Measurement Invariance Analyses PreviousNext
Mplus Discussion > Confirmatory Factor Analysis >
Message/Author
 Robert Latzman posted on Saturday, March 02, 2013 - 6:22 pm
Hello,

I am working on some measurement invariance analyses based on both gender and race and have been running into some trouble. Although the two most restrictive models are running and fitting fine, the two less restrictive models are not. I am receiving the following error message:


THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED. CHECK YOUR MODEL.PROBLEM INVOLVING PARAMETER 201.

THE CONDITION NUMBER IS -0.577D-09.

There seems to be something wrong with the latent covariance matrix...it seems the matrix is singular (as indicated by the small condition number) leading to two models not being identified. When looking at the parameter estimates, there does not seem to be anything glaringly wrong. Interestingly, the same problem occurs in the PSI matrix when running separate multi-group analyses based on gender and based on race. I have tried fixing the parameter identified as being the problem to 0 but this simply results in the program identifying a different parameter with the same problem in the next run.


Thank you very much for your help with this.
 Linda K. Muthen posted on Saturday, March 02, 2013 - 6:38 pm
I would guess that you are freeing the first factor loading by mentioning it in the group-specific MODEL commands. You should not do that, for example,

MODEL:
f BY y1 y2 y3;

MODEL male:
f BY y2 y3;

If this is not the case, please send the output and your license number to support@statmodel.com.
 Orpha de Lenne  posted on Thursday, September 14, 2017 - 7:35 am
Dear Dr. Muthen,

I am testing measurement invariance by using the "MODEL = CONFIGURAL METRIC SCALAR" command.

The tests were significant. Therefore I would like to aim for partial measurement invariance by finding the source of variance.

1. Is it possible to ask for modification indices using the "MODEL = CONFIGURAL METRIC SCALAR" command?
2. How do I add freely estimated parameters in the syntax?

Thank you in advance.

Below you can find my syntax:

TITLE: Performance Pressure Prof Soc Sex Rom Measurement Invariance

DATA: FILE IS Dataset Clean Mplus.dat;

VARIABLE: NAMES =
Country BirthD BirthM BirthY Gender DegreeF DegreeM BirthC Ethn NativeL Sexual LivingS
CWF CWF1-CWF4 CWP CWP1-CWP4
AcadAch AcadAch1 AcadAch2
PERF PERF1-PERF5
MH MH1 MH2 MH2R MH3 MH4 MH5 MH5R
PP1-PP11
SM FB Insta Audiovis Snapchat Whatsapp Other
TV TVSport TVMusic TVRom TVPT TvPorn TVWMag TVMMag TVSMag
PASSFB PFB1-PFB5
ACTFB AFB1-AFB8
PR PR1 PR2 PR3 PR4 PR5 PR5R PR6 PR7;

USEVARIABLES = PP1 PP2 PP3 PP4 PP8 PP9 PP10 PP11;

GROUPING = Country (1=Austria 2=Belgium 3=Spain 4=Korea);

MISSING = all (-999);

ANALYSIS: MODEL = CONFIGURAL METRIC SCALAR;

MODEL:
ProfPP by PP1-PP2;
SocPP by PP3-PP4;
SexPP by PP8-PP9;
RomPP by PP10-PP11;

OUTPUT:SAMPSTAT STDYX TECH1 MODINDICES;
 Bengt O. Muthen posted on Thursday, September 14, 2017 - 4:01 pm
Q1: Try it.

Q2: I think you can free up some, but not all of them. If the rules assume that certain parameters are equal for model=metric, for example, freeing them up could confuse things. I don’t think there are any restrictions though so it’s just a matter of trying it out.
 Orpha de Lenne  posted on Friday, September 15, 2017 - 6:30 am
Dear Dr. Muthen,

Thank you for your answer.

I now tested the configural, metric and scalar model separately. I used class-specific syntax to free up certain factor loadings for the metric model.

Yet, I get no model estimation and the following error:

THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED. CHECK YOUR MODEL. PROBLEM INVOLVING THE FOLLOWING PARAMETER:
Parameter 107, Group KOREA: ROMPP WITH PROFPP
THE CONDITION NUMBER IS -0.247D-06.

What is the correct command in the syntax to free up parameters while using the metric or scalar command?

Below my syntax:
USEVARIABLES = PP1 PP2 PP3 PP4 PP8 PP9 PP10 PP11;

GROUPING = Country (1=Austria 2=Belgium 3=Spain 4=Korea);

MISSING = all (-999);

ANALYSIS: MODEL = METRIC;

MODEL:
ProfPP by PP1-PP2;
SocPP by PP3-PP4;
SexPP by PP8-PP9;
RomPP by PP10-PP11;

MODEL Korea:
SocPP by PP1;
SocPP by PP2;
 Bengt O. Muthen posted on Friday, September 15, 2017 - 5:57 pm
We need to see your full output - send to Support along with your license number.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: