STATISTICAL PACKAGE CLUSTERING MAY NOT BE BEST FOR GROUPING CONSUMERS TO UNDERSTAND THEIR MOST LIKED PRODUCTS

Ensuring that new products satisfy specific groups of consumers can impact successful product development. In sensory studies, cluster analysis has been used to segment consumers. Researchers often analyze mean values of products for consumer segments, presuming that the segmented consumers like or dislike similar products. This study investigates how well most/least liked products match for individual members of clusters using various cluster methods in two sensory studies. Four statistical package clustering (SPC) methods were used with hedonic data and data transformed to ranks. Next, the products most frequently rated/ranked highest in each cluster were examined. Four manual clustering groups were extracted and compared with the results of the SPC methods. Standard SPC was not found to separate consumers appropriately to understand their ranking/rating of most/least liked products. For these data, additional manual clustering was necessary to produce consumer cluster segments where the consumers within each group had the same highest/lowest scoring products. PRACTICAL APPLICATIONS Statistical package clustering (SPC) is a common method for determining consumer clusters, but it may not be the best method for separating consumers or for understanding their most or least liked products. The findings from this research show that the assumption that cluster analysis will produce clusters containing consumers who have the same most or least liked products is false. This is important because it shows that clustering consumers using typical SPC may not produce the homogeneous segments that researchers would like to obtain. This can impact further analyses of data for product optimization and preference mapping. SPC with further manual clustering is recommended for more homogeneous segments. In addition, new SPC methods should be developed that generate cluster segments that are more homogeneous.