Have an idea?

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

Maxdiff segmentation - one group that chooses less preferred items

I have done a Maxdiff and have now run a Latent class segmentation within Sawtooth.

When I start looking at the 2 segments, there is one group that scores high on all items that are chosen most often (when looking at the average scores for 1 segment), and one group that scores "high" on all items that are chosen less often.
I was wondering whether this is biased: do they just segment together because they choose often/not often chosen items; or are the 2 groups distinct because of the content of the items?

The segment that chooses the items that are chosen less often also shows a very low standard deviation across all items  (0.8) compared to the other group (2.5).

This group that chooses the less-important items also keeps existing until I go up to 6 segments.
I guess I'm just wondering if this is a "real" segment based on the content of the items.
asked Oct 3, 2022 by Tina Van Regenmortel Bronze (615 points)

1 Answer

0 votes
Hi Tina,

It might be worth checking the RLH of the respondents within this segment, my guess is that these will be respondents who werent answering the survey with particular care.

Have a read into the below article on a setting within Latent class which could help to smooth of some of the effects of respondents having a high level of error in their responses (and therefore a lower scale factor in scores).


Hope this helps,

answered Oct 3, 2022 by Dean Tindall Bronze Sawtooth Software, Inc. (4,125 points)
Hi Dean,
thanks for your reply.
Indeed, if I look at the fit statistic for this group, these are the lower RLH values (average of 0.4; 50% of the group with a RHL of less than 0.269 - I have 5 items within each task).
Am I correct that ticking off the box 'constrain groups to a common scale' is the setting that helps smooth out some of these effects?
Is this a setting that I should tick off standard? Could it have a negative impact in some way? Or should I just always use it?
That is the correct box to tick yes.

I would run it both ways in order to check that you are happy with what the constraints are doing with the data.  

Using this method should help with some of the offending respondents, but will lead to a slightly lower model fit.
Hi Dean,
unfortunately, I now get the error "Index was outside the bounds of the array."
I have sent the Sawtooth helpdesk an email to hopefully help me further.
This issue was resolved by updating to the latest version of LHS