In April, 2004 we completed a second wave of our annual customer feedback survey. For those who participated, thank you! Aside from asking users how we might improve, we also collected some information that may be of interest to you. Customers specified what percent of projects over the last 12 months employed which specific conjoint methods.
The results (Exhibit 1) show that the momentum continues to shift (from 2003 to 2004) in favor of CBC (Choice-Based Conjoint). Considering only the projects that used CBC, ACA or CVA, the relative use of CBC increased from 50% to 61%. (Data are weighted by the number of projects completed.)
In the 1990s, ACA was the most widely used conjoint method (according to two industry surveys). Although we cannot be certain exactly when, in about the year 2000 CBC began to be used more often than ACA. One of the main reasons for this shift was the availability of HB methods (starting in the late 1990s) to estimate individual-level part worths from CBC data. Previously, only group-level estimation was available.
Users often test multiple methods to estimate part worth models, often selecting the one that best meets some criterion, such as predictive accuracy. Exhibit 2 shows that 62% of CBC users are using HB for their final models. Using HB to analyze CBC data importantly leads to individual-level part worth estimates.
ACA (Adaptive Conjoint Analysis) and CVA (Traditional Full-Profile Conjoint Analysis) have always supported individual-level estimation. Still, HB estimation can improve estimates for ACA and CVA beyond the classical OLS approaches. The use of HB in these more traditional contexts is growing, with 33% of ACA users and 25% of CVA users now relying on HB estimation.