Have an idea?

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

Usefulness of BYO without price attribute in ACBC

Dear community,

In our ACBC study we do not have any price attribute which means that respondents may abuse the BYO configurator and just select the best attribute levels straight away (there is no "penalty" to select high levels). From my understanding, even if BYO is abused as supposed, this should not have a significant impact on the results because BYO just prepares more relevant options for the Screener Section and Choice Tournament. The respondent will still have to make trade-off choices.

To my question: How "bad" is the influence of potential abusing in BYO on the overall results?
asked Jan 27 by danny Bronze (1,260 points)

1 Answer

0 votes
Best answer
When no Summed Price attribute is involved in the BYO, then respondents can just pick their preferred levels within each attribute irrespective of price premiums or discounts that might be associated.  For attributes like brand and color, this makes sense.  For "ordered" attributes where every respondent would agree regarding which level is preferred, we recommend you select the setting that drops those attributes from the BYO section.  Thus, you are only asking the non-ordered attributes in the BYO section.

When an attribute is dropped from the BYO section, then ACBC's design strategy is to sample these levels with equal frequency for the  subsequent "screener" section.   For attributes included in the BYO section, ACBC's design strategy is to oversample the BYO-selected levels in the subsequent "screener" section.
answered Jan 27 by Bryan Orme Platinum Sawtooth Software, Inc. (184,140 points)
selected Feb 13 by danny
Good questions.  

1.  Don't think this is up to date.  BYO without price seems like a similar idea.  But, our BYO uses data augmentation during analysis that makes it look like the respondent saw all the levels of an attribute and preferred their chosen BYO level.  Thus, one shouldn't use language in a non-Price BYO that asks them "given what you know about your budget...", because we want people to make choices in a non-price BYO congruent to their main-effect preferences (preference for each attribute, assuming everything else including price is equal).

2.  If not using the BYO question, the choice of conjoint concepts to evaluate in the Screener section will be different across respondents (level-balanced and nearly orthogonal; but a different "version" of the concepts for each respondent).  In that case, ACBC is acting like other conjoint methods to generate an efficient experimental design for the Screener section, unique to each person (like multiple "versions" or "blocks").  Of course, once respondents indicate "possibility/not possibility" for each concept in the screener, only the "possibility" concepts are carried forward to the Tournament section.  So, at that point it's an adaptive design.

3.  You could try the HB run with no constraints versus using constraints on all the 8 attributes.  If they all have rational preference order, it could make sense to fully constrain the model if especially you were interested in individual-level accuracy or classification.
That is so valuable. Thank you Bryan!

In the beta phase I tested the hypothesis "most respondents will select the highest level". Analyzing the beta data, to my surprise 2-3 of the 8 attributes do not have a strong preference order, contrary to my previous assumption.

Here some example data from the beta study (n<15) to show how often the levels were selected.

Attribute 1: Level 1 = Feature A (47%)
Attribute 1: Level 2 = Feature B (6%)
Attribute 1: Level 3 = Feature A+B (47%)

Attribute 2: Level 1 = Feature A (34%)
Attribute 2: Level 2 = Feature B (13%)
Attribute 2: Level 3 = Feature A+B (53%)

Other attributes have a clear preference order which is why they will be excluded from BYO. Example:
Attribute n: Level 1 = Feature A (7%)
Attribute n: Level 2 = Feature B (13%)
Attribute n: Level 3 = Feature C (80%)

Now, for those 2-3 attribute that do not seem to have a clear preference order... Would it make sense to have a BYO with only 2-3 attributes? Or is that considered too little, given that there are 7-8 attributes in total? Would you recommend to include them, or rather omit BYO at all in this case?
Highly appreciate your thoughts on the inclusion of 2-3 attributes in BYO. Just worrying whether include 2-3 of 7 attributes will add a bias
I think it makes sense to ask about the subset of attributes that are non-ordered in the BYO of an ACBC...even if it means only showing a very few of your attributes in the BYO.  However, if Summed Pricing is included, then you'll want to work on the BYO such that the respondent understands that most of the attributes are not displaying and that they need to think about the pricing tradeoffs among just those attributes shown in the BYO.
Thanks a lot for sharing your thoughts Bryan!! Helps a lot.