Chapter 4 Checklist for Measurement Invariance with MGCFA
In this section we describe the necessary steps for invariance testing.
4.1 Start by having a model
Any statistical model is only as good as the theory it is built on.
Run CFA in each group to detect any large deviations.
4.2 Test the configural model
Run a MGCFA without cross group equality constraints
The model should show a good fit.
There are two approaches of model identification: marker indicator or reference group (see Little et al., 2006)
4.3 Test the metric model
Fix the factor loadings to be equal across groups
Compare the model fit to the configural model
4.4 Test the scalar model
In addition to the factor loadings, constrained the intercepts to be equal across groups
Compare the model fit to the metric model
4.5 Got noninvariance? It happens often.
Several options:
- exclude groups
- remove items
- distinguish several subgroups of countries
Still not getting invariance?
Conclude that the construct has different meaning across groups
Alternatively, consider the decision tree presented in Figure 3.7.