Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

What about test persons, who are not clearly defined in one Latent Class?

Hi there,

for my master thesis I do a research with CBC - Latent Class. In my first step I chose a 4-segment-solution. As I looked through the tables I noticed, that some probands where not clearly defined in one cluster, e.g:

Prob_Membership 1/4    Prob_Membership  2/4    Prob_Membership 3/4    Prob_Membership4/4
0,00001                             0,42120                             0,44858                            0,13021
0,00012                             0,48381                             0,01536                            0,50071
0,00665                             0,51897                             0,00120                            0,47317
0,50288                             0,49712                             0,00000                            0,00000

How do I have to handle those? Simply delete? Just keep them? With a deletion I will get better results.
If I will delete them which one should I delete (percentage difference less than 5% or 10%)?

Thanks for your support!
Best regards
asked Apr 12, 2013 by anonymous

1 Answer

+1 vote
The following response doesn't  take into account the goal of your work, but it applies to most cases:

Your situation is absolutely normal. A LC segmentation with a perfectly separated clusters of respondents rarely happen. And degree of separation is also an information about your data structure. So I wouldn't exclude anybody and worry unless the degree of uncertainty of segment membership is very large. Unfortunately I don't know any guides for "very large", I usually judge it by fell - overal quality of the model. But if it had been too large it would lead to one segment solution probably.
Anyway If you leave those uncertain cases in your data you must remember that segment description should take this uncertainty into account.
answered Apr 15, 2013 by lkomenda Bronze (2,830 points)