Is the absolute scaling model the basis for the 9-point hedonic scale? Evidence from Poulton’s Stimulus Range Equalizing Bias

Abstract

Besides confirming earlier work, this study provided evidence for the theoretical position that consumers use cognitive strategies based on an absolute behavioral model, for the traditional verbal version of the 9-point hedonic scale. This is opposed to the relative model, which is appropriate for the less frequently used purely numerical and unstructured version of the 9-point hedonic scale. After establishing a constant rank order of liking for six chocolate stimuli, consumers rated the chocolates using the traditional verbal version and an unstructured numerical version of the 9-point hedonic scale. The numerical data produced by each scale were different and incompatible for significant majorities of consumers, after correction for their unimodal and bimodal designs. This occurred with both rank-rating and serial monadic protocols. This was confirmed using pattern change scores, a parametric measure of the discrepancy between ‘words only’ and ‘numbers only’ scaling responses. It also checked for ‘scaling drift’, inconsistency in scale use. The fact that numerical scales based on the relative model are prone to Poulton’s Stimulus Range Equalizing Bias and that scales based on an absolute model are not, was used to investigate the model for the traditional verbal 9-point hedonic scale. The verbal scale was much less prone to Poulton’s bias than the numerical scale but it was still partially prone. Poulton’s bias had demonstrated that the model for the traditional verbal 9-point scale was different from the relative model for a purely numerical scale. Yet, it had failed to demonstrate that the different model was the absolute model. It had partial absolute and partial relative properties. Descriptions of the nature of the model are discussed. Compared with rank-rating, the effects of high memory load, which is systemic for the serial monadic protocol, was evidenced by higher error rates due to forgetting, higher amounts of scaling drift and discrepancies between the order of ranked and rated data.