Have an idea?

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

Detect outliers for Latent Class Analysis

I want to conduct a Latent Class analysis for CBC data to identify reasonable segments. Before that I want to detect potential outliers. I already deleted respondents with a low response time, incomplete surveys and straight liners.
Are there others ways to detect outliers with Sawtooth that may influence the results of the Latent Class analysis negatively?
Is it possible to conduct an HB Analysis before Latent Class and look for low RLH values (< 0.33) and delete those respondents? Or is it not reasonable to combine these methods?

And another question:
From 248 respondents, 23 respondents always click on the No choice option (There are 3 choice sets and 1 no choice option in one task.) Latent Class analysis provides one segment with these 23 respondents. Does this influence the results of the other segments? Is it better to exclude the 23 respondents from the latent class analysis and evaluate them (with regard to demographics and personal characteristics) separately?

Thanks for your help.
Best regards
asked Mar 28, 2017 by Sophia

1 Answer

0 votes

First, those 23 respondents who only choose None are like straightliners, aren't they?  I might well kick them out of the latent class model:  they have essentially told you that they are not in whatever market you are trying to model.  

I would hesitate to exclude low RLH respondents unless you have some additional criterion on which to exclude them:  low RLH doesn't necessarily mean "bad," it just means "inconsistent."  It is entirely possible for consumers to be inconsistent in the marketplace, too, so I tend to keep them in my models unless I have some additional reason to kick them out.
answered Mar 28, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (62,700 points)
Dear Keith,
thank you very much for your fast response and support!
Best, Sophia