE PROBLEMATIC VARIABLE IN THE 3RD GROUP:

MODEL: F1 - F5 BY C1 E1 A1 O1 N1
A2 C2 E2 O2 N2 C3 E3 A3 N3 (*1);

[f1 - f5@0];

model G1:

O2 WITH O1;

Model G2:

F1 - F5 BY C1 E1 A1 O1 N1
A2 C2 E2 O2 N2 C3 E3 A3 N3 (*1);

[C1 - N3];

Model G3:

F1 - F5 BY C1 E1 A1 O1 N1
A2 C2 E2 O2 N2 C3 E3 A3 N3 (*1);

[C1 - n3];
 Bengt O. Muthen posted on Tuesday, June 20, 2017 - 5:23 pm
Try saying e.g.

a1@0;

within the group-specific Model statements.
 Tom Hess posted on Wednesday, June 21, 2017 - 6:11 am
I tried that and adding that single statement to the program results in it not running--the only output is a copy of the program. Without it, the program runs fine.

This where I included the code:

Model G3:

F1 - F5 BY C1 E1 A1 O1 N1
A2 C2 E2 O2 N2 C3 E3 A3 N3 (*1);

A3@0;

[C1 - n3];
 Linda K. Muthen posted on Wednesday, June 21, 2017 - 3:45 pm
Instead of this, try using MODEL CONSTRAINT to keep the parameter greater than zero. See Example 5.28.