Testing for measurement invariance using AMOS (Nov 2020)
HTML-код
- Опубликовано: 11 сен 2024
- In this video I provide demonstrations of how to set up and test for measurement invariance using AMOS. This video is a supplement to the main presentation found in a Powerpoint that can be downloaded at:
drive.google.c...
The zip file referenced in the video containing the data for the examples, AMOS files, and the Excel calculator for comparing models can be downloaded here: drive.google.c...
Hi,
I'm currently revising a journal manuscript and one of the reviewers actually asked me to perform MGCFA Measurement invariance testings with my newly developed scales. Your video is basically everything I needed and you saved my life for revising my manuscript!! Thank you so much
You are very welcome! I'm glad you found it helpful. Cheers!
Love to see your contents, it is very useful and I fill my thirst and hunger for knowledge and research techniques. Love you Sir
Hi there, thanks for your comment. I hope you find the presentation helpful. By the way, I made a couple of updates to the powerpoint this morning. So be sure to download the latest edition!
@@mikecrowson2462 thanks for letting me know about it
Thank you so much. This is a very clear video demonstration. I was desperate looking for the Scalar invariance demonstration in AMOS for ages. This is very helpful.
You are very welcome. Thanks for visiting! Cheers!
Incredibly useful, saved me weeks or more
Thanks, Matthew. I'm really glad you found this useful! Thanks for visiting! Best wishes.
Thank you so much for such an amazing video. You saved me many times in the past as well. Mike, how do we do this if we have samples from two countries? What will be the variable name?
Great and really helpful. Thank you.
Hi Mike,
Thank you for your video!
I have to validate an instrument and assess the measurement invariance between 2 groups (Italian VS international students). I have administered the questionnaires in 2 different languages (italian and english).
I wonder whether there are differences between the following approaches?:
1) perform a multigroup CFA to assess measurement invariance
2) perform the CFA, model comparison and, only after, the measurement invariance between italian and dutch students (as you showed in this youtube video) (So in this second case I would not perform the multigroup CFA but only measurement invariance).
Which one do you think it would be the best approach to follow?
Thank you in advance
Hi Mike, Thanks for these videos. They are super helpful. I had a question when running 2nd order or higher-order models in AMOS. When testing invariance (especially scalar invariance), do I adjust all the information for the 2nd order factor only. Because some information is blanked out on the first-order factor as they have error terms that are technically classed as reflective measures of the latent 2nd order factor.?
Sir, thank you so much for the video, and you have explained clearly for every step. Just wondering, if I want to test the measurement invariance for the second-order construct, how should I do?
+1
Hello, thank you very much for this video. I still have one question: If I got a non-significant result in both difference tests, is it still necessary to calculate the partial invariance?
thank you very much!
Thanks!
Hey Jason, great to hear from you! Hope you find it helpful! btw, I had to add in a slide this morning (inadvertently left out a test). So download the fresh copy :)
Is it necessary to constraints all those parameters? some videos constrained only factor loading and covariance, but not the mean of the latent factors.
Hi there, and great question. The invariance testing approach you take seems to depend largely on what your aims are with the measure you are testing. If you check out some of the references I have in the Powerpoint, a number of the authors (such as Milfont & Fischer, 2010) describe two aspects of invariance: measurement and structural invariance. Measurement invariance pertains to the set of procedures to address configural, metric (factor loadings), scalar (intercepts), and potentially strict (error variances) invariance. If you are are broadly testing measurement invariance, then you'd want to go through all of these. If your aim is ultimately to use the measurement to compare groups on latent means (or if you use scaled scores), then you'd need to test at least up through the intercepts, since you want to ensure that your groups are equivalent on the construct and that that differential responding is not due to factors related to the item that are irrelevant to measurement of the construct. The test of differences in latent means is a test of one type of structural invariance. (and this is actually what I'm addressing in the video). Other types of structural invariance one might test for are variances and covariances between groups on latent factors. This might be of particular focus if you are not testing differences in means but differences between groups in relationships among variables - and where perhaps you have questions pertaining to moderation of relationships among latent constructs as a function of group membership. In that case, you would not necessarily need to go through testing for scalar invariance as the item intercepts do not bear on the structural relations among latent variables. So, the main test of invariance would go up through tests of metric invariance (i.e., factor loadings). Byrne (2010) does a good job of breaking these different purposes out. Chapter 7 involves testing up through metric invariance and then testing for invariance of the variances and covariances; whereas the model in her Chapter 8 involves testing for latent means and she goes all the way up through tests for scalar invariance to justify the comparison of means. As you can see, there are some 'finer points' behind the strategy you take invariance testing. But that's my read of this literature. Check out the articles by Milfont & Fischer (2010) and Putnick, D.L., & Bornstein, M.H. (2016) in the references of the Powerpoint. Cheers!
@@mikecrowson2462 Thank you for your sincere answer. My focus of research is to compare relationships among variables, that is, I want to check moderating effect. I guess, I can just go through test of metric invariance!
@@jennisoo-heelee1563 I think that's about right, if you are mainly planning on moderated relations among latent variables. I believe you may need to proceed with testing equality of variances of your latent variables first before testing equality of relationships among them. You might check out those articles to make sure that is the sequencing you should take after metric invariance. Cheers!
@@mikecrowson2462 Thanks! The articles look really helpful.
Hello professor, after watching your video, I ran the analysis and found that factor loadings and variance were invariant across two groups, but covariance was not invariant. For the result of covariance, I interpret that since these two groups are structurally different, within multigroup path analysis, there must be some differences in path coefficients. But, after running the multigroup analysis, it shows that path coefficients were invariant across two groups --> I also followed the steps of your video (multigroup analysis 2). Is this result possible? Or do I made some mistakes? Only difference between your multigroup analysis video and my analysis is that individual path coefficient comparisons were conducted along with factor loadings constraints (since I used the measurement invariance test model right before this analysis). Thanks!!
I was thinking that to establish that you have partial scalar invariance model , you should compare your partial scalar model chi square value with your metric invariance model, so that you can claim no difference????
Hi there. You are absolutely correct. I inadvertently left that out of the presentation. I have just updated the Powerpoint with a slide demonstrating that the partial scalar invariance model was not significantly different from the metric model. Thanks for the catch! Cheers.