I am conducting an LPA and need to clarify whether the variables must be normally distributed before modeling.
In LPA, it is assumed that the variables are normally distributed within each class (cluster/profile), not overall (not before modeling). So you're fine to start out with non-normally distributed variables in the overall population.
In Latent Profile Analysis (LPA), your variables do not need to be normally distributed beforehand. The model assumes normality within each latent class, not in the raw data.
Key Points:
• Mild non-normality is okay.
• Standardize variables if on different scales.
• Check for outliers/extreme skewness—they can affect model fit.
• Consider transformations or robust methods if non-normality is severe.
You can use R packages like tidyLPA, mclust, or lavaan for LPA.
Thank you Maha Fouad Ahmad and Christian Geiser for your comments and feedback.
I had already done the analysis with tidyLPA and mclust and wanted to be sure.
- Badges
- Science topic
- Similar topics
- Social Science
- Media
- Journalism