Comparing Derived Importance Weights Across Attributes: An Experimental Study Using Conjoint Analysis

In this study we investigate the possibility that derived attribute importance weights, as commonly computed in conjoint analysis applications, may be systematically influenced by the number of levels on which an attribute is defined for constructing hypothetical stimuli. Specifically, the research hypothesis is that the importance weight increases as the number of levels increases, even when the range of variation is constant for the attribute. Forced preference rank order judgments about hypothetical summer jobs were collected from first-year MBA students at a private university. The students were randomly assigned to one of four experimental conditions. The experimental groups differed in the number of levels used for two of the summer job attributes and in the method of data collection used. The range of variation was held constant for one of these attributes. The results suggest that the experimental conditions have a substantial and statistically significant impact on the magnitude of relative derived attribute importance weights. A mathematical adjust- ment for this phenomenon is suggested, and is shown to eliminate a great deal of the variation across experimental conditions.