latent class

#1
For a market research study I am using latent class. I want to find out what kind of things are important to different people. I will use regression.
The problem I experience is that the betas I get (in Latent Gold) are nicely explaining differences accross clusters, but do not reflect relative importance within a cluster. This is because the variables have different scale types. How do I get around this problem? I want to end up with say the three most imoportant measures for each cluster.
Another point is class size. These are based on probabilities. In my case the modal clusters are quite different from the probabilistic clusters. In one cluster the number of respondents is only 45 (from 500), but the class size is 22%!!! How do I interpret class size in this case?
Hope someone can help me on these two questions...

Ivo
 

JohnM

TS Contributor
#2
ivoeman said:
The problem I experience is that the betas I get (in Latent Gold) are nicely explaining differences accross clusters, but do not reflect relative importance within a cluster. This is because the variables have different scale types. How do I get around this problem? I want to end up with say the three most imoportant measures for each cluster.
You'll need to normalize each variable into z scores, based on the normal distribution:

z = (x-mu)/sigma

mu = mean
sigma = standard deviation