|
Abstract:
|
Food and beverage products are characterized by sensory attributes whose intensities are measured by an expert panel to determine how products of interest differ from one another. Thus, knowing how these attributes drive consumer liking feeds salient information to the product development process. Sensory attributes are highly collinear; thus, principal components regression (PCR) has become a popular tool for modeling consumer liking vs. sensory attributes. PCR involves a preliminary principal components analysis of the attributes; liking is then regressed vs. the first k principal components (PCs) that explain high data variability, e.g., at least 70%. Attributes with low variability are expected to have low loadings on the first k PCs, and thus excluded in the PCR. But what if their variability is enough to drive consumer liking? Likewise, attributes with high variability may be those that consumers do not care about, but included in the PCR due to their high PC loadings. This presentation features a study that describes the above scenario and alternatives to PCR to increase the chances that drivers of liking become part of the regression, even those with low variability.
|
