Have an idea?

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

My HB graph does not converge after 50k iterations, should I keep increasing iterations?

Hi,

at the moment I am analyzing ACBC respondent data using HB. The manual states that "the important point is that there be no noticeable trend by the time convergence is assumed". The amount of iterations is already increased to 50k now. The lower values in the Parameter Estimates graph are still not in a repetitive stable "wobbling" state. Should I keep increasing iterations untill there is no noticeable trend anymore or is there a more fundamental problem going on?

Thanks in advance for your replies!

Best regards,

Rui
asked Apr 2, 2013 by Rui (120 points)
retagged Apr 2, 2013 by Walter Williams

1 Answer

0 votes
Are you doing piece-wise summed price?  If so, we recommend constraining price to be negative.  Also, with standard summed pricing in ACBC, we generally recommend to constrain linear or log-linear estimation to be negative for the Price slope.  This aids considerably in proper convergence.

If it isn't the issue with constraining price, then I'm perplexed, because ACBC usually converges very nicely.  Could mean you should have one of our tech support consultants look more closely at your data.
answered Apr 2, 2013 by Bryan Orme Platinum Sawtooth Software, Inc. (138,915 points)
Thank you for your fast reply. I am sorry but in my language your sentence "we recommend constraining price to be negative" can be interpreted either way. So you do mean that price should not be negative?

Thanks you,
Rui
...