Hello,
Is there a way to test for group invariance in the context of latent class analysis (LCA) in Stata 15 outside of the sem... group() subcommand? One thing I have noticed in the literature is differing advice about whether to include a variable as a covariate and then turn around and use it to test measurement invariance. I have fairly reasonable results from an 3-4 class LCA (five indicators, 8-9 covariates) but I want to know if there is measurement invariance across some of the indicators by race/ethnicity (before I follow up up with latent class regression to examine the way class membership affects a distal outcome. I would like to do something as indicated in the diagram below (not sure if I am following the correct protocol for posting a diagram but I will try):
LCA Figure for Statalist.pdf
where I can examine the impact of race/ethnicity specifically on each indicator.
I have been using the following LCA command:
Stata will not allow you to include a variable as a covariate and then test for measurement invariance. I use this code and get the following error message when I try:
I am looking to do something similar to what is done in this piece:
doi/full/10.1111/cdev.13691
and also this piece where they test model fit for various versions of dif:
https://doi.org/10.1080/10705511.2016.1254049
I have tried diflogistic after the LCA with the following command:
I get reasonable results, though I want to compare two different ethnic group two a single reference group. With the diflogistic command, I would have to lump the two groups together and compare them to the reference group. I don't think I can compare them separately.
Also, how would I go about testing model fit for the diflogistic?. I know some folks also do this in the context of MIMIC models (I guess the distinction for me between MIMIC and LCA in this particular case is semantic).
So in sum, how can I: 1) test ethnicity as a covariate and source of group invariance 2) get model fit statistics for measuring group invariance on individual indicators?
I would appreciate any help you could offer. My data example follows below. Thank you.
input long(stu_id race white hsub lang) double ses long aca double wt long(behave parex save etex mtex) double flun float(select colgpa highdeg f1gpa30)
101101 5 1 4 1 -.25 0 185.478 0 5 . 6 3 33.56 0 0 . 0
101104 8 1 . 1 -.85 0 649.7293 0 7 0 5 5 33.56 1 1 1 0
101105 3 0 . 1 -.8 0 248.753 . . . 3 3 33.56 0 0 0 0
101106 4 . 4 0 -1.41 0 160.605 0 2 0 3 2 33.56 0 0 . 0
101107 4 . 4 0 -1.07 0 220.1699 0 3 0 3 3 33.56 0 . . 0
101108 4 . 4 1 .27 0 172.7619 . 5 . 3 5 33.56 0 0 . 0
101109 8 1 . 1 -.16 1 719.55 0 5 0 5 5 33.56 1 1 1 1
101110 4 . 4 1 -1 0 0 0 5 1 3 2 33.56 . . . 0
101111 3 0 . 1 -1.22 0 224.2212 . 2 . 5 3 33.56 0 0 . 0
101112 3 0 . 1 -.18 0 30.205 0 5 0 6 5 33.56 1 0 1 1
101113 4 . 4 0 -.03 0 150.7823 0 3 1 3 4 33.56 0 1 . 0
101114 3 0 . 1 -.8 0 219.5727 0 6 . 5 3 33.56 0 1 0 0
101117 3 0 . 1 .22 0 0 0 3 0 2 2 33.56 . . . 0
101118 3 0 . 1 .06 0 206.5884 1 5 1 . . 33.56 0 1 . 0
101119 4 . 4 0 -1.35 0 0 . . . 1 3 33.56 . . . 0
101120 8 1 . 1 -.57 0 719.55 0 3 1 2 2 33.56 . . . 0
101122 4 . 4 0 -.63 0 173.3694 0 7 0 3 3 33.56 0 1 0 0
101123 4 . 4 0 -.69 0 183.7529 0 3 1 3 3 33.56 0 0 . 0
101124 5 0 4 1 -.69 0 0 0 6 1 2 2 33.56 . . . 0
101125 5 1 4 0 -1.22 0 0 1 6 1 5 2 33.56 . . . 0
101129 8 1 . 1 .26 1 719.55 1 6 1 6 5 33.56 1 1 1 0
101131 8 1 . 1 .48 1 671.9371 0 5 0 5 5 33.56 1 1 1 0
101132 4 . 4 1 -.16 0 0 0 2 . . 3 33.56 . . . 0
101133 4 . 4 1 -1.16 0 0 0 5 1 3 2 33.56 . . . 0
Is there a way to test for group invariance in the context of latent class analysis (LCA) in Stata 15 outside of the sem... group() subcommand? One thing I have noticed in the literature is differing advice about whether to include a variable as a covariate and then turn around and use it to test measurement invariance. I have fairly reasonable results from an 3-4 class LCA (five indicators, 8-9 covariates) but I want to know if there is measurement invariance across some of the indicators by race/ethnicity (before I follow up up with latent class regression to examine the way class membership affects a distal outcome. I would like to do something as indicated in the diagram below (not sure if I am following the correct protocol for posting a diagram but I will try):
LCA Figure for Statalist.pdf
where I can examine the impact of race/ethnicity specifically on each indicator.
I have been using the following LCA command:
Code:
gsem (f1gpa30 aca select colgpa highdeg<-_cons) (Tra <- ses etex mtex parex behave flun save lang) [pweight=wt], family(bernoulli) link(logit) lclass(Tra 3)
Stata will not allow you to include a variable as a covariate and then test for measurement invariance. I use this code and get the following error message when I try:
Code:
gsem (f1gpa30 aca select colgpa highdeg <- _cons) (Tra <- ses white etex mtex parex save behave flun lang ) [pweight = wt], group(white) ginvariant(coef) family(bernoulli) link(logit) lclass(Tra 3) invalid path specification; group() variable white is not allowed to be an independent variable r(198);
I am looking to do something similar to what is done in this piece:
doi/full/10.1111/cdev.13691
and also this piece where they test model fit for various versions of dif:
https://doi.org/10.1080/10705511.2016.1254049
I have tried diflogistic after the LCA with the following command:
Code:
diflogistic f1gpa30 aca select colgpa highdeg, group(white)
I get reasonable results, though I want to compare two different ethnic group two a single reference group. With the diflogistic command, I would have to lump the two groups together and compare them to the reference group. I don't think I can compare them separately.
Also, how would I go about testing model fit for the diflogistic?. I know some folks also do this in the context of MIMIC models (I guess the distinction for me between MIMIC and LCA in this particular case is semantic).
So in sum, how can I: 1) test ethnicity as a covariate and source of group invariance 2) get model fit statistics for measuring group invariance on individual indicators?
I would appreciate any help you could offer. My data example follows below. Thank you.
input long(stu_id race white hsub lang) double ses long aca double wt long(behave parex save etex mtex) double flun float(select colgpa highdeg f1gpa30)
101101 5 1 4 1 -.25 0 185.478 0 5 . 6 3 33.56 0 0 . 0
101104 8 1 . 1 -.85 0 649.7293 0 7 0 5 5 33.56 1 1 1 0
101105 3 0 . 1 -.8 0 248.753 . . . 3 3 33.56 0 0 0 0
101106 4 . 4 0 -1.41 0 160.605 0 2 0 3 2 33.56 0 0 . 0
101107 4 . 4 0 -1.07 0 220.1699 0 3 0 3 3 33.56 0 . . 0
101108 4 . 4 1 .27 0 172.7619 . 5 . 3 5 33.56 0 0 . 0
101109 8 1 . 1 -.16 1 719.55 0 5 0 5 5 33.56 1 1 1 1
101110 4 . 4 1 -1 0 0 0 5 1 3 2 33.56 . . . 0
101111 3 0 . 1 -1.22 0 224.2212 . 2 . 5 3 33.56 0 0 . 0
101112 3 0 . 1 -.18 0 30.205 0 5 0 6 5 33.56 1 0 1 1
101113 4 . 4 0 -.03 0 150.7823 0 3 1 3 4 33.56 0 1 . 0
101114 3 0 . 1 -.8 0 219.5727 0 6 . 5 3 33.56 0 1 0 0
101117 3 0 . 1 .22 0 0 0 3 0 2 2 33.56 . . . 0
101118 3 0 . 1 .06 0 206.5884 1 5 1 . . 33.56 0 1 . 0
101119 4 . 4 0 -1.35 0 0 . . . 1 3 33.56 . . . 0
101120 8 1 . 1 -.57 0 719.55 0 3 1 2 2 33.56 . . . 0
101122 4 . 4 0 -.63 0 173.3694 0 7 0 3 3 33.56 0 1 0 0
101123 4 . 4 0 -.69 0 183.7529 0 3 1 3 3 33.56 0 0 . 0
101124 5 0 4 1 -.69 0 0 0 6 1 2 2 33.56 . . . 0
101125 5 1 4 0 -1.22 0 0 1 6 1 5 2 33.56 . . . 0
101129 8 1 . 1 .26 1 719.55 1 6 1 6 5 33.56 1 1 1 0
101131 8 1 . 1 .48 1 671.9371 0 5 0 5 5 33.56 1 1 1 0
101132 4 . 4 1 -.16 0 0 0 2 . . 3 33.56 . . . 0
101133 4 . 4 1 -1.16 0 0 0 5 1 3 2 33.56 . . . 0
