DIF with bifactor model PreviousNext
Mplus Discussion > Categorical Data Modeling >
Message/Author
 Len Simms posted on Wednesday, September 07, 2011 - 2:02 pm
I would like to test DIF across two groups using the bifactor model. in particular, I need the ability to test for differences in both loadings and difficulties across the two groups along the general factor. How can this be done with mplus?

Also, when mplus does DIF, how does it deal with groups that have different mean levels on the underlying general factor? In a typical IRT analysis, some adjustments are needed on the betas if there is a mean theta difference across groups. How is this done in mplus?

Thanks,
Len Simms
 Bengt O. Muthen posted on Wednesday, September 07, 2011 - 5:12 pm
You give labels to the parameters of the loadings and thresholds that you are interested in, and make those labels the same vs different in the groups you want to analyze. 2 times the loglikelihood difference for the two models gives the chi-square test of DIF if you use ML. With WLSMV you use DIFFTEST.

The different groups have different factor means and variances as the default. No adjustements are needed.
 Len Simms posted on Wednesday, September 07, 2011 - 9:07 pm
Thank you Dr. Muthen. That should get us started.
 mohammad ahmadi posted on Wednesday, November 30, 2016 - 3:52 am
Hi Dear Linda/Bengt,
I would like to fit a bifactor IRT model in Mplus. I have come up with the following syntax , items are mixed items format test, Is it correct model has been set
Thank you.
Sincerely,
mohammad ahmadi

VARIABLE : NAMES ARE i1-i147;
USEVARIABLES ARE i2-i147;
CATEGORICAL ARE i2-i147;
Missing are all (9)
ANALYSIS : ESTIMATOR IS ML;
LINK IS LOGIT;
MODEL: G by i2* i3-i147;
G@1;
[G@0];
F1 BY i2-i14*;
F1@1;
[F1@0];
.
.
F10 BY i129-i147*;
F10@1;
[F10@0];
! uncorrelated factors
G with F1-F10@0;
F1 with F2-F10@0;
.
.
F9 with F10@0;

! Item thresholds free
! First threshold
[i2$1];
.
.
[i147$1];

! Second threshold
[i2$1];
[i3$1];
.
[i146$1];
[i147$1];

PLOT : TYPE IS PLOT1 PLOT2 PLOT3;

SAVEDATA : SAVE = FSCORES;
FILE IS bimono.dat;

OUTPUT : STDYX;
RESIDUAL TECH1 TECH5 TECH8 TECH10;
 Bengt O. Muthen posted on Thursday, December 01, 2016 - 10:50 am
This looks correct but note that the thresholds are free by default.
 mohammad ahmadi posted on Monday, January 16, 2017 - 5:09 am
Hi Dear Linda/Bengt
I would like to fit a Bifactor Multidimensional Item Response Theory (MIRT) Differential Item Functioning (DIF) Model With a Bayesian Estimation in Mplus. But I want to do with An iterative item purification procedure and In addition, the set of anchor items contain at least one item from each testlet that free of DIF to ensure that the anchor set does not contain items that show DIF.
1-Is there such a facility in this software?
2-can compute impact(difference between latent ability of groups?
3- can compute testlet effect?
4-can compute variance of testlet effect?
5- can compute mean of ability distribution?
Thank you.
Sincerely,
mohammad ahmadi
 Bengt O. Muthen posted on Monday, January 16, 2017 - 9:52 am
Mplus does not automate "an iterative item purification" but perhaps that is something you do.

1. Yes you can specify that certain items don't have DIF.

2. yes.

3-4. I don't know what that is but if a testlet effect is captured by a factor - then yes.

5. yes
 mohammad ahmadi posted on Tuesday, January 17, 2017 - 7:22 am
Thank you Dr. Bengt O.
 mohammad ahmadi posted on Sunday, January 22, 2017 - 9:12 pm
Hi Dear Linda/Bengt
I have implemented a bi-factor model with ten Testlet. But two Testlet has items with positive and negative factor loading on Testlet, actually any testlet measure two different trait:
1-Should any Testlet be converted into two Testlet and implemente a bifactor model with twelve testlet?
2- Also a number of items have Non-significant factor loadings on any Testlet, Should in final analysis, these items be removed

Please guide me in this case
Thank you.
Sincerely,
mohammad ahmadi
 Bengt O. Muthen posted on Monday, January 23, 2017 - 3:39 pm
You want to ask these general analysis strategy questions on SEMNET.
 mohammad ahmadi posted on Friday, January 05, 2018 - 12:38 am
Hi Dear Dr. Linda/Bengt,
I used the syntax of generalized partial credit model from your site. But I'm facing this error, please guide me.

TITLE: this is an example of a generalized partial credit
item response theory (IRT) model
DATA: FILE IS ex5.5part1.dat;
VARIABLE: NAMES ARE u1-u20;
CATEGORICAL ARE u1-u20 (gpcm);
ANALYSIS: ESTIMATOR = MLR;
MODEL: f BY u1-u20*;
f@1;
OUTPUT: TECH1 TECH8;
PLOT: TYPE = PLOT3;

*** ERROR in VARIABLE command
Categorical variable I20 contains invalid and/or non-integer values.
Problem with: GPCM
Thank you.
Sincerely,
mohammad ahmadi
 Linda K. Muthen posted on Friday, January 05, 2018 - 6:34 am
It sounds like you are using an older version of the program that does not have the GPCM setting of the CATEGORICAL option. If not, please send the output and your license number to 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: