Have an idea?

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

HB analyses dual response none option

Dear,

is there anything you can do with the response of the dual response none option regarding analyses? For example, can you observe the relative importance of attributes in the CBC task for choosing yes?
If yes, how would we do this best?

Many thanks!
asked Aug 3, 2016 by ariane (220 points)
retagged Aug 3, 2016 by Walter Williams

1 Answer

0 votes
The responses of "yes" or "no" to the dual-response None question in CBC are already automatically being used and combined with the other CBC questions to refine and estimate the utilities for levels--and those utilities are what Importances for attributes are calculated on.  The more a respondent is being influenced to choose "yes" based on her preferred level being present for a given attribute, the more important that attribute becomes.
answered Aug 3, 2016 by Bryan Orme Platinum Sawtooth Software, Inc. (132,290 points)
thank you bryan for your response. So it is not possible to first examine the relative importance and utilities for choosing any option, and secondly, determine which factors are most influencing the actual decision to "buy the product"?
Ah, I see you are assuming that the propensity to choose (and the part-worths associated with choice) is not the same as the propensity to buy.

It would seem at first glance possible to use only the buy/no buy answers to build a different conjoint model (a binary logit)...until you realize that the experimental design for such a model would be crummy.  Respondents are only asked the buy/no buy question for the concepts they chose.  So, the design would be dominated by mostly good levels and there is potential endogeneity bias as well.
Hi Bryan, we have already conducted data collection, we are just trying to figure out how we could analyse these data regarding the none option. So what i have done now is extracted the information regarding the different factors present in each choice task for each respondent (via the interaction research tool). Then merged this information with the none option (yes/no). Now i would do a logistic regression in another software package like spss and observe which factor has most influence on not buying the product.
do you think that is a suitable option? Or are there similar analyses we could do, for example within sawtooth itselve? Maybe i can merge it with information like descriptive info such as gender etc collected in the questionnaire, and observe whether intention to buy is maybe more influenced by gender than factors included in the choice task? What do you think?
Just using the HB utilities that our software automatically estimates and tabulating the None utility by different respondent characteristics I think would be a solid approach.  That avoids the issues I brought up.  Remember that the None weight is estimated based only on the buy/no buy questions.
...