Enriching sensory and consumer datasets with temporal metadata
Descriptive analysis provides valuable information about the sensory properties of consumer products, but this information lacks the temporal dimensionality of real-world sensory experiences. Type II error occurs when the descriptive sensory panel fails to differentiate between products known to be discriminable. Findlay (2000) reported no meaningful reduction in beta risk when descriptive analysis on manipulated salad dressings was augmented by order of initial perception data. Attributes on the ballot appeared in fixed order but panellists were required to respond in order of perception, and responding to order of perception increased the complexity of descriptive analysis. Pineau (2004) discussed the Temporal Dominance of Sensations (TDS) methodology, which also recognizes the promise of temporal data.
Guessing models for difference tests assume that correct responses are given by discriminators, who perceive true differences, and non-discriminators, who guess correctly. Panellists might adopt strategies that enhance discriminative ability in a triangle test (Rousseau, 2001), although experimenters cannot verify strategy use.
Enriched datasets might provide insights in these and other areas. Computers facilitate data collection, and have potential to gather temporal metadata, providing contextually enriched data without detriment to existing analyses. Enriched datasets might include irregularly spaced temporal data representing discrete, sometimes dependent events that might better model and provide insights into sensory experiences. Myriad opportunities for investigation arise.
A descriptive panel might respond to attribute intensity multiple times – leaving their task largely unchanged – permitting supplemental analysis of either incomplete or interpolated data within specific time intervals. Relationships between response time, decision, and accuracy in difference testing might provide new information. Relationships between correctness and decisiveness in descriptive panel training also merit exploration. Consumers might be grouped for analysis according to temporal response patterns.
To assess the potential value of temporal metadata, twelve panellists were selected and instructed to respond to dual attribute time intensity (DATI) where anchors were “Same” and “Different”. Samples A and B were Premium salted soda crackers (32.6 mg Na / cracker) and Premium unsalted soda crackers (21.4 mg Na / cracker), respectively. Each panellist received four pairs (AA, AB, BA, BB) according to Williams Latin square design (four treatments).
There were 40/48 (83%) correct identifications (p<0.001). Panellist indicated “same” after an average of 23.2 s (sd=13.3 s) when correct and 25.3 s (sd=16.3 s) when incorrect. Panellists indicated “different” after an average of 14.5 s (sd=9.3 s) when correct and 17.7 s (sd=10.4 s) when incorrect. Panellists took significantly longer to declare “same” than “different” according to one-way ANOVA (p=0.019) in this preliminary test: further investigations are planned to determine relationships between d’, decision, correctness, and response time.
Computerized sensory and consumer data collection systems allow rapid implementation of tests based on standard methodologies, cost reductions in the testing process, global reach, and increased throughput. Historical data, rich with temporal and other contextual information, could further increase utility by providing a relationship-rich data store, which can be mined in interesting ways, and provide a basis for knowledge discovery.
Castura, J.C., & Findlay, C.J. (2006). Enriching sensory and consumer datasets with temporal metadata. In: 8th Sensometrics Meeting. August 2-4, Ås, Norway.