Have an idea?

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

Utilities of conditional list in ACBC conjoint analysis.

Dear Sawtooth
We have set up a ACBC with 6 attributes. The first attribute (brand) is constructed with a conditional list: The participant first chooses one, two or three attribute levels (brands) out of 8 possible levels. Then the ACBC starts with these three levels (brands). However the output file shows me utilities for each of the 8 attribute levels in each participants individual answer. What are these values and how do I treat them? What are the consequences for the other utilities? The relative importance values are calculated on the base of all the utilities. Hence the relative importance of the first attribute is probably overstated?
I thank you in advance for your help.

Best Christoph
asked Nov 22, 2011 by anonymous

1 Answer

+1 vote
This is one of the very nice things about ACBC.  For the respondent, the brands that the respondent rejected (prior to the beginning of the ACBC survey) are indeed included in the utility estimation.  Additional "synthetic" tasks are added to the data for that respondent, indicating that this respondent compared the rejected brands to the accepted brands, and chose the accepted brands.  This informs utility estimation that this respondent strongly dislikes those other brands.  But, it allows you to have a complete data set with all brands present for each respondent whereby to conduct market simulations.

I'd suggest you don't look at attribute importance scores.  They are not very meaningful, especially in this case.

Preferably, use the market simulator to conduct "sensitivity analysis" on each attribute, so you can see the impact of each attribute on choices (while considering a relevant competitive base case scenario).  The ACBC utilities for missing brands are quite appropriate for including in this type of analysis.

If you allow respondents to drop levels of an attribute as inferior, then indeed traditional "importance" scores coming out of ACBC exercises will have lots of weight thrown to attributes that allowed such exclusions.
answered Nov 22, 2011 by Bryan Orme Platinum Sawtooth Software, Inc. (150,615 points)
When using constructed lists, the researcher can provide instruction on how these synthetic tasks are used to influence the final utility scores.  From our help files:


If an attribute used a constructed list, then you need to specify how HB or Monotone Regression should treat any missing levels. The selections are:
    
 Inferior to Included Levels: Any missing levels are assumed to be inferior to included levels. For HB estimation, we add information to the BYO tasks indicating that the respondent compared excluded levels to included levels, and selected the included levels. For Monotone Regression, we extrapolate the utility of missing values as a certain percentage below the worst level within that attribute, where the percentage is based on the average utility range of the attributes.  
 
  
 
 Unavailable (Zero Probability of Choice): Any missing levels are assumed to be not available to the respondent, and therefore have a very low part-worth utility. If you select this option, you must specify the low part-worth utility (that essentially drives share of preference to zero for a product containing that level). We recommend using -10 or lower. You can tune this constant based on market simulation results.  
 
  
 
 Missing at Random (Imputed from Population Information): (Only available for HB.) Missing levels are imputed based on draws from population means and covariances. We generally suggest avoiding ACBC designs that omit attributes at random.
...