Have an idea?

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

Chip allocation design and analysis in SSI Web 8.3.10

Hello,

I'm considering conducting a CBC conjoint using a chip allocation response format. I've completed CBC conjoints using the standard discrete choice response format but have never used chip allocation. I'm not clear on the exact steps to follow with chip allocation so would like to know:
(1) Is there a technical paper or other resource explaining design and analysis procedures in detail step-by-step?
(2) Are my design and analysis steps below correct? If not, can you please provide correct steps?

I will be using an outside vendor to program and host the conjoint - so will need to use the paper-and-pencil data accumulation approach to prepare data for analysis.

Any assistance you can provide would be greatly appreciated. Thank you!


Conjoint design info:
-5 attributes, each with 5 levels.
-8 products will be shown per screen and would also like to have a "choose/buy None" option.
-Would like to NOT have allocation responses sum to a constant value (in other words, respondents can say they would purchase 1 or up to 20 total units per conjoint screen).
-10 conjoint questions/screens will be completed by each respondent.
-Full profile design with no prohibitions will be used.

 
Design procedure in SSI Web:
-Set up study in SSI web as normally would (including the Test design procedure/rules of thumb) for CBC BUT do the following:
1. On "Response type" tab, select constant sum/chip allocation response type and under "Additional Options" enter 20 for Total, and under "Additional Options- Advanced" indicate Amount below the total value=19 and Amount above the total value=0.
2. On "Design" tab, select "Traditional-None" option in order to have analysis estimate a None utility in HB estimation.   


Analysis procedure in SSI Web:

1.Create SSI Web's automated accumulated data template file as normally would for CBC and do the following to prepare data for analysis:
-Each conjoint question/screen will be coded as 9 columns in the accumulated data template file (for the 8 products per screen/question plus the "choose None" option) and I need to enter number of units purchased for the 8 products per screen (which will be a value from 0 to max 20 for each of the 8 products - with sum of max 20 across all 8 products per question/screen). How do I code response data if a respondent selected "choose None" (for example, do I place a 9 in the 9th column and zeroes in columns 1-8 for the product)?

2. Estimate CBC HB utilities as I normally would using SSI Web HB analysis defaults. In other words, no additional modifications are needed in order to accommodate chip allocation.

3. Use resulting raw utilities, rescaled utilities/zero centered diffs and attribute importance scores as normally would for CBC analysis and simulation (I am planing to use the share of preference approach for simulations within an Excel simulator and would also like to have the option of including or excluding the "None" in simulation as with discrete choice CBC simulation). In other words, I can use utility and related results SSI Web produces with no additional modifications needed in simulation or otherwise in order to accommodate the chip allocation.
asked Jul 14, 2015 by anonymous

1 Answer

+1 vote
What you're trying to do is something called "volumetric CBC," not "chip allocation".  In chip allocation, we usually think of having a constant number of chips to allocate in each choice task.

But, it would be strange to ask respondents to specify how much volume they'd purchase across multiple alternatives and then also to include a None option.  Usually, you do not include a "None" option for the design of the questionnaire and for the fielding of the questions.

There are multiple ways to analyze the results, the more sophisticated methods are described by Tom Eagle in a 2010 Sawtooth Software Conference Proceedings article.

A relatively simple approach that can work quite well for straightforward CBC datasets (assuming respondents can actually do volumetric CBC questionnaires well, which is totally up for debate) is to field a CBC questionnaire allowing respondents to type values for each alternative.  Do not include a None in the fielding of the questionnaire.

But, on the back end, you modify the CBC design to "trick" CBC software into thinking that a None concept was included in each choice task.  For each respondent, scan across the tasks to find the largest total volume "purchased" of product within that task.  Let's say for your study respondent #1 bought 20 products in one of the choice tasks (and that was equal to their max purchased across all choice tasks).  For that choice task, the None alternative gets "0" chips.  But, for another choice task for Resp#1 where the respondent only "purchased" 10 units, a full 10 chips is assigned to the None.  You can export the .CSV file including the Raw CBC data and make modifications to it to accomplish this.

Then, proceed with HB estimation as usual using our software (running CBC/HB standalone software on the modified .CSV file).  But, in the market simulator, you must weight each respondent by the maximum number of purchases the respondent made across their choice tasks.  For example, above, Resp#1 would be weighted by "20".
answered Jul 14, 2015 by Bryan Orme Platinum Sawtooth Software, Inc. (138,915 points)
...