We are attempting to validate or revise a theoretical model. The model groups etiological risk factors that variously lead to problem gambling. We started with 120 items and reduced them through EFA and used CFA to obtain six psychometrically strong factors. We now want to see how these factors break out into groups (that is, what constellation of factors combine to form each subgroup)? Do we use latent class analysis or some other technique to devise these groups? no agreement among the stats folks. Thanks in advance
Hello Lia,
CFA groups variables. Latent Class categorizes individuals into discrete groups based on similarity or proximity scores on the variables. Factors don't break out into groups, people do. It is best to work with the raw variables for Latent Analysis as opposed to aggregated factors from CFA otherwise your interpretation risks being meaningless. Hope this helps. Nice to see you continue to work on the theory building.
Best
Neda Faregh
Dear Dr. Nower:
One option would be to use factor mixture modelling. You could use the factor model that you have developed already and then derive latent classes based on the factor scores.
Lubke, G. H., & Muthén, B. (2005). Investigating population heterogeneity with factor mixture models. Psychological methods, 10(1), 21.
Hope this Helps,
Paul
Hello Lia,
Whatever you do, you need strong rationale for that, as mistakes in methodology could be very costly. Factors pertain to the items (e.g. questions in surveys) and latent classes assign participants to the classes (bear in mind that there may only one class but the homogeneity like this is not very likely) and this assignment is based on probabilities. Mplus software could be a good tool to apply in my opinion.
Good luck with your research
- Badges
- Science topic
- Similar topics
- Correction