In the 1970s, leading academics suggested that conjoint analysis be limited to six attributes or fewer (for traditional, ratings-based conjoint which is rarely used today). But real-world needs for decision-making are more demanding. We recently surveyed some of our most prolific customers and found that one-third of their recent conjoint analysis studies involved 10 attributes or more, with 6% involving 20 attributes or more. (The reporting covered 952 projects conducted by 39 researchers.)
Conjoint analysis has come a long way since the ’70s, making possible what previously seemed daunting or downright impossible. Adaptive methods (e.g., Adaptive CBC– ACBC) let us ask respondents upfront which attributes matter to their buying decisions and carry only those attributes forward into the conjoint analysis exercise. Another strategy is the partial-profile approach, where each respondent evaluates each attribute and level, but typically only five to seven attributes at a time as they are rotated into each choice question. These design strategies combined with advances in HB utility estimation have increased the accuracy and expanded the possibilities for conjoint analysis.
Megan Peitz and Abby Learner (Numerious, Inc.) recently conducted a comprehensive research-on-research comparison between different approaches to CBC involving both 10 and 18 attributes. The subject matter was cellphone choice. They compared full-profile CBC (where all attributes are always shown), partial-profile CBC, and two approaches that involve only taking the most important attributes (plus brand and price) forward into the conjoint exercises (ACBC and their own programmed “bespoke CBC”).
They found that all methods produced similar results and predictions, though partial-profile CBC was the least similar (of note, showing lower price sensitivity). Previous researchers have also found that partial-profile CBC leads to lower price sensitivity and price importance. For their 18-attribute experiment, respondents preferred the methods that didn’t show all 18 attributes in each choice question.
Interested in learning more? You can subscribe to Megan and Abby’s video presentation (along with over 30 other video presentations and tutorials by a variety of industry researchers) as presented at the 2020 Sawtooth Software European Virtual Conference.
Bryan Orme – Sawtooth Software