Range and Number-of-levels Effects in Derived and Stated Measures of Attribute Importance*

We study how the range of variation and the number of attribute levels affect five measures of attribute importance: full profile conjoint estimates, ranges in attribute level attractiveness ratings, regression coefficients, graded paired comparisons, and self-reported ratings. We find that all importance measures are affected by the range manipulation. The number of attribute levels affects only two measures. The results allow us to benchmark the magnitude of the number-of-levels effect against the range effect: conjoint importance estimates were approximately equally affected by a threefold increase in the range of attribute variation and by the insertion of two intermediate attribute levels. Our findings show that the number-of-levels effect is most likely due to respondents’ tendencies to distribute their mental stimulus representations and their responses uniformly over the corresponding continua.