MIMIC model for DIF detection PreviousNext
Mplus Discussion > Structural Equation Modeling >
Message/Author
 Chayut Piromsombat posted on Monday, October 11, 2010 - 1:53 pm
Dear Dr.Muthen,

I played with the MIMIC model for DIF detection and found that the results from significance test of the coefficient for grouping variable on the studied item and the chi-square test for nested model comparison (i.e., No DIF vs DIF models) are different. For example, model comparison suggested that DIF occurs in an item but the coefficient of grouping variable on that item is not statistically significant.

Here's the command I used for DIF model.

MODEL:
ability BY i1-i20;
ability ON group;
i7 ON group; !uniform DIF

For No DIF model, the last line was deleted.

Is it possible that two approaches (sig. test and nested model comparison) result in different ways? If so, which one should be used?

Thank you
 Linda K. Muthen posted on Monday, October 11, 2010 - 2:15 pm
I think rather than deleting the last line, you need to specify:

i7 ON group@0;
 Chayut Piromsombat posted on Monday, October 11, 2010 - 4:26 pm
I think another question has arisen. I tried to run No DIF model by (1) deleting the "i7 ON group;" line and (2) constraining the parameter "i7 ON group@0;" as you suggested.

Here's the result:
(1)
Chi-Square Test of Model Fit

Value 304.999
Degrees of Freedom 21
P-Value 0.0000

Loglikelihood

H0 Value -1123.327
H1 Value -970.828

Information Criteria

Number of Free Parameters 21


(2)
Chi-Square Test of Model Fit

Value 304.999
Degrees of Freedom 21
P-Value 0.0000

Loglikelihood

H0 Value -1123.327
H1 Value -970.828

Information Criteria

Number of Free Parameters 21


What is the difference I should expect from using "@0" instead of deleting the command line? And how using "@0" can explain the different results between the significance test and model comparison that I've asked in previous post?

Thank you so much,
 Linda K. Muthen posted on Monday, October 11, 2010 - 5:06 pm
If that makes no difference, I would need to see the relevant outputs and your license number at support@statmodel.com.
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: