Mplus CFA Measurement Invariance (configural metric scalar)

[@attachmentfocusedgrowth]

Oh my goodness. I’ve been looking EVERYWHERE for the appropriate syntax for my measurement invariance analysis. This video FINALLY showed me what I needed! Thanks ❤

[@kangma4462]

thanks for keeping sharing these tutorials. They have been helpful. Also, any posts on longitudinal measurement invariance be shared? thx

[@Gaskination]

Oooh, nope. I have never done that. However, I suppose you could do it in a similar way if you can group the early and late responses and then use that timing variable as the grouping variable.

[@evagromme4688]

thank you for the great explanation! I tried to run this with imputed data and it gave me an error. Do you know how measurement variance can be used with imputed data? Thank you!

[@Gaskination]

Sorry, I'm not sure.

[@yuheiinoue7616]

Hi @James Gaskin, thank you very much for posting this. I have Mplus 7.0 which does not have the option for configural metric scalar. Could you let me know if there is a way to perform the measurement invariance test without this option?

[@Gaskination]

You could do a chi-square difference test in Excel after running each different model. I have a chi-square test in my Excel stats tools on the homepage of statwiki.

[@antonbastian7005]

One Question: What did you do to get an output index (its called output section in your output) to navigate in the output? I wasn't able to find something like that...

[@Gaskination]

Go to the Mplus menu (between View and Plot) and select the HTML output. This will produce the linked guide at the top. That allows you to link to sections and also zoom in and out.

[@nasseralresaini8176]

Hi, I tried the option "model is configural metric scalar" but it didn't work with higher-order factors. So, I returned to the conventional way by writing all constraints manually.

[@Gaskination]

That's unfortunate. You could also just run it without the higher order factor, since invariance is mainly concerned with the measures.

[@Danielcb1966]

Hello. Thank you very much for the tutorial. Can I ask a question? How to prepare a data file to run the factorial invariance method in MPLUS?

[@Gaskination]

The only preparation is to make sure your categorical (grouping) variable is coded correctly (as a number representing the group memberships). For example, if your grouping variable is industry, you might have it coded as 1, 2, and 3 where 1=manufacturing, 2=retail, and 3=service (but only the numbers are in the dataset).

[@xiaozhang3083]

Hello James, I wonder in the model with one latent variable and three measured variables, is it okay to just have measurement invariance for the latent variable? Will the measured variables will pass just for alpha? Thanks in advance!

[@Gaskination]

If doing a multigroup analysis, then measurement invariance should be tested on all measures of latent factors. So, single item measures (not part of latent factors) do not need to be tested in the invariance check.

[@xiaozhang3083]

@@Gaskination Thank you so much!

[@irinavanzhula9266]

The video mentions that most of the code except for analysis should be familiar, but it is not the case for me. Is there a different video that explains the model code?

[@Gaskination]

haha yes. It would be hard to start from this video. Here is an ordered playlist with relevant videos starting from the beginning. Hope this helps: kzbin.info/aero/PLnMJlbz3sefL983xxEgbGjRc6mVIsIcAg

[@MrEls2010]

How do you get Mplus to view the output to look like you have done in the video?

[@Gaskination]

You can increase the font size by holding CTRL while scrolling up, or by clicking CTRL+. The HTML output comes from the View menu.

[@xiaozhang3083]

Thanks for your video! Quick question: why you set CR - PART @1 and what's the meaning to it? Is that necessary to do so? Thanks in advance!!

[@Gaskination]

This is to set the variance of those latent factors to equal 1.00. I do this instead of having a regression weight constraint on one of the indicator paths per factor (which is the default that I'm canceling with the astrisk).

[@xiaozhang3083]

@@Gaskination Thank you very much for the explanation! So it's okay to use both way (default or your way)? Thank you that's so helpful!

[@xiaozhang3083]

BTW I also wanted to know (might be a stupid question), since I have 7 latent variables in two countries, does it matter to do ME together or separately (one latent at a time; grouping or using separated data file) or it will be the same? Thank you so much!

[@Gaskination]

@@xiaozhang3083 Yes, either is fine. Often a researcher will move the constraint to the factor variance so that they can test measurement differences for all indicators.

[@Gaskination]

@@xiaozhang3083 I'm not sure what "ME" is. Usually models should include all latent factors together, not in separate models.

[@drewporter1318]

I am confused by the difference in degrees of freedom between metric and configural models. If you are constraining the factor loadings to be equal between groups and you have 12 factor loadings in the overall model, then shouldn't the difference in degrees of freedom be 12 and not 9?

[@Gaskination]

Yes, except the forced constraint of variance at 1 for each of the three latent factors reduces the DF by 3.

[@drewporter1318]

@@Gaskination How does placing a constraint reduce DF? You are estimating fewer parameters when placing constraints.

[@Gaskination]

@@drewporter1318 My mistake. I misunderstood your question. I'm not sure why the difference is only nine. It should be 12, as you say, but perhaps shifting the constraint around (from the first indicator to the latent factor) has tampered with the DF.

[@theCOSlives]

THANK YOU SO MUCH!!!!!!!!!!!!!!!!!!!!!!!!!!