Abstract

Conjoint analysis is a widely used marketing research tool for analyzing consumer preferences for a set of product/service attributes and estimating part-worths for the levels of each attribute. Although numerous studies have confirmed the reliability and validity of part-worth estimates, several studies have reported what was called the Number of Levels (NOL) effect, especially in full-profile conjoint designs. The NOL effect occurs in conjoint analysis when the estimated relative importance weight of an attribute increases as a function of the number of levels by which it is defined. This study investigates the NOL effect for Adaptive and Choice-Based Conjoint Analysis (ACA and CBC), two increasingly popular data collection methods. Three experiments were conducted with 684 physicians in Canada to test the hypothesis that NOL effect will be observed for both ACA and CBC but the effect will be stronger for CBC than ACA. The findings suggest that while CBC suffers from the NOL effect, ACA does not. The results also suggest that a less important attribute may be more susceptible to the NOL effect. Given the findings, it is recommended that when CBC is used it is better to have equal number of levels across all attributes in the conjoint design whenever possible. Also, relative importance weights from two different studies where the number of levels differ should not be compared. ACA does not seem to have these limitations.