Sawtooth Software: The Survey Software of Choice

Avoiding the “Blank” Concept in CBC/Web

Oftentimes in CBC research, we include “blank” levels within attribute lists, such as for binary (on/off) attributes where we don’t want to display anything for the “off” level. To further illustrate, some researchers are asked to use conjoint analysis to study advertising statements and/or graphics that could be assembled in print ads or packaging.

One approach is to group different statements (levels) within attributes where one level within each attribute is “blank,” and to show two to four attributes (partial profile display) within the same CBC task:

Which of these groups of statements about Shampoo X makes you most want to buy it?
Leaves your hair shiny   You'll love the shine
Get noticed with Shampoo X Turn heads with Shampoo X Captivate the senses with Shampoo X
You’re worth it--don’t skimp on your hair Your hair is worth it  

Current versions of CBC/Web haven’t let you do this, as it is not possible to prohibit a completely blank concept from appearing (unless you import your own experimental design into CBC/Web). To avoid this situation, the next version of CBC/Web (within SSI Web v6) will allow you to specify the minimum and maximum number of non-blank attributes that may appear together within a concept. This is a specialized type of prohibition.

The layout above reflects the default CBC layout. We’d recommend you randomize attribute order (an automatic option in CBC/Web) to control for order biases for this type of study. There are other layout options in CBC/Web as well. If you use Free Format questions to customize the look of your CBC questions (with your own HTML, Flash, or CSS), you could develop attractive composite displays of graphics and statements, even mimicking the look of an ad or the product packaging itself. With this type of research, it may be critical to approach such realism.

Other improvements for v6 of CBC/Web include: cross-concept prohibitions; up to k-way prohibitions, from k=2 attributes to k=total attributes in the study (both within- and cross-concept); constant-sum response; volumetric response (no constant sum requirement); and dual-response None (requires CBC/HB v4 for analysis).