I am doing a comparative study, investigating the relationship between independent and dependent variables across two samples from different cultures. I have performed multi-group CFA in AMOS. While doing Measurement Invariance analysis, I have found that there is factorial invariance, but the model lacks metric invariance. This means that the groups are not comparable using multigroup CFA?
In this case, what are the options left for doing a comparative study on these samples? Are there alternative statistical techniques that may be used to compare these samples?
You can perform the MIMIC analysis that allows you to use the prediction of categorical variables (groups, age, sex, etc.) to determine the heterogeneity of the model considering these variables. Another alternative is to perform a network analysis and compare both groups to demonstrate the differences according to the correlations between the variables of the model.
Hi Ahmad,
did you not even find partial metric invariance (i.e., having two indicators invariant)? If that is the case, I see three options:
1) You could sell THIS as the main result. Different cognitive conceptualizations across cultures could indeed bye a very interesting result.
2) Beyond the mere statistical difference across groups: If the actual differences in loadings are not substantial, you could continue with your original analysis and discussing the differences as a limitations. No need to throw away studies when things don't run perfectly.
3) There are some approaches to avoid testing altogether and apply Bayesian analysis to metric invariance, for instance,
Cieciuch, J., Davidov, E., Schmidt, P., Algesheimer, R., & Schwartz, S. H. (2014). Comparing results of an exact versus an approximate (bayesian) measurement invariance test: A cross-country illustration with a scale to measure 19 human values. Frontiers in Psychology, 5(AUG), 1-10. doi:10.3389/fpsyg.2014.00982
HTH
Holger
P.S. Cristian Ramos-Vera , I doubt that you can do a MIMIC model after having found (substantial) metric non-invariance, as this would at least require that you measure the same thing across the levels of the categorical predictors....
Dear Holger Steinmetz , thanks for your valuable suggestions. I haven't checked for partial invariance yet. I will try this solution now.
regards
Ahmad
Dear Holger Steinmetz
I have achieved partial invariance for the model. In the total 11 constructs, i have got partial invariance in 9 constructs with 2 or more indicators invariant, but two constructs have only left with one indicator invariant. Is it acceptable as a partial invariant measurement model?
Hi Ahmad,
except for the 2 constructs with noninvariance, yes. See
Byrne, B., Shavelson, R. J., & Muthén, B. (1989). Testing for the equivalence of factor covariance and mean structures: The issue of partial measurement invariance. Psychological Bulletin, 105(3), 456-466.
Best,
Holger
- Badges
- Science topic
- Similar topics
- Indexes