Have an idea?

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

Occasion Based conjoint


Just a general question to ask whether it is possible to conduct 'occasion based' conjoints and what the best practices are for that. Clearly, we could make respondents go through multiple exercises about different occasions but this could get cumbersome/would need lots of sample if there are many occasions. I guess we can also think of volumetric CBC as another way around it because asking about someone's next X purchases should hopefully capture different occasions? Just wondering whether there are any neat tricks that people are doing to get at occasion based conjoints
asked Nov 7, 2013 by chris
I'm planning a CBC study for potato chips market and don't know how is the best way to phrase the question. It can simply be "If these were your only options, which would you choose?" or "If these were your only options, which would you choose for consuming at home when watching TV?", where "consuming at home when watching TV" is an occasion customized to each respondent. In the end, the utilities will be computed by total, not by each occasion. Do you think this type of question (occasion-based) is more relevant? Perhaps putting the respondent in more contextual landscape it will generate more relevant answers.
Many thanks,

1 Answer

+1 vote
That's a very good question.  Clear back in the early 1990s, our founder, Rich Johnson, very much liked and advocated the idea of using different "prologues" at the top of the CBC question to get people thinking about different situations/occasions.   

Asking people to allocate "chips/percentages" across the different products maybe gets them to think in aggregate across multiple purchase situations, but there is quite a lot of opinion mounting that this is problematic for respondents to navigate mentally.

Models can be built separately to account for different occasions (and estimate separate parameters for each occasion).  This could be done by estimating models separately for the tasks involving each occasion.  Or, occasion-based covariates could be introduced via HB (with each respondent broken into multiple records that collect the common occasions under each respondent replicate).  You could weight each of the replicated respondent records by the volume purchased by that respondent (usually) for the different occasions (asked as a separate question outside the CBC).  You could also deal with occasions via interaction effects if using some sort of pooled analysis (such as LC or aggregate logit).

Occasions could be customized per respondent and the prologue at the top of the task could be modified.  Or, if using Free Format CBC, you could specify a multi-row grid beneath each choice task.  In the first row, the respondent answers the question with respect to his/her first occasion, the second row is for answering the choice task with respect to the second occasion, etc.   You could then build separate models for each of the occasions, or introduce as covariates or interaction terms as mentioned above.

To be honest, I've never conducted research into these different methods of obtaining occasion-based results.  

Jon Pinnell makes a brief mention about occasion-based approaches for CBC in his paper: https://www.sawtoothsoftware.com/downloadPDF.php?file=pinnell.pdf

Right now, I cannot remember any other papers presented at our conference that go into this.  Maybe somebody else can.
answered Nov 8, 2013 by Bryan Orme Platinum Sawtooth Software, Inc. (134,015 points)