Presented at the 2013 Pangborn Symposium in Rio de Janeiro, Brazil.
Multivariate mapping techniques are valuable to sensory scientists because of their ability to visually represent differences among products in a multidimensional space and to connect these differences to hedonic assessments of the products. There are many mapping methods in use and their respective analysis outcomes often differ because of the disparate process assumptions that underlie them - without an understanding of these assumptions, sensory scientists are not in an informed position to interpret their results. In this presentation, we use a data-driven approach to explore the assumptions that underlie two of the multivariate mapping methods in common use: External Preference Mapping and Probabilistic Unfolding. Both techniques use the same data, but their processes are reversed: Probabilistic Unfolding first creates the product space from hedonic data and regresses sensory information into the space second, while External Preference Mapping first creates a space from the sensory information before regressing hedonic information into the space. To understand the effects of these different processes, we examined 9 large-scale category appraisals and found that External Preference Mapping can have limitations when the products are spread on more than 2 sensory dimensions. In that case, more dimensions (4 or 5) are required to define the individual model - such an approach increases the difficulty of interpreting the results. In contrast, reasonable interpretations of these same datasets can be obtained from Probabilistic Unfolding in lower dimensionality (2 or 3 dimensions). This result has important implications for the ease of communication of sensory results with management, especially given the widespread use of External Preference Mapping.
Colleagues can download the poster here:
The Effect of Dimensionality on Multivariate Mapping of Hedonic Data
Not a Colleague? Click here to join for free!