Shig ONO posted on Friday, September 28, 2007 - 5:33 am
Hello, I'm afraid this may be a naive question...
I frequently do cluster analysis or LCA for marketing segmentation. To my understanding, in k-means cluster analysis we can look at RMS standard deviation of a cluster to see how the cluster is "cohesive", i.e. how members of the cluster are similar to each other.
In LCA, do we have any easy way to check the "cohesion" of a class? Or is it a nonsense idea in the context of LCA?
I'd like to hear your suggestion in particular for LCA with binary indicators.
It seems like class cohesion would be relevant in LCA as well. I am not familiar with literature on this - is anybody else? We have information on how well the classes are separated, but that's another matter. For LCA with binary indicators, I guess one can see how similar people classified into a class are with respect to their observed 0's and 1's. Perhaps doing a factor analysis within class to see if there is any factor variance.
Shig ONO posted on Saturday, September 29, 2007 - 2:45 am
Thank you for your comment! Let me confirm, at last line you mean one-factor CFA for members of a class, fixing all factor loadings as one?