Hello, in the latent profile analysis, as is well known, a sample is divided into subgroups. I would like to see how high the probability is for an individual case from diagnostic practice to belong to one of the subgroups. For example, in a given study, 5 subgroups were found based on e.g. NEO PI-R. Now, I would like to use the NEO PI-R for an individual case and then calculate how well this individual case fits into one of the subgroups - that is, how high the agreement of the individual case is with each subgroup. How should I best calculate that? I was thinking about calculating z-Values and then compare them or even Cohen's or Fleiss Kappa to see the agreement eventhough my question is not about reliability.
Greetings
Dear Maximilian Fritz Schwarz
Have a look:
1. http://www.statmodel.com/discussion/messages/13/155.html?1141671585
2. https://journals.sagepub.com/doi/full/10.1177/0095798420930932
3. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6913642/
Kind Regards
Qamar Ul Islam
.
in latent profile profile analysis, you have fitted a mixture model to your data
it seems you want to estimate the probability that a new data "belongs" to a specific component of the mixture
from bayes rule :
p(k|x) = p(x|k)*p(k)/p(x)
with p(k|x) the probability of x belonging to the component k, p(x|k) the distribution of the component k (a simple gaussian if you're using mixture of gaussians), p(k) the probability of the component k (the coefficient of compoennt k in the mixture) and p(x) the probability of data x, which can be ignored when it comes to find the component with the highest p(k|x)
.