CBC Chip Allocation

Question

Hi,

I am setting up a chip allocation cbc, where the respondent will have 10 points to spend on a number of different options. I was thinking of setting up the conjoint model with 2 attributes (Option and Point Cost), then the respondent would see something like the below, where they will select options in order to total a spend of 10.

For example: You have 10 chips to spend on the following options, how do you spend them:

Option A - 3 points
Option B - 2 points
Option C - 1 point
Option D - 5 points
Option E - 2 points
Option F - 4 points
Option G - 2 points
Option H - 1 point

i.e, Respondent 1 chooses: Option D, F and H

Can I fix the conjoint to show the options above, so that the options have a fixed cost for the respondent to assign chips too?

I think the ideal solution for me is to use Menu Based Conjoint, but we do not have the software and budget constraints might stop this being an option so I am trying to find a workaround solution to this.

Any advice would be very appreciated!

Thanks,
Dave

Comments

What version of SSI Web / Lighthouse Studio are you running?

Answer 1

Dave,

This is like a constrained menu-based choice, which you could indeed just program up in Lighthouse Studio as a free-format CBC question (allowing multi-check response, and with some fancy Javascript verification on each question, to force the respondent to spend exactly 10 dollars or no more than 10 dollars on the alternatives).

Then, you could conduct the modeling and analysis using our CBC/HB software (which you have a license to, if you have a license to our CBC system within Lighthouse Studio).

So, you have two attributes: brand & price.  You would allow multi-checks in the CBC task.  And, you would need to reformat the data in a special way for CBC/HB analysis (not hard, as you can do this in Excel and create .CSV file to submit to CBC/HB).  If you constrained the respondent to spend exactly 10 dollars, then you'd just model it up as brand effects plus price effects.  But, if you allowed respondents to spend less than 10 dollars, then you'd use the "synthetic None" approach of assigning any slack in demand to the None alternative.  The respondent wouldn't see the None alternative in the questionnaire...but we'd model the data with a None option as if the respondent had seen it.

Comments

Thanks for this, makes sense!

In terms of the end simulator, I am assuming it will look and behave in the same way as a 'standard' discrete choice conjoint simulator with the ability to toggle options and cost which will produce the Share of Preference based off of utility values.

Something that I may have to determine is working out the share of each package. By this I mean that once I have set the optimum pricing for each option, I will need to work out how people will spend their 10 points. Is this a case of working out the possible 10 point combinations and adding up the individual option SOP's then divide by the total number of combination SOP's?

For example,
Option 1 - 5 points          SOP: 20%
Option 2 - 4 points          SOP: 40%
Option 3 - 1 point            SOP: 15%
Option 4 - 4 points          SOP: 25%

So the two options here would be:
PACKAGE 1: Option 1,2,3 summed SOP: 75%
PACKAGE 2: Option 1,3,4 summed SOP: 60%

Then to work out the uptake of each package:
Package 1 Uptake = 75%/135% = 56%
Package 2 Uptake = 60%/135% = 44%

This seems to make sense in my head, but not sure if it's the correct way to approach this.

Thanks,
DT

---

If your only requirement on the analysis end is to determine the percent of each item that is predicted to be selected across the population, then a standard conjoint simulator as we offer will give it to you.

However, if you need to produce a simulator that predicts the combinatorial choices (the packages) that are selected, this involves something greater.   This could involve using our simulator to report the individual-level results per respondent.  The individual-level results show you the shares of preference that sum to 100% across the options.  You would then need to develop some logical rules that would allow you to predict the most likely choice of package for each individual that doesn't exceed each individual's budget (is constrained to spend no more than 10 points).  Then, you can tabulate across those predictions.  

So, I'm saying that you would need to develop another layer of simulator that might be able to be written in Excel.  In fact, once you knew you had to create the "combinatorial" predictions of the packages, then it would probably be better for you to take the HB utility results into Excel and write your own simulator that used the logit equation to predict what percent of respondents picked each item on the menu; and also referred to the individual-level predictions to tabulate the combinatorial selection of package compositions (subject to spending no more than 10 points).

Indeed, this is doing MBC--menu-based choice.  MBC as a methodology is more complicated than standard CBC.  So, it is not for the feint of heart.  But, if you have enough experience in discrete choice methods, you may be able to figure it all out and make it work well.  However, if you are relatively new to these methods or are concerned about the complication of what you are proposing, you would probably want to hire a consultant with experience in MBC projects.

And, the modeling approach I suggested to you above (building a single model, assuming an allocation approach) is just one of the simpler of multiple approaches you could take to modeling the MBC data.   I encourage you to look at the following white paper that provides more details about models like you are proposing:  http://www.sawtoothsoftware.com/download/techpap/mbcconf2010.pdf

In this white paper, the approach I suggested above is called the "Volumetric CBC Model" and is described on pages 11-14.