A Basic Conjoint Analysis to Six Sigma Design of Experiments (DOE) Translational Dictionary

This article provides a quick cross reference between some commonly used conjoint analysis terminology and Six Sigma terminology and/or classical Design of Experiments (DOE) terminology. It gives a simple conjoint terminology background for Six Sigma-trained individuals who would like to research conjoint articles further.

Since conjoint analysis tools were developed by marketing professionals, they developed their own terminology for concepts and items that have corresponding Six Sigma offsets. This is not surprising, since the development groups worked independently and at different times. Consequently, each group developed different terms for the same or similar concepts. Most conjoint terms have direct or equivalent Six Sigma terms as shown in Table 1. Six Sigma practitioners would call a Design of Experiments an experiment, whereas a conjoint practitioner would call the Design of Experiments a study or investigation.

Table 1: Some Corresponding Six Sigma to Conjoint Analysis Terms

A typical Six Sigma Design of Experiments has factors that may be continuous or discrete. A traditional conjoint investigation has attributes or features. Normally, attributes or features are discrete in nature and particular discrete levels may not have a logical or categorical order (i.e. Brand: Sony, Samsung, Sanyo, Hitachi, Sharp, etc. or Color: red, yellow, brown, blue, orange, etc.). Other attributes may have logical or ordered relationships such as size, temperature or resolution. Currently, many conjoint references will state that only one continuous attribute is used (e.g. cost), although examples may be found with several continuous attributes. These situations utilize variable conjoint analysis, a more recent evolution of the tool set. Figure 1 is an example of a typical attribute/level layout showing possible attributes and levels for digital televisions. This is a very condensed and limited example for illustrative purposes only and is not meant to capture all possible attributes and levels. (Click on diagrams to enlarge.)

Figure 1: Potential Attributes and Levels for Digital Televisions


Figure 2: Typical Highly Fractional Experimental Design



Figure 2 is a highly fractionalized experimental design; 7/72 or ~1/10 fraction. Due to the physical nature of conducting a conjoint analysis (sorting and ranking cards), there are human limitations that restrict the size of the study. Typically, the number of questions asked is limited to 30. Therefore, conjoint designs are highly fractionated and highly confounded. Since conjoint analysis assumes that main effects are significant while two-way and higher level interactions are insignificant, highly fractionated designs are acceptable. In a conjoint study, each line (Design of Experiments run) is converted into a card, profile, stimuli or panel. Figure 3 shows a card that was generated for the first run or row in Figure 2. In a traditional full profile conjoint study, cards are generated for each run or row in the experimental design. They are called cards because they are often physically printed out in the form of note cards. Sometimes cards are called stimuli because the cards are presented to a respondent as stimuli and the amount of respondent desire, interest or preference is quantified. When the term panel is used it often refers to the entire card set or the complete experimental design. (Click on diagram to enlarge.)

Figure 3: Physical Representation of the First Experimental Design Run Shown in Figure 2


Another common conjoint term is "full" vs. "partial" profile, which hasn’t been discussed above. A full profile study means that all attributes in the conjoint study where included in each card or profile. A partial profile means that not all attributes are included on every card or profile. Normally partial profiles occur in conjunction with conjoint screening methods like partial profile choice experiments (PPCE).

The basic concepts between Six Sigma and conjoint Design of Experiments are very similar, but they have offsetting terms. Both examine the effect of independent variables on dependent variables. Terminology differs slightly, but once well-versed in both Six Sigma and conjoint analysis, practitioners can easily work in both realms.

Read Kris Stark and Hank Sanftleben's first article on conjoint analysis here.