Comparing rapid sensory approaches Christian Dehlholm, chde@teknologisk.dk Senior Consultant, Danish Technological Institute 1
Why descriptive! 1900 quality scoring and grading Focus on differentiating products 1930 Grading Foods by a Descriptive Method Early descriptive whole product system Minimising assessor subjectivity 1930s and 40s Selecting the right assessors Development of scales 1950 Modern sensory evaluation The Flavor Profile by Cairncross & Sjostrom C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 3 TDS/TOS PM/Napping Sorting Pick-any/ CATA 2
Validity and reliability Comparisons between methods and panels Holistic! C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 5 Overall configurations compared C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 6 3
Confidence Conventional Profile Panel A Panel B C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 7 Confidence Global Napping Panel A Panel B C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 8 4
Confidence Overview Panel A Panel B Conventional Profile 9-10 h Flash Profile 3 h Partial Nappings 80 min Free Multiple Sorting 60 min Global Napping 40 min C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 9 Confidence Overview Panel A Panel B Conventional Profile 9-10 h Flash Profile 3 h Partial Nappings 80 min Dim 2 (20.59 %) -4-2 0 2 Appearance P5 Free Multiple P3 P7P6 P4 Sorting P8 P1 P2 P9 Global -4-2 Napping 0 2 4 Dim 1 (44.94 %) Dim 2 (22.23 %) -3-1 1 2 3 Taste P5P1 P9 P7 P4 P2 P8 P6 P3-3 -1 1 2 3 4 Dim 1 (27.59 %) Dim 2 (19.71 %) -2 0 1 2 3 4 P3 P9P4 Mouthfeel P5 P1 P2 60 min P7 P8 P6 40 min -3-1 0 1 2 3 Dim 1 (34.03 %) C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 10 5
Semantics Overview Panel P1 P2 P3 P4 P5 P6 P7 P8 P9 Conventional Profile Flash Profile Free Multiple Sorting Global Napping Appearance Partial Nappings Conventional Profile Flash Profile Free Multiple Sorting Global Napping Taste Partial Nappings Conventional Profile Flash Profile Free Multiple Sorting Global Napping Mouthfeel Partial Nappings C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 11 Conclusions 1 Correlation between evaluation time and reliability Semantic differences Free Multiple Sorting Comparable to projective mapping Provides categorical data instead of Euclidean Partial Napping Closer relationship to conventional profiling Possibility to isolate single modalities -> directing perception Small changes could have large effects 6
Consensus vs. Individual Panelists vs. Experts Group no. Assessor type Evaluation type Evaluation method G0 Trained panel Individual Conventional profiling G1 Trained panel Individual Napping (GN/PN) G2 Trained panel Individual Napping (GN/PN) G3 Trained panel Consensus Napping (PN) G4 Trained panel Consensus Napping (PN) G5 Trained panel Consensus Attribute rating G6 Trained panel Consensus Attribute rating G7 Experts Consensus Attribute rating G8 Experts Consensus Attribute rating G9 Experts Consensus Attribute rating C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 13 Conclusions 2 Product experts less in common with references than trained panel Semantic differences Not recommended to compare consistency between consensus groups Product training could change the picture Horizontal vs. vertical frame dimensions Geometry matters theoretically Rectangular geometry promotes two-dimensionality Semantics improve the model 7
PM on sound News speaker Lisa Ekdahl Rage Against the Machine C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 15 PM on sound with naïve assessors (projection strategies) C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 16 8
PM on sound with naïve assessors (projection strategies) News Rage Against First Second Third Projection strategy Lisa Ekdahl Total (%) recording The Machine session session session 1) Linear (one-way) 45% 30% 30% 31% 36% 37% 35% 2) L or T (two-way) 12% 20% 16% 14% 18% 16% 16% 3) Dispersed 16% 13% 22% 19% 18% 13% 17% 4) Lumpy 14% 11% 10% 8% 14% 12% 12% 5) Categorical 6% 10% 7% 10% 6% 7% 8% 6) Categorical dbl. linear 6% 5% 10% 7% 4% 10% 7% 7) Other 1% 11% 6% 10% 4% 5% 6% Total (n) 83 83 83 83 83 83 249 C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 17 Conclusions 3 Possible for naïve assessors to use projective mapping (on sound) Variations in projection strategy Categorisation of strategies One- and two-dimensional projection strategies > 50% Dispersed projections < 1/6 Might affected by Sample complexity Method experience Data analytical approach does not match evaluation method Might explaining noise in data 9
Choosing the right method! Main considerations C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 20 10
Sensory assessor types C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 21 Training The attribute Personal constructs Aligned panel Misaligned panels C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 22 11
Conclusions 4 Configurations Methods separate in similar ways (main dimensions) Less time -> weaker separations Semantics Same methods provide different descriptions Predetermined vocabularies -> predetermined results Product approach Holistic approaches allow a closer-to-real product separation A rapid approach is easy to administer General issues Simpler methods might be difficult to understand (simple instructions) Holistic methods does not provide intensity data Analytical methods does minimalize conceptual reality Future developments Concept Scaling C. Dehlholm, Nordic Workshop, 22 May 2013 Slide 24 12
Use rapid approaches For an overview of unknown products As a basis for sampling to larger experiments In NPD, in idea and exploration phases New semantics are wished -> Proj. mapping (Global Napping, consumers) Directed semantics are wished -> Partial Napping, panel Pre-known semantics are wished -> Flash Profile (with introduced vocabulary) Thank you! 13