Hi!
I would like to do a latent class analysis with ranking data (in MPLUS).
Adolescents ranked 5 areas of life according to their personal importance from 1-5 (e.g. 1 school, 2 peers, ...).
My problem is: using a LCA with categorial variables does not include the info that the rankings are interdependent (if school =1 is ranked nothing else can be ranked as 1). How can I take this dependency into account or is this not a problem at all?
Many thanks for hints!
Lara
Hello Lara,
Yes, forced choice ranking makes for problems...the respective rank values will be negatively correlated. The preferred way would have been to allow respondents to rate each of the five areas on some common scale (which could be as simple as "important to me" vs. "not important to me", or a 1-whatever rating scale).
Your options, given the way data were collected would be:
1. Recast the rankings as "low" or "high" (perhaps ranks 4-5 would be low, and ranks 1-3 would be high) for a given area. This will reduce, but not eliminate, the degree of negative dependencies in the data.
2. Consider a mixed model, in which rankings are nested within respondents, and both are random factors. I don't know enough about Mplus to say whether this would be viable or possible.
3. Ask the nice folks at Mplus...their web pages and previous posts would suggest that they do a good job of supporting users who have tricky questions: https://www.statmodel.com/support/index.shtml
Good luck with your work.
Hi David,
thank you so much for your advice! That helped me a lot. I think I will go with your second suggestion and use a mixed model.
Kind regards
Lara
- Badges
- Science topic