Segmentation of BIB Consumer Liking Data of 12 Cabernet Sauvignon Wines

1,028 views

Published on

Presented at University of Stellenbosch, September 2010

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
1,028
On SlideShare
0
From Embeds
0
Number of Embeds
2
Actions
Shares
0
Downloads
0
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Segmentation of BIB Consumer Liking Data of 12 Cabernet Sauvignon Wines

  1. 1. Segmentation of BIB Consumer Liking Data of 12 Cabernet Sauvignon Wines Chris Findlay, PhD Chairman Compusense Inc. Guelph, Canada
  2. 2. Background • Consumer testing of beverage alcohol has a number of serious challenges. The effect of consumption of alcohol is a limiting factor in obtaining complete block data. • Collecting consumer data over several days affects the quality of the consumer response. By the third day, most consumers are behaving like experts, a conclusion that is supported by the decrease in first position effect. Segmentation of BIB Consumer Liking Data August 26, 2010
  3. 3. The Effect of Order and Day on Consumer Liking 12 White Wines, 115 Consumers, CBD 12:12 over 3 Days 70 60 50 40 1 2 3 30 4 20 10 0 1 2 3 All Segmentation of BIB Consumer Liking Data August 26, 2010
  4. 4. Ensuring Quality Data • Careful selection of appropriate consumers • Sensory selection of test products • Diligent application of Best Practices • Typically, segmentation of consumer liking data requires a complete block. Segmentation of BIB Consumer Liking Data August 26, 2010
  5. 5. The Study • In this study, 12 Cabernet Sauvignon wines were evaluated by over 600 red wine consumers in a 12 present 3 Balanced Incomplete Block design. Each consumer tasted 3 of the wines in a single 10 minute session, with demographic questions providing a break between samples. • A total of 11 sessions were conducted at 5 LCBO store locations. Segmentation of BIB Consumer Liking Data August 26, 2010
  6. 6. Factor Scores plot : dimension 1 versus 2 2.58 PC 1 vs PC 2 Sensory Map #1 W3 Astringent W11 W6 W7 W2 Bitter W12 Raisin FruityWoody Eucalyptus Leather Vanilla Sour Pepper -2.58 W10 Floral T obacco W5 2.58 W4 W8 Green Sweet Asparagus Coffee Smoke W9 W1 -2.58
  7. 7. Factor Scores plot : dimension 3 versus 4 2.58 PC 3 vs PC 4 Sensory Map #2 W3 W2 W10 W9 Leather W1 Eucalyptus Sweet Asparagus W6 Coffee Floral Green Bitter Fruity -2.58 W12 Pepper Woody Astringent 2.58 Raisin T obacco Vanilla Smoke W5 Sour W11 W7 W4 W8 -2.58
  8. 8. Data collection at LCBO stores Segmentation of BIB Consumer Liking Data August 26, 2010
  9. 9. RAW RESULTS Mean Liking across All Consumers W6 W1 Overall Liking W4 (n=614) W10 W12 W8 W7 W11 W9 W5 W3 W2 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 Segmentation of BIB Consumer Liking Data August 26, 2010
  10. 10. Segmentation Procedure • Substitute missing data with for each panelist 1. Panelist Mean for the 3 products tested 2. Product mean for each product 3. Grand mean for all products • Cluster using Qannari method using Senstools 3.3.1 (OP&P, Utrecht) • Apply the cluster solutions to the original data to create segments Segmentation of BIB Consumer Liking Data August 26, 2010
  11. 11. Segmentation of BIB Consumer Liking Data August 26, 2010
  12. 12. T1 Clustering results Increase Relative Stress By Decreasing Number Of Clusters 0.20 The average liking response for each panelist was inserted 0.15 into the missing data points. The Qannari Clustering method (Senstools 3.3.1) provided a four cluster solution. 0.10 0.05 0.00 20 19 18 17 16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 Segmentation of BIB Consumer Liking Data August 26, 2010
  13. 13. Clusters 28% 23% Cluster 1 - 172 Cluster 2 - 142 Cluster 4 - 105 Cluster 3 - 195 17% 32% Segmentation of BIB Consumer Liking Data August 26, 2010
  14. 14. W2 W4 W10 W1 Cluster 2 Liking Cluster 1 Liking W3 (n=142 or 23%) W6 (n= 172 or 28%) W1 W7 W11 W8 W5 W9 W8 W12 W4 W11 W12 W2 W6 W10 W9 W5 W7 W3 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 W9 W6 W6 W8 Cluster 3 Liking Cluster 4 Liking W7 (n= 195 or 32%) W3 (n = 105 or 17%) W12 W5 W3 W1 W2 W11 W5 W4 W10 W9 W1 W12 W8 W10 W11 W7 W4 W2 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 3.5 4.0 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 Segmentation of BIB Consumer Liking Data August 26, 2010
  15. 15. Cluster 1 Astringent Sensory Contrast Bitter Positive Sour Sweet Leather Raisin Fruity Eucalyptus Woody Pepper Vanilla Floral Coffee Asparagus Smoke Tobacco -3 -2 -1 0 1 2 3 Segmentation of BIB Consumer Liking Data August 26, 2010
  16. 16. Cluster 2 Astringent Sensory Contrast Bitter Positive Sour Sweet Leather Raisin Fruity Eucalyptus Woody Pepper Vanilla Floral Coffee Asparagus Smoke Tobacco -3 -2 -1 0 1 2 3 Segmentation of BIB Consumer Liking Data August 26, 2010
  17. 17. Cluster 3 Astringent Sensory Contrast Bitter Positive Sour Sweet Leather Raisin Fruity Eucalyptus Woody Pepper Vanilla Floral Coffee Asparagus Smoke Tobacco -3 -2 -1 0 1 2 3 Segmentation of BIB Consumer Liking Data August 26, 2010
  18. 18. Cluster 4 Astringent Sensory Contrast Bitter Positive Sour Sweet Leather Raisin Fruity Eucalyptus Woody Pepper Vanilla Floral Coffee Asparagus Smoke Tobacco -3 -2 -1 0 1 2 3 Segmentation of BIB Consumer Liking Data August 26, 2010
  19. 19. Statistical Challenge • We need a valid approach to segmentation of consumer BIB data • Possibly a combination of sensory best practice, experimental design and advanced statistical analysis Segmentation of BIB Consumer Liking Data August 26, 2010
  20. 20. • Let’s consider a sensory space Factor Scores plot : dimension 3 versus 4 Factor Scores plot : dimension 1 versus 2 2.58 2.58 W3 W2 W3 W10 Astringent W11 W9 W6 W7 W2 Bitter Leather W12 Raisin W1 Eucalyptus Sweet FruityWoody Eucalyptus Leather Asparagus Vanilla Sour Coffee Pepper W6 Floral Green Bitter Fruity -2.58 W10 Floral T obacco W5 -2.58 2.58 W12 Pepper Woody Astringent 2.58 Raisin Tobacco Vanilla W4 W8 Green Smoke Sweet Asparagus W5 Coffee Smoke Sour W11 W7 W9 W4 W8 W1 -2.58 -2.58 • Can we find logical contrasts to test Segmentation of BIB Consumer Liking Data August 26, 2010
  21. 21. • Quadrangles and Triangles Factor Scores plot : dimension 1 versus 2 Factor Scores plot : dimension 3 versus 4 2.58 2.58 W3 W2 W3 W10 Astringent W11 W6 W7 W2 Bitter W9 Leather W12 Raisin W1 FruityWoody Eucalyptus Sweet Eucalyptus Leather Sour Asparagus Vanilla Pepper W6 Coffee Floral Green Bitter Fruity -2.58 W10 Floral T obacco W5 -2.58 2.58 W12 Pepper Woody Astringent 2.58 Raisin T obacco Vanilla W4 W8 Green Smoke Sweet Asparagus W5 Sour W11 Coffee Smoke W7 W9 W4 W8 W1 -2.58 -2.58 Segmentation of BIB Consumer Liking Data August 26, 2010
  22. 22. Is there a Preference? • To state a true preference a consumer must see a real difference. • Otherwise it’s just a guess. • Considerations: – Do all attributes influence liking equally? – For all consumers? – In all contexts? Are you ready for a new era of sensory & consumer research? April 23, 2010
  23. 23. Is there a Better Clustering Method? • Model-based clustering is being developed to improve the validity of the clusters. • Ongoing research by Dr. Paul McNicholas and his team at the University of Guelph and Compusense is attempting to combine sensory design and effective clustering to provide robust solutions to the challenge of BIB segmentation of high fatigue products. • Please refer to the Sensometrics.org for proceedings of the 10th Conference held in Rotterdam, July 2010. Are you ready for a new era of sensory & consumer research? April 23, 2010
  24. 24. Conclusion • The selection of sensory contrasts may be used to strengthen the design of BIB experiments. If random incomplete blocks are chosen it is possible that groupings of quite similar products may be received by some assessors and widely different products by others. • Prior knowledge of the sensory properties is essential in assigning blocks that will emphasize the inherent differences in the products and improve the quality of data from the study. Segmentation of BIB Consumer Liking Data August 26, 2010
  25. 25. For more information, please contact Chris Findlay, PhD Chairman +1 800 367 6666 (North America) +1 519 836 9993 (International) cfindlay@compusense.com

×