Multidimensional IRT script PreviousNext
Mplus Discussion > Confirmatory Factor Analysis >
Message/Author
 J C posted on Friday, January 24, 2014 - 4:25 pm
Hi Dr Muthen,

I was reading your paper on IRT modeling:
Muthén & Asparouhov (2013). Item response modeling in Mplus: A multi-dimensional, multi-level, and multi-timepoint example.

The paper said that scripts are available on the website, but I wasn't able to find it. I was able to find relevant information in the following link (https://www.statmodel.com/recentpapers.shtml ) , but couldn't locate the script itself.

Would you please help and point me to the right place to look? This would help me a lot! Appreciated!
 Bengt O. Muthen posted on Friday, January 24, 2014 - 4:46 pm
You have the Table 6 run on our website:

http://www.statmodel.com/download/11-22_IRT.out

But I can also post some further runs in the paper.
 Wendy Sun posted on Monday, March 23, 2015 - 2:26 pm
Dear Dr. Muthen,

Can you post the script for Table 2 in the paper: A multi-dimensional, multi-level, and multi-timepoint example?

Many thanks!

Wendy
 Bengt O. Muthen posted on Monday, March 23, 2015 - 2:37 pm
This is done in line with UG ex 4.7, except declaring the items as categorical.
 Wendy Sun posted on Monday, March 23, 2015 - 3:41 pm
Thanks Dr. Muthen.
Can you guide me any example to do a multi-dimensional irt/Rasch model using Mplus?

Thanks!
 Bengt O. Muthen posted on Monday, March 23, 2015 - 4:14 pm
There is nothing to it. See UG ex 5.5 and for 2 factors, each having the equal loadings, change to

f1 by y1-y10* (1);
f2 by y11-y20* (2);
f1-f2@1;
 Wendy Sun posted on Monday, March 23, 2015 - 4:40 pm
Thanks for providing the example syntax. I am a beginner to use Mplus for IRT analysis, and I have another question. In your example, if the the response category from y1 to y20 are different, like some of them are dichotomous, and the others are polytomous with different scales (e.g., 3 point and 4 point likert scale), do I need to do anything, like defining/grouping the items if I want to see the estimation of thresholds?
Thanks.
 Bengt O. Muthen posted on Monday, March 23, 2015 - 5:09 pm
No, you don't need to do anything. Mplus checks how many categories each item has and assigns as many threshold parameters (related to the difficulties) as needed.
 Shanshan Wang posted on Friday, June 12, 2015 - 9:31 am
Hi Dr. Muthen,

Will you please direct me to the scripts for section 6: 2-level item bifactor explanatory factor analysis, section 7: 2-level bifactor confirmatory factor analysis, and section 8: 2-level item bifactor confirmatory factor analysis with random factor loadings?

Also, I wonder if you can share the data file for this paper.

Thank you!
 Bengt O. Muthen posted on Friday, June 12, 2015 - 5:58 pm
You find them in "download output files" on our website:

Muthén, B. & Asparouhov, T. (2013). Item response modeling in Mplus: A multi-dimensional, multi-level, and multi-timepoint example. Forthcoming in Linden & Hambleton (2013). Handbook of item response theory: Models, statistical tools, and applications.
download paper download Table 6 output download output files show abstract
 Shanshan Wang posted on Sunday, June 21, 2015 - 1:58 pm
Thank you Dr. Muthen! Can you give me an example to calculate intraclass correlations as showed in Table 5?

Thank you for your help as always.
SW
 Bengt O. Muthen posted on Monday, June 22, 2015 - 5:57 pm
The approach is shown on page 15.
 Yu Hui Zhang posted on Friday, June 22, 2018 - 10:08 am
Hi Drs. Muthen,

I have several questions regarding nominal item response models:

a. It seems that item information is not provided by nominal item response models in Mplus, is my understanding correct?

b. Does Mplus estimate i) second order, and ii) bifactor multidimensional nominal item response models?

Are there examples of a and b in the manual if one could get the item information and estimate those models with Mplus? I don't seem to be able to find them.

Thank you.

Yuhui
 Tihomir Asparouhov posted on Friday, June 22, 2018 - 4:10 pm
a. Yes - it is provided for the types listed here
https://www.statmodel.com/download/MplusIRT.pdf

b. Yes - you will have to combine features from user's guide examples 5.2, 5.6, 5.30
 Yu Hui Zhang posted on Friday, June 22, 2018 - 11:00 pm
Thank you, Dr. Asparaouhov.

Regarding the item information curves for nominal IRT, where would one find them? I opened view plot on the Plot menu after running the model and did not see any ICC and information curves options.

When I specified the variables as categorical rather than as nominal, the ICC and information curves were there. Thank you.
 Tihomir Asparouhov posted on Saturday, June 23, 2018 - 7:30 am
By Yes I meant ... you are correct that they are not available for nominal variables.
 Yu Hui Zhang posted on Saturday, June 23, 2018 - 7:50 am
I see what you meant now, Dr. Asparaohov, thanks. Would that feature be added to the program in the near feature? Thanks.
 Tihomir Asparouhov posted on Monday, June 25, 2018 - 10:17 am
We will add this to our list but you could consider the alternatives: the partial credit model, the ordered polytomous model or using N-1 binary variables instead of one nominal variable.
 Yu Hui Zhang posted on Monday, June 25, 2018 - 7:32 pm
Thank you, Dr. Asparouhov.
 Yu Hui Zhang posted on Thursday, March 26, 2020 - 9:43 am
I ran a multidimensional IRT using the WLSMV-Theta analysis option. The Mplus output, unlike the unidimensional IRT model output, did not contain estimates of item difficulty. and discrimination.

I have a few questions:

1. Do I understand correctly that Mplus 8.0 does not output item difficulty and discrimination parameters for multidimensional IRT?

2. Could one use the same formulae for univariate IRT models for MIRT?

a = discrimination = loading
b = difficulty = threshold/loading

3. I could get test information curve (TIC) for each of the subscale, but not for the multidimensional scale. Are the individual TIC's for MIRT? Or unidimensional IRT?

4. I read online the following quote, made in 2015:

That conversion should get you something close to what you would get from an IRT program but only in the unidimensional case (unless Mplus has updated their conversion documentation accordingly). . In multidimensional IRT models the discrimination parameter isn't as readily available and it is often best to switch to intercepts.

I wonder whether any changes have been in Mplus 8.0 made since then.

Thank you!
 Bengt O. Muthen posted on Thursday, March 26, 2020 - 11:26 am
See the FAQ on our website:

IRT parameterization using Mplus thresholds

where it says that the slope-intercept parameterization (used by Mplus) is more
general.
 Bengt O. Muthen posted on Thursday, March 26, 2020 - 1:05 pm
Regarding 3.: You get the item information curves for each (item, factor) but you provide conditional values for all other factors. It's available in plots.
 Yu Hui Zhang posted on Friday, March 27, 2020 - 12:29 am
Thank you, Dr. Muthen, the FAQ is very helpful.
 Jeff Williams posted on Tuesday, October 20, 2020 - 4:05 am
Following up on the multi-dimensional conversion, the formula for converting the threshold to a difficulty parameter is given by

MDIFF = -d / sqrt(a1^2 + a2^2)

Does this formula change depending on whether the factors for a1 and a2 are correlated or not?

Thanks.
 Bengt O. Muthen posted on Wednesday, October 21, 2020 - 11:19 am
If sqrt(a1^2 + a2^2) represents the SD of the factor contribution, you would have to add 2*a1*a2*Cov(f1, f2) inside the sqrt.
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: