Testing equality of coefficients in multigroup model

186 views
Skip to first unread message

Felix B.

unread,
Mar 7, 2024, 3:21:51 AM3/7/24
to lavaan
Dear all,
my main model is a RI-CLPM which I estimated using lavaan, which works fine. Now I want to extend the model and compare 2 groups (boys vs girls). This also works fine. Next I would like to compare some regression coefficients between the two groups. Say, the coef of interest for boys is 1.4 and for girls 1.7. I would like to conduct something like a Wald-Test that tells me whether these two coefficients are significantly different between the two groups. I researched this a lot but all I found are either global model fit comparisons or coefficient tests between regressors (but not between groups). Help is much appreciated!

Daniel Morillo Cuadrado

unread,
Mar 7, 2024, 4:08:23 AM3/7/24
to lav...@googlegroups.com
If I understand you well, I think what you need is a likelihood-ratio test comparing two models: One with your coefficient of interest constrained to be equal across groups, and another one without that constraint.

Although this is a global model fit comparison between two models, you are only performing a contrast on that precise parameter as it is the only difference between the two; you will notice that the statistic is distributed as chi-squared with precisely one degree of freedom.

Please LMK if this answers your question.

--
Daniel Morillo, Ph.D.
GitHub | ORCID


El jue, 7 mar 2024 a las 9:21, 'Felix B.' via lavaan (<lav...@googlegroups.com>) escribió:
Dear all,
my main model is a RI-CLPM which I estimated using lavaan, which works fine. Now I want to extend the model and compare 2 groups (boys vs girls). This also works fine. Next I would like to compare some regression coefficients between the two groups. Say, the coef of interest for boys is 1.4 and for girls 1.7. I would like to conduct something like a Wald-Test that tells me whether these two coefficients are significantly different between the two groups. I researched this a lot but all I found are either global model fit comparisons or coefficient tests between regressors (but not between groups). Help is much appreciated!

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/lavaan/2ea61218-d8fe-4761-a962-e0747014cf28n%40googlegroups.com.

Felix B.

unread,
Mar 7, 2024, 4:46:31 AM3/7/24
to lavaan
Thanks for the reply. If you think this is the best way to achieve my goal, sure, I am open to this approach. How would this work? In the documentation I see that I can restrict all "regressions" in total, see https://lavaan.ugent.be/tutorial/groups.html#constraining-groups-of-parameters-to-be-equal-across-groups ("regressions: all regression coefficients in the model"). But how would this look like for a single regression parameter, say the effect of x1 on y2?

Daniel Morillo Cuadrado

unread,
Mar 7, 2024, 5:23:59 AM3/7/24
to lav...@googlegroups.com

I think this is what you are looking for.

--
Daniel Morillo, Ph.D.
GitHub | ORCID

Felix B.

unread,
Mar 7, 2024, 8:50:35 AM3/7/24
to lavaan
Thanks, I will test this.

Christian Arnold

unread,
Mar 7, 2024, 8:55:58 AM3/7/24
to lavaan

Felix B.

unread,
Mar 8, 2024, 2:52:07 AM3/8/24
to lavaan
Thanks a lot. Finally it clicked in my head and how this notation works. I tested both approaches and they gave sensible results.
Reply all
Reply to author
Forward
0 new messages