Have an idea?

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

CBC: How would one keep a single attribute constant for each design in a given task?

Hello,

I'm looking to set up a CBC wherein each concept in a given task has one attribute constant (but that attribute can vary between tasks).

Example:

Shirts can be $10, $15, or $20
Shirts can be Green, Orange, or Red (G, O, R)
Shirts can be Buttondown, Collared, or V-neck (B, C, V)

How do I make it to where each task shows 3 random shirt concepts that all have the same price, like this:

Task 1:
$10, G, V
$10, G, C
$10, O, C

NOT

Task 1:
$10, G, V
$15, G, C
$10, O, C

Thanks!

MT
asked Feb 5, 2014 by mjt127 (120 points)

1 Answer

0 votes
This would require some custom setup and manipulation, but indeed may be done within our CBC software (Advanced Design Module).

I would think of the problem in a new way (for setup in our CBC software for generating the experimental design).  For the purposes of experimental design, you would think of each task (which contains 3 concepts) as if it were a single concept.  That design would have the following attributes:

Attribute 1 (Concept 1's Color):
  Level 1: Green
  Level 2: Orange
  Level 3: Red

Attribute 2 (Concept 1's Style):
  Level 1:  Button-down
  Level 2:  Collared
  Level 3:  V-Neck

Attribute 3 (Concept 2's Color):
  Level 1: Green
  Level 2: Orange
  Level 3: Red

Attribute 4 (Concept 2's Style):
  Level 1:  Button-down
  Level 2:  Collared
  Level 3:  V-Neck

Attribute 5 (Concept 3's Color):
  Level 1: Green
  Level 2: Orange
  Level 3: Red

Attribute 6 (Concept 3's Style):
  Level 1:  Button-down
  Level 2:  Collared
  Level 3:  V-Neck

Attribute 7 (Flag for if price is constant for the task):
  Level 1: True
  Level 2: False

Attribute 8 (Price for all three concepts, if Attribute 7 is True)
  Level 1: $10
  Level 2: $20
  Level 3: $30

Attribute 9 (Price for Concept 1, if Attribute 7 is False)
  Level 1: $10
  Level 2: $20
  Level 3: $30

Attribute 10 (Price for Concept 2, if Attribute 7 is False)
  Level 1: $10
  Level 2: $20
  Level 3: $30

Attribute 11 (Price for Concept 3, if Attribute 7 is False)
  Level 1: $10
  Level 2: $20
  Level 3: $30

You set up the design as an alternative-specific design, where attributes 8-11 depend on the outcomes of attribute 7.  

Also, you prohibit certain 3-way attribute combinations from occurring:

Green + Green + Green for attributes 1, 3, 5 (a 3-way prohibition)
Orange + Orange + Orange for attributes 1, 3, 5 (a 3-way prohibition)
Red + Red + Red for attributes 1, 3, 5 (a 3-way prohibition)
BD + BD + BD for attributes 2, 4, 6 (a 3-way prohibition)
C + C + C for attributes 2, 4, 6 (a 3-way prohibition)
V + V + V for attributes 2, 4, 6 (a 3-way prohibition)
$10 + $10 + $10 for attributes 9, 10, 11 (a 3-way prohibition)
$15 + $15 + $15 for attributes 9, 10, 11 (a 3-way prohibition)
$20 + $20 + $20 for attributes 9, 10, 11 (a 3-way prohibition)

This gives you an experimental design (for as many versions of the questionnaire as you desire).  You can export that design to a .csv file that may be opened in Excel.  You manipulate the design in that file to make it look like there are actually 3 concepts per task rather than 1.  Then, you collapse the columns as well so that there are just 3 total attributes.   Now, you've got something that looks like a more standard design for CBC software.

Last, create a different CBC project using our software, with just 3 attributes, 3 concepts per task (as you had originally thought of the problem).  Ask the software to generate a regular experimental design, for as many versions as you desire.  You're going to replace this design with the one you composed in the previous step.  Run Test Design on this standard design to get a benchmark of D-efficiency.  Export that design to .csv file.  Replace the contents of that .csv file with the custom design you created.  Import that new design (your custom one).

Then, run the Test Design again to make sure the precision of the design is still quite good.  Compare the D-efficiency of the new design to the previous benchmark.  You'd hope to have D-efficiency about 80% as good with your custom design as the standard CBC design, I would think.  But, you're going to lose some design efficiency due to your desire to constrain price constant in half the tasks.
answered Feb 6, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (150,315 points)
...