We assessed different risk behaviors and consequences associated with alcohol use. We asked the participants if they were involved in a given behavior (e.g: driving under the inffluence) the previous month/year. So our data consite of many yes/no answers. Now I want to explore if some people tend to engage in a constellation of risk behaviors. What do you think f the best multivariate statistica technique for this question.
If you have a bunch of yes/no indicators I would look at latent class analysis. There are several R packages for this (e.g., PoLCA and LCCA).
If, on the other hand, your yes/no variables are dependent variables then I would use multinomial logistic regression. This is the case in most medical risk factor studies. An example from the 2016 Joint Statistical Meeting Proceedings is given below. Best wishes, David
Latent class analysis is a suitable option for nominal variables. LCA assumes that the covariation between manifest indicators arises by virtue of their association with underlying classes. LCA facilitates the extraction of distinct and meaningful subgroups based on the unobserved heterogeneity within a population and on the similarity between their response profiles. So, it is an adequate option if you are trying to find patterns or profiles of behaviors under the influence of alcohol
Thanks, to all of you. Can you recommend me a good introductory reference for LCA?
Uebersax, J. (1994). Latent class analysis of substance use patterns. In L. M. Collins, & L. A. Seitz (Eds.), Advances in data analysis for prevention intervention
research. NIDA research monograph, No. 142. Rockville, MD: National Institute on Drug Abuse.
A couple of my colleagues wrote a gentle LCA introduction. Worth look before the more technical resources out there.
Chapter Latent Class Analysis and Latent Profile Analysis
Thank you, Thom. It seems that a drat version of the chapter is available online. I´ll check it.
- Badges
- Science topic