|
|
Identifying "the class-defining u ind... |
|
Message/Author |
|
|
I am using cross-sectional LCA to look at dietary patterns based on 24 food groups (categorical indicators with 3 thresholds). Clearly it would be advantageous to remove food groups from the analysis if they aren't acually helping us to discriminate classes. In another thread you mention "reducing down to the class-defining u indicators" in the ASB example on the website, but I can't sort out how you actually do this, even after looking at the course handout for LCA/ASB example. http://www.statmodel.com/discussion/messages/13/92.html?1256660822 Can you explain the process you use to remove useless indicators (or point me to where this information is)? Many thanks, DL |
|
|
An item that is not helpful in defining classes is an item that has similar conditional item probabilities (similar logit estimates) for all the classes, for instance with 3 classes: 0.33 0.33 0.33 Instead you want something like: 0.7 0.1 0.2 where the item helps to discriminate between class 1 (first column) and the other two classes. That item is a class-defining u indicator. |
|
Back to top |
|
|