You know what you like, but what about everyone else? A Case study on Incomplete Block Segmentation of white-bread consumers.

Transcription

1 You know what you like, but what about everyone else? A Case study on Incomplete Block Segmentation of white-bread consumers.

2 Abstract One man s meat is another man s poison. There will always be a wide range of consumer liking response across any product category. Cluster analysis can provide consumer segments based upon common liking that reflect underlying sensory preferences. To determine valid population segments requires a large sample of consumers. As the number of products tested by each consumer increases experimental bias degrades the data quality. A limited number of well-designed incomplete sample sets can provide cleaner responses. The challenge in this study was to collect 200 valid consumer responses for each of 12 commercial white breads whilst minimizing fatigue.

3 What we know about consumers There is no such person as an average consumer. There is no product that is universally liked, even water. Traditional demographics are no indicator of consumer preference Consumers Lie!

4 What we know about products There is no product that is universally liked, even water. Optimization of products is desirable from the point of view of efficiency and market To optimize, you must have a clear target. Unless you segment your consumers based upon their sensory preference you will not have a clear target.

5 Consumer Segmentation of BIB liking data of 12 Cabernet Sauvignon wines 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.

6 The Effect of Order and Day on Consumer Liking 12 White Wines, 115 Consumers, CBD 12:12 over 3 Days All

7 Typically, segmentation of consumer liking data requires a complete block. 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.

8 Data collection at LCBO stores

9 Mean Liking across All Consumers W6 W1 W4 Overall Liking (n=614) W10 W12 W8 W7 W11 W9 W5 W3 W

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 (OP&P, Utrecht) Apply the cluster solutions to the original data to create segments

11 1 2 3

12 Increase Relative Stress By Decreasing Number Of Clusters 0.20 T1 Clustering results

13 Clusters 28% 23% Cluster Cluster Cluster % 32% Cluster 3-195

14 W4 W1 W6 W7 W8 W9 W12 W11 W2 W10 W5 W3 Cluster 1 Liking (n= 172 or 28%) W2 W10 W3 W1 W11 W5 W8 W4 W12 W6 W9 W7 Cluster 2 Liking (n=142 or 23%) W9 W6 W6 W7 Cluster 3 Liking (n= 195 or 32%) W8 W3 Cluster 4 Liking (n = 105 or 17%) W12 W5 W3 W1 W2 W11 W5 W4 W10 W9 W1 W12 W8 W10 W11 W7 W4 W

15 Astringent Bitter Sour Sweet Leather Raisin Fruity Eucalyptus Woody Pepper Vanilla Floral Coffee Asparagus Smoke Tobacco Cluster 1 Sensory Contrast Positive

16 Astringent Bitter Sour Sweet Leather Raisin Fruity Eucalyptus Woody Pepper Vanilla Floral Coffee Asparagus Smoke Tobacco Cluster 2 Sensory Contrast Positive

17 Astringent Bitter Sour Sweet Leather Raisin Fruity Eucalyptus Woody Pepper Vanilla Floral Coffee Asparagus Smoke Tobacco Cluster 3 Sensory Contrast Positive

18 Astringent Bitter Sour Sweet Leather Raisin Fruity Eucalyptus Woody Pepper Vanilla Floral Coffee Asparagus Smoke Tobacco Cluster 4 Sensory Contrast Positive

19 Statistical Challenge We need a valid approach to segmentation of consumer IB data Possibly a combination of sensory best practice, experimental design and advanced statistical analysis

20 Factor Scores plot : dimension 1 versus PC 1 vs PC 2 W3 Astringent W11 W6 W2 Bitter W12 Raisin W W10 Fruity WoodyEucalyptus Leather Vanilla Sour Pepper Floral Tobacc o W5 W4 W8 Sweet Green Asparagus 2.58 Coffee Smoke W9 W1-2.58

21 Factor Scores plot : dimension 3 versus PC 3 vs PC 4 W3 W2 W10 W9 Leather W1 Sweet Eucalyptus Asparagus W6 Fruity Green Floral Bitter Coffee W12 Pepper Astringent Woody 2.58 Raisin Tobacc ovanilla Smoke W5 Sour W11 W7 W4 W8-2.58

22 Quadrangles and Triangles Factor Scores plot : dimension 1 versus Factor Scores plot : dimension 3 versus W3 W2 W3 W10 Astringent W11 W6 Raisin W7 W2 Bitter W9 Leather W12 W1 Fruity WoodyEucalyptus Leather Sweet Eucalyptus Vanilla Sour Asparagus Pepper W6 Fruity Green Floral Bitter Coffee W10 Floral Tobacc o W W12 Pepper Astringent Woody 2.58 W4 W8 Raisin Tobacc ovanilla Sweet Green Smoke Asparagus W5 Coffee Smoke Sour W11 W9 W4 W7 W8 W Consumer Research February 2012

23 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.

24 Segmentation of Sensory-Informed Designed Incomplete Block of Consumer Liking Data

25 The Bread Study 12 Consumer Sliced White Breads Calibrated Descriptive Analysis for sensory profiles Sensory selection of the test products Efficient Incomplete Blocks of 12 present 6 A Nested experiment of 3 and 4 of category consumers 200 observations per product Segmentation using a model-based approach