Have an idea?

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

missing prices in the MBC design matrix 2.0

Dear Sawtooth-Team,

if you have a chance, could you have a look at my follow-up comment on the post regarding the coding of missing prices in the MBC design matrix?


I was futzing with an exported simulator tool and experienced the same phenomenon that I expected / described in my comment:
With the linear and loglinear specification there is always a point/value (input for simulator) where the transformed coding (model tab) results in a 0 (reflecting that no utility is assigned at that value). When I set that input value to n/a the resulting shares are the same as before.
(btw: What is (or can be) expressed by "n/a" in the simulator?)

Now with regard to the question in my earlier post:
What is the reasoning behind the following?:
The absence (i.e. non-availability) of an item or price level (reflected by the 0 coding of the respective price level in the design matrix) results in the same utility level of the choice option (containing the non-available item) as when the item is available at that particular price level described above.

Thank you,
asked Nov 8, 2012 by alex.wendland Bronze (2,430 points)
retagged Jun 29, 2014 by Walter Williams

1 Answer

+1 vote
Best answer
In choice-based conjoint, if alternative-specific designs are coded, then an attribute may be not applicable or is indeed "blank" for certain choice alternatives.  The common way to code these is to leave them as zero in the design matrix when they are not applicable to the alternative (row).  Indeed, this means the result is the same as if the attribute is "on" and at the average price point.  But, all the estimation accounts for this in the ASC utility for the different alternatives.  This is a common point to trip on for people starting to think about alternative-specific designs, but it just works.

Now, it would not be appropriate to say an attribute suddenly disappears for respondents in the simulator unless you had also made that possible in the experimental design and in the questionnaire for the respondents.  Thus, it is strange to think about "turning off" (making not applicable) an attribute when we had never provided that contingency within the questionnaire.  So, that wouldn't be appropriate.

Here's another instance in which MBC can appropriate use the n/a specification in the simulator.  I had a dataset where 4 items on the menu were only available in half the choice tasks.  There was an availability variable (an independent variable) that signaled whether the items were on the menu or not.  In the simulator, when the availability is specified as "off" then I still need to type something into the price places for the missing items on the menu (the software requires some sort of entry).  The n/a entry is the appropriate thing to type, because these products are just not available.
answered Nov 8, 2012 by Bryan Orme Platinum Sawtooth Software, Inc. (180,515 points)
selected Nov 12, 2012 by alex.wendland
Thanks so much again for taking the time to explain!