Have an idea?

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

ACBC price utility turns positive when I constrain other attributes

Hi all,

I am trying to calculate utilities for a ACBC conjoint.

It has 4 levels plus price. Each of the levels has a 'not included' option (at £0.00), and I want to constrain the utilities so that 'not-included' has a lower utility than the other levels.

I am running piecewise price utility with 5 levels, constrained negative.

When i try to calculate utilities without constraints, it takes a fair few iterations to calculate but eventually converges quite neatly.

Running the utilities with constraints appears to be fine for 2 of my 5 user segments.

However, when I try to run 3 out of 5 of my user segments with the constraints, for the first few iterations it appears to work fine, but then all of the price level utilities (including at the top of the price range) suddenly jump up and end up after 5k iterations plus to be heavily positive- much more positive than the other attribute levels.

Obviously this shouldn't be the case, so I was wondering if anyone knows anything that might be causing it, or anything that I could do to prevent it.

Using v. 6.6.18

asked Mar 9, 2014 by JamesWood (120 points)
retagged Mar 9, 2014 by Walter Williams

1 Answer

0 votes
I don't know exactly what is going on, but I'd check the following:

1) Make sure with your Summed Pricing options that you gave enough random shock to the final prices shown that price sensitivity can indeed be separated with precision from preferences for other levels.

2) Rethink your constraints for your non-price attributes to make sure that they are logically correct (that all respondents would agree with them).

3) Check your RLH (or Pct. Cert) fit for constrained models vs. unconstrained models.  The constrained models should slightly reduce the fit.  But, if you see a really big drop in fit (say from Pct. Cert 0.60 to Pct. Cert 0.50), then something is probably going wrong with your constraints.

4) Check predictions of constrained vs. unconstrained utilities of holdout choice tasks (if you have them).
answered Mar 10, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (136,265 points)