Have an idea?

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

Exclude respondents from LCA with low RLH?


I'm running a CBC exercise. I have run the HB model and there is a small fraction of respondents with low RLH values.

Since my exercises had 2 product options plus a none, there is a ~50% chance you could guess right in predicting their behavior if you knew nothing about them. So the target RLH for randomness would be 50%.

I have since moved on to running latent class analysis, but I was concerned before I did that that I would not be able to remove the respondents who had RLH in the 50s or 60s (i.e. the HB model has not done much better than random).

Is there a way to create a filter or set an option (I'm running Lighthouse Studio 9.11) that would exclude respondents with an RLH below a certain cutoff value?

asked Aug 4 by alexrosenberg (180 points)

1 Answer

0 votes
Alex, two things:

First, 50% is the RLH for random data but there's a white paper that gives better advice about setting a cutoff:  https://www.linkedin.com/pulse/identifying-consistency-cutoffs-identify-bad-respondents-orme/?trackingId=g9ylG8GeasHXn79cZcC5JQ%3D%3D

And here's a recent update:  https://sawtoothsoftware.com/resources/technical-papers/diagnostics-for-random-respondents-in-choice-experiments

Second thing:   I usually export my data, then in Excel I create a binary variable for whether a respondent is below or above the RLH cutoff, then I copy that vector of zeros and ones into Lighthouse Studio to use as a filter.  I know there's a way to do this for MaxDiff using the MaxDiffRLH function in the software, but I don't think we have an equivalent CBC function built into Lighthouse Studio.
answered Aug 4 by Keith Chrzan Platinum Sawtooth Software, Inc. (107,050 points)