Baseline model in Mplus PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Qi posted on Thursday, October 02, 2008 - 1:48 pm
is the "Chi-Square Test of Model Fit for the Baseline Model" in Mplus comparable to the "Chi-Square for Independence Model" in LISREL 8.53? I want to make sure in both software I am running the same model by comparing the baseline model's df and Chi-square value, but it seems that I heard that the baseline model in Mplus is a little different (so it can result in different CFI fit stat value). How different is it? Thanks for your advice!
 Linda K. Muthen posted on Friday, October 03, 2008 - 9:00 am
Where y represents dependent variables and x represents independent variables, the Mplus baseline model is:

y with y is zero
y with x is zero
x with x is not zero

The LISREL baseline model is:

y with y is zero
y with x is zero
x with x is zero

Because the LISREL baseline model fits very poorly, model fit is exaggerated.
 Yue Yin posted on Thursday, November 09, 2017 - 1:13 pm
How about if there are two x's? Is the correlation with x1 and x2 also 0? Or they stay the same ?
 Bengt O. Muthen posted on Friday, November 10, 2017 - 4:37 pm
"x with x is not zero" implies x1 with x2 not zero (but equal to sample covariance).
 Jing Davis posted on Thursday, June 21, 2018 - 8:38 am
Hi Dr. Muthen,

I tried to run a bifactor model using all categorical variables. But, I got the resutls:
The chi-square value for MLM, MLMV, MLR, ULSMV, WLSM and WLSMV cannot be used
for chi-square difference testing in the regular way. MLM, MLR and WLSM
chi-square difference testing is described on the Mplus website. MLMV, WLSMV,
and ULSMV difference testing is done using the DIFFTEST option.

I was wondering how to perform the DIFFTEST for the model fit. Do I need to run two different model twice - one is the bifactor model and the other is the baseline model so that the DIFFTEST can be computed? If so, what is the Baseline model of Bifactor model in Mplus?

Thank you for your time in advance!
 Bengt O. Muthen posted on Thursday, June 21, 2018 - 3:44 pm
DIFFTEST is needed only if you compare two "H0" models. If you want the chi-square for an H0 compared to the usual unrestricted H1 model, then you already have that in the regular output.
 Jing Davis posted on Friday, June 22, 2018 - 6:43 am
Hi Dr. Muthen,


Thank you for the quick response. It is really helpful.

In the Bifactor model analysis, the estimator method is WLSMV. So, I guess that is why in the output, there is a comment line under the Chi-square test of Model Fit. Please see the following comments.

Chi-Square Test of Model Fit

Value 232.077*
Degrees of Freedom 187
P-Value 0.0139

* The chi-square value for MLM, MLMV, MLR, ULSMV, WLSM and WLSMV cannot be used
for chi-square difference testing in the regular way. MLM, MLR and WLSM
chi-square difference testing is described on the Mplus website. MLMV, WLSMV,
and ULSMV difference testing is done using the DIFFTEST option.

RMSEA (Root Mean Square Error Of Approximation)

Estimate 0.015
90 Percent C.I. 0.007 0.020
Probability RMSEA <= .05 1.000


CFI/TLI

CFI 0.970
TLI 0.963

If I interpreted correctly, it looks like the Chi-square test indicates that the Model does not fit. But, the RMSEA says that the model fits well. Did I missing anything here?

Again, thank you for your time.
 Bengt O. Muthen posted on Friday, June 22, 2018 - 3:54 pm
Fit indices often disagree. Look for ways to improve the chi-square fit by working with Modindices.
 Jing Davis posted on Monday, June 25, 2018 - 2:58 pm
Dr. Muthen, thank you for the clarification! I'll run the output with Modindices function to see how the results will look like.
 Jing Davis posted on Wednesday, June 27, 2018 - 8:42 am
Hi Dr. Muthen,


I ran the command Modindices. In the output, there are 999 in almost everywhere under ON, WITH, and BY statement. The following are the excerpts from the output. From your answers to other posts' questions regarding the 999 in Modindices, I found that 999 means MI can not be computed. However, since my data are categorical, the estimation method is WLSMV. My question is: Can I still trust the results? How can I interpret it? Is 999 in this case due to the MI missing or there are some issues going on in the model I tried to fit. Thank you.

ON/BY Statements

V1 ON F1 /
F1 BY V1 999.000 0.000 0.000 0.000
V1 ON G /
G BY V1 999.000 0.000 0.000 0.000

ON Statements

F1 ON V1 999.000 0.000 0.000 0.000
F1 ON V2 999.000 0.000 0.000 0.000

WITH Statements

G WITH F1 999.000 0.000 0.000 0.000
G WITH F2 999.000 0.000 999.000 999.000
V1 WITH F1 999.000 0.000 0.000 0.000
V1 WITH G 999.000 0.000 0.000 0.000
V2 WITH F1 999.000 0.000 0.000 0.000
 Bengt O. Muthen posted on Wednesday, June 27, 2018 - 4:11 pm
Send your output 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: