Negative variance in ESEM for categor...
Message/Author
 Stéphanie Baggio posted on Friday, January 18, 2013 - 6:02 am
Dear Mrs. and Mr. Muthen,

I am performing an exploratory structural equation modeling with categorical variables, but unfortunately, 2 of my loadings are up to 1 or very close to 1.
I suppose I have negative residual variance, but I'm unable to fix the problem.
I know I have to switch to theta parameterization, test if the variance if significant, and fix it to 0 if not, but I don't know how to do.
Can you help me?

My model is:
MODEL:
F1-F3 BY V1-V10 (*1);
[F1@0 F2@0 F3@0];
{V1-V10@1};

Thank you
 Linda K. Muthen posted on Friday, January 18, 2013 - 2:03 pm
 Sylia Wilson posted on Tuesday, April 04, 2017 - 8:00 am
I am attempting to use ESEM to first establish fit of a model at two ages and in two groups, and to then test for measurement invariance. Variables are continuous, the sample is clustered in families.

I've used the below at both ages successfully.

Analysis:
Type = COMPLEX;
Rotation = TARGET(OBLIQUE);
Model:
F1 by
V1 V2~0 V3~0 V4 V5~0 V6~0 V7~0 V8~0 V9~0 V10~0 (*1);
F2 by
V1~0 V2 V3 V4~0 V5~0 V6~0 V7~0 V8~0 V9~0 V10~0 (*1);
F3 by
V1~0 V2~0 V3~0 V4~0 V5 V6 V7 V8~0 V9~0 V10~0 (*1);
F4 by
V1~0 V2~0 V3~0 V4~0 V5~0 V6~0 V7~0 V8 V9 V10 (*1);

However, when I try to examine the model in one of the groups, variable V3 has negative residual variance (nonsignificant). I tried constraining the residual variance for that variable:

! fix residual error variance to 0
V3 (vV3)
Model constraint:
vV3=0;

However, the residual variance for V3 is still negative (though smaller than in the original model).

I also tried constraining the residual variance for that variable to be greater than 0:

! fix residual error variance greater than 0
V3 (vV3)
Model constraint:
vV3>0;

But this model doesn't converge.

Do you have any suggestions for how to proceed from here?
 Tihomir Asparouhov posted on Thursday, April 06, 2017 - 9:16 am
Can you send the examples and the data to support@statmodel.com.

You can probably resolve the problem by updating your version of Mplus.

Also in the model you can use
V3@0;
V3 (vV3);
 Clayton McClintock posted on Wednesday, April 18, 2018 - 4:16 pm
I am trying to save factor scores for an ESEM using the code below. However, I received the message, "FACTOR SCORES COULD NOT BE COMPUTED. AT LEAST ONE OF THE RESIDUAL VARIANCES FOR THE CATEGORICAL VARIABLES IS NOT POSITIVE." The problem variable is Med_YN_Days.

USEVARIABLES ARE
Bel_Sali
Intr_Rel
Rel_Enga
Med_YN_Days
MindB_YN_Days
Spir_Nat
Eco_Awar
Universa
Psy_Love
Ont_Love
Soc_Lov
Compassi
Hum_Enga
;

CATEGORICAL = Med_YN_Days MindB_YN_Days;
MISSING ARE ALL (-99);

ANALYSIS:
ESTIMATOR IS WLSMV;
ROTATION IS QUARTIMIN;
ITERATIONS = 5000;
CONVERGENCE = 0.00005;

MODEL:
F1-F5 BY
Bel_Sali - Hum_Enga (*1);

SAVEDATA:

FILE = CategoricalFactorScores.sav;
SAVE = FSCORES;

When I tried to fix the variance of Med_YN_Days to zero or constrain the variance to be greater than 0, the model did not converge.

Is there any way to derive factor scores from this model?