Dear Statalisters,
In his (excelent!) book Discovering Structural Equation Modelling Using Stata, Alan Acock teaches us how to test for measurement invariance in Stata.
Using data from the National Longitudinal Survey of Youth (1997), three items about depression are linked to one latent factor and four items about government responsibility are linked to a second latent factor. In Figure 4.2 of Section 5.3 he shows us the solution when women and men are combined into one group using both the "group(female)" and the "ginvariant(all)" options:
The latter command gives us the fit indices.
Comparing the SEM model to the situation when ignoring the grouping variable, using
gives us exactly the same model (= as expected: hurray!!), but - except for the Coefficient of determination - very different fit indices (= very unexpected: huh?!).
Why?
Kind regards, Adriaan Hoogendoorn
In his (excelent!) book Discovering Structural Equation Modelling Using Stata, Alan Acock teaches us how to test for measurement invariance in Stata.
Using data from the National Longitudinal Survey of Youth (1997), three items about depression are linked to one latent factor and four items about government responsibility are linked to a second latent factor. In Figure 4.2 of Section 5.3 he shows us the solution when women and men are combined into one group using both the "group(female)" and the "ginvariant(all)" options:
Code:
net from http://www.stata-press.com/data/dsemus/
net get dsemus
use multgrp_cfa, clear
sem (Depress -> x1 x2 x3) ///
(Gov_Resp -> x4 x9 x10 x12), ///
standardized group(female) ginvariant(all)
estat gof, stats(all)
Comparing the SEM model to the situation when ignoring the grouping variable, using
Code:
sem (Depress -> x1 x2 x3) ///
(Gov_Resp -> x4 x9 x10 x12), ///
standardized
estat gof, stats(all)
Why?
Kind regards, Adriaan Hoogendoorn
