Have an idea?

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

Respondent Weight in CBC/HB Estimation

Dear All,
is it possible to use a respondent weight for estimating utilities with CBC/HB?

Many Thanks
asked Oct 30, 2011 by anonymous

1 Answer

+3 votes
Best answer
HB does not utilize respondent weighting.  To date there has not been an effective means found of weighting respondents in HB.

Weights can be applied in simulations after estimating utilities.
answered Oct 30, 2011 by Walter Williams Gold Sawtooth Software, Inc. (22,905 points)
Thanks for your prompt reply. Reason why I am asking is, that I would like to weight the different versions (resp. respondents who received different versions) of a design so that after the survey all versions are equally often represented in the sample.

Does such an approach make sense and if so are there any other approaches how to handle this issue? Maybe it would make sense to give the user the functionality to include a respondent weight in future versions or is it rather a technical/statistical problem to include a weight?

This makes little sense.  You can have really quite substantial proportions filling out different CBC (or MaxDiff) versions without any harm to your utilities.  In my opinion, don't worry about this.
If the attributes are the same for each version, there is no reason to weight the respondents by the version they saw.  The nature of the HB algorithm makes it very robust given different designs.

The question of including a respondent weight is a statistical validity question.  John Howell did a paper regarding this issue in the 2007 Sawtooth Software conference and found there was no benefit to adding weights.

How to use weights during counts analysis?