Model Test PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 S.Arunachalam posted on Thursday, June 28, 2012 - 11:03 am
Hello Prof. Muthen

A clarification on usage of Model Test in multi-group analysis.I have two groups, and trying to check if the regression parameter with label a is equal to parameter b.

When I ran the code
Model test:
a = b
wald-chi sq test is non-significant
Wald Test of Parameter Constraints
Value 0.981
Degrees of Freedom 1
P-Value 0.3219

However if use
Model test:
a - b = 0

the results are
Wald Test of Parameter Constraints
Value 7.280
Degrees of Freedom 1
P-Value 0.0070

I thought in both ways am testing the same difference in parameter test. Please help
 Eric Deemer posted on Friday, July 18, 2014 - 1:08 pm
I have a question similar to the previous poster. I'm doing a multiple group analysis and I want to test the noninvariance of a particular path coefficient. Do I need to label the coefficient in the overall model command? My input is below:

MODEL: st BY svs1-svs8;
scientid by atmr sdcr sciid;
tapp BY newagq1 newagq4 newagq7;
sapp by newagq2 newagq5 newagq8;

tapp on scientid st;
sapp on scientid st;
scientid on st;

MODEL INDIRECT:
tapp ind scientid st;
sapp ind scientid st;

Model other:
scientid on st(a);
Model science:
scientid on st(b);

Model test:
a=b;

Eric
 Bengt O. Muthen posted on Friday, July 18, 2014 - 4:13 pm
The way you label them is correct. Model test should have

0 = a-b;
 Eric Deemer posted on Friday, July 18, 2014 - 4:37 pm
Thanks so much, Bengt.

Eric
 Aidan posted on Monday, May 11, 2015 - 8:37 am
Dear Bengt,

I have run a multigroup latent growth model and am trying to compare the two groups for significant differences in the mean and variance of the two latent factors. Using the code below (I have also tested an alternative form of the MODEL TEST command, where p1=p21, and so on) produces a Wald test with a very low, significant, p-value for the omnibus test of all four parameters. However, when I examine each parameter one-by-one, none of them show a significant difference between groups.

Model Group 1:
[int] (p1);
[slope] (p2);
int (p3);
slope (p4);

Model Group 2:
[int] (p21);
[slope] (p22);
int (p23);
slope (p24);

Model test:
0 = p1-p21;
0 = p2-p22;
0 = p3-p23;
0 = p4-p24;

Can you tell me if I am doing anything wrong? It seems odd that none of the individual parameters are close to significance while the omnibus test is significant - I am not sure how one would report this in a useful fashion, since no specific difference between the two groups can be identified.

I have searched the forum archives for an answer, but haven't seen this described before - apologies if I have overlooked anything. Thank you.
 Bengt O. Muthen posted on Monday, May 11, 2015 - 6:15 pm
The joint test has more power to reject than each individual test so the outcome is quite possible. You can try to use a likelihood-ratio chi-2 difference test instead of the Wald test to check if you get the same result.
 Pia H. posted on Thursday, October 06, 2016 - 1:02 am
Dear Prof Muthén

I want to compare regression coefficients of latent (interaction) variables between two groups using the model test command, but I get an error message in my output:

ANALYSIS:

TYPE = COMPLEX MIXTURE RANDOM;
ITERATIONS = 5000;
PROCESSORS = 8;
model = nocovariances;
ALGORITHM=INTEGRATION;
INTEGRATION = MONTECARLO (5000);
CHOLESKY = OFF;
MCONVERGENCE = .01;

MODEL:
%OVERALL%

(input omitted)

g_PSYCHO on g_TRAUMA (a);
g_PSYCHO ON age_cen (c);
g_TRAUMA ON age_cen (e);

agextrauma| age_cen XWITH g_TRAUMA;
g_PSYCHO ON agextrauma (g);

%MAOA#2%

g_PSYCHO on g_TRAUMA (b);
g_PSYCHO ON age_cen (d);
g_TRAUMA ON age_cen (f);
g_PSYCHO ON agextrauma (h);

Model test:
0=a-b;
0=c-d;
0=e-f;
0=g-h;


*** ERROR in MODEL command
Random effect variables can only be declared in the OVERALL model.

Thank you!
Best, Pia
 Bengt O. Muthen posted on Thursday, October 06, 2016 - 10:39 am
We need to see the full output - send to Support along with your license number.
 Qiong Wu posted on Thursday, June 15, 2017 - 9:41 pm
Dear Bengt,

I'd like to follow up with the discussion where you suggested "The joint test has more power to reject than each individual test so the outcome is quite possible. You can try to use a likelihood-ratio chi-2 difference test instead of the Wald test to check if you get the same result." I am testing the difference of mediation between two groups. The joint test had more power than individual tests about a1b11 - a2b12, a1b21 - a2b22. What is the syntax to use a likelihood-ratio chi-2 difference test instead of the Wald test?

MODEL constraint:
new(a1b11 a1b21 a2b12 a2b22);
a1b11=a1*b11;
a1b21=a1*b21;
a2b12=a2*b12;
a2b22=a2*b22;

Model test:
0 = a1b11 - a2b12;
0 = a1b21 - a2b22;
 Bengt O. Muthen posted on Friday, June 16, 2017 - 5:58 pm
The likelihood-ratio chi-2 test is obtained from 2 runs. One run with equality and one without.
 Rimantas Vosylis posted on Thursday, May 24, 2018 - 12:56 am
Hi,
I am doing multi-group path analysis with 3 groups. I want to test if one of the paths in the model is different in three groups.

I label the paths in three groups as:
crosa101 crosb101 crosc101

To do the test, I am using MODEL TEST command, which I set up like this:
0 = crosa101 - crosb101 - crosc101;

The result is this:
Wald Test of Parameter Constraints

Value 0.194
Degrees of Freedom 1
P-Value 0.6596

My question is why do I get only one degree of freedom. It's supposed to be two, isn't it?

thank you so much
 Bengt O. Muthen posted on Thursday, May 24, 2018 - 3:48 pm
Use

0 = crosa101 - crosb101;
0 = crosa101 - crosc101;
 Rimantas Vosylis posted on Thursday, May 24, 2018 - 10:33 pm
Hello,

I also have another question on Wald's test:
I am using MLR in my analysis, and I am wondering if Wald's test does produce a corrected Chi-square (adjusted by scaling correction factor), or is it uncorrected ML-based chi-square?
Thank you so much for your responses!
 Bengt O. Muthen posted on Friday, May 25, 2018 - 12:55 pm
Yes, the Wald test is corrected with MLR, that is, it is using the MLR-based (not ML-based) covariance matrix for the parameter estimates.
 Georgios Sideridis posted on Friday, March 22, 2019 - 6:56 am
Hi,
Can we get multiple Wald tests with 1df in a single run, not joint tests?
Thank you
 Bengt O. Muthen posted on Friday, March 22, 2019 - 2:57 pm
Not in a single run.
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: