Have an idea?

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

ACBC - HB regression using piecewise price attribute coding

Hi everyone !

I'm doing an ACBC study with 45 respondents,

Yesterday, with 42 respondents, I used an HB regression and a piecewise method for the price attribute with 20 price points.
Today, just by adding 3 more respondents to the initial 42 respondents and by changing the price points for the HB piecewise regression, the importance of price decreased from 50% down to 35% (the minimum price point and the maximum price point tested did not change).

Are there any reasons explaining this quite strong decrease of the price attribute apart from the low number of respondents (hence the impact of 3 more respondents, knowing that these 3 new respondents were quite price sensitive in addition)?

Thank you very much for your help

asked May 30, 2018 by luclepage2017 (225 points)

1 Answer

0 votes
That's unusual.  To get a sense for the amount of variation in the results just due to random starting seeds, I'd run your HB four or five times (should be fast, given your small sample size), using seeds of 1, 2, 3, etc. (you can change the starting seed on the Estimation Settings link within the ACBC HB settings area.  If you're seeing quite a bit of variation (in price importance) when only the seed is changing (everything else is held constant), then perhaps this is due to lack of convergence.  Remedy would be to triple or quadruple the number of burn-in and used iterations.
answered May 30, 2018 by Bryan Orme Platinum Sawtooth Software, Inc. (201,565 points)
Thank you very much for your answer Bryan!

Sorry I forgot to say that by "changing the price points" I meant "adding 10 more price points" without changing the minimum and the maximum, but I will try to follow your advice.