Effect of Knowledge and Sensory Expertise — 275
Preliminary Study of the Effect of Knowledge and Sensory Expertise on Liking for Red Wines Michael Bom Frøst1* and Ann C. Noble2 Twelve inexpensive red wines were evaluated by descriptive analysis in which 12 trained judges rated 14 attributes in duplicate. Fifty-seven subjects rated liking for these wines on a 9-point hedonic scale. Subjects were segmented based on their level of wine knowledge (written test) and sensory expertise (sensory test) and on overall expertise, as estimated by the combined scores. Segments were formed by assigning lowest approximately 25%, medium approximately 50%, and highest approximately 25% to different groups. There was no correlation in performance between the two tests (Spearman’s ranked correlation = 0.057), indicating that sensory performance and wine knowledge are two distinctly different types of expertise and that one cannot be inferred from the other. The combined scores better reflect wine expertise, as true “wine experts” should possess knowledge in both domains. Significant differences in liking were found across all subjects. Based on the sensory expertise test, an overall significantly lower rating (p = 0.044) was found for the medium performance group. No overall significant differences in liking were found across groups based on performance in the wine knowledge test or the overall expertise. Despite this, significant differences in preference scores were found for several wines between groups. Individual differences in liking were significant, suggesting that individual preferences play a far larger role than mere level of wine knowledge or sensory expertise. External preference mapping was used to relate subjects liking to descriptive profiles. Although liking could not be modeled well from the sensory properties, some conclusions regarding liking for these subjects could still be made. Wines high in vanilla/oak aroma and high in either berry or canned vegetables aroma were liked significantly more than wines high in leather aroma and/or sour taste. Key words: Wine, experts, novices, expertise testing, hedonic evaluation, liking rating, sensory descriptive analysis, preference mapping
Preference or liking of a wine is influenced by many extrinsic factors in addition to flavor and appearance. Price, grape variety, vintage, origin, expert recommendations, winery reputation, bottle appearance, and awards can influence quality perception of wine. In addition, consumer experience with wine may influence preference and perception. In several food and beverage categories a difference in preference between consumer panels and trained panels has been observed [2]. In 1955, Gibson and Gibson [3] noticed that with increased experience, individuals tend to integrate details and recognize patterns, such as varietal identity.
descriptions of a set of wines to the same wines than were less experienced tasters [6]. Solomon [16] showed that descriptions by experts were more successfully matched to wines by other expert judges than by novices. When subjects were restricted in their use of descriptive terms, nonexperts used more sensory descriptors than the experts, but the experts were more consistent with one another in their use of terms. Novices are more apt to apply sensory terms listed on ballots to products than experienced judges, showing their greater susceptibility to the power of suggestion when first describing products, as suggested by Lawless [7].
Subjects experienced with wine flavor description have been shown to use more terms to describe the odor qualities of wines than inexperienced tasters. Furthermore, experienced judges, or “wine experts,” were better able to match previously written
It is not only in the domain of description of wine that the expert wine taster differs from the novice. Professional wine tasters performed better than control subjects in discrimination and identification tests to identify a group of nonwine stimuli (various common odors), suggesting that perceptual odor learning in nonlaboratory situations generalizes to performance in discrimination or identification [1]. Performance of wine tasters and novices may differ because of different cognitive strategies. Experts may differentiate the wine sensory properties more finely and may perceive more dimensions in the wines. Because of their greater experience with wines, they may integrate perceived flavors into more differentiated categories of wines. Solomon [15] suggested that these two kinds of differentiation may work in concert. Although these studies showed
1
Department of Dairy and Food Science, Royal Veterinary and Agricultural University, Rolighedsvej 30, 5, DK-1958 Frederiksberg, Denmark; 2Department of Viticulture and Enology, University of California, Davis, CA 95616. *Corresponding author [Email:
[email protected]] Acknowledgments: The donation of wines by participating wineries (Beringer Wine Estates, E&J Gallo, Hess Collection, Kendall-Jackson, Meridian, and Stag’s Leap) and distributors (Leestown Company and Seagram Chateau & Estate Wines) is gratefully acknowledged. The efforts of David Gillespie, Timothy Milos, Marianne Swaney, Gregory Winter, Christopher Simons, and Matthew Meyer in conducting the descriptive analysis and of Seung Joo Lee for providing valuable assistance during consumer tests are greatly appreciated. Two anonymous reviewers are thanked for their helpful comments on an earlier version of this manuscript. Manuscript submitted: August 2000; revised May 2001, July 2002
275 Am. J. Enol. Vitic. 53:4 (2002)
276 — Frøst and Noble
that variation in subject experience affects perception and ability to describe perception per se, they did not address the effect of level of experience on wine preferences.
training sessions were conducted. At each training session, descriptors for two or three wines were discussed and reference standards presented. (Descriptors and standards are listed in Table 2.) The 14 terms for the descriptive analysis were chosen by consensus. In the final training session, judges rated the wines under the same conditions as those used in the formal evaluations. Before each formal session, panelists smelled the reference standards. Intensity was rated on a 10-point category scale anchored at the end points with the terms “none” to “very intense.” All wines were expectorated. The wines were evaluated in duplicate with six wines presented in a balanced random order within each session. Samples (30 mL) were presented in coded black wine glasses. All evaluations were conducted in individual booths under incandescent white light.
The goal of this study was to examine differences in preference among groups of subjects with varied levels of wine knowledge and sensory expertise. Sensory expertise is here defined as the ability to discriminate among different aromas, to recognize different aromas by cued recall, and to describe a wine aroma by free recall. Subjects were tested for wine knowledge and sensory expertise, rather than assuming that some were experts and others novices in the domains of wine knowledge and, especially, sensory expertise. Prior studies examining differences between “wine experts” and “novices” in flavor perception [1] or aspects of wine perception [6,16] have drawn their “expert” subjects from wine-tasting societies or professional Tasting sessions. Wine liking was rated using a 9-point associations. They have assumed wine-tasting expertise based hedonic scale anchored at the end points with the terms “like either on subject background [16] or on testing of wine knowlextremely” and “dislike extremely” [8,12]. Six wines were preedge [1,6]. However, as the two types of expertise are different domains, it is necessary to test both in order to achieve a good measure of overall wine expertise. ExTable 1 Wines used for the study: Codes, varieties, region of origin, pertise in wine should be reflected in good performance vintage, and retail prices in June 1998. in both domains. Furthermore, it seems unlikely that level Abbreviation Variety Origin Vintage Price of wine knowledge, mainly achieved through reading A-CS Cabernet Sauvignon California 1995 $9.99 books and other sources of explicit knowledge, will alone B-CS Cabernet Sauvignon California 1995 $10.99 affect how a person likes a wine in a completely blind C-Mer Merlot California 1995 $9.99 tasting. Thus, the overall goals of the experiment were C-Zin Zinfandel Paso Robles, CA 1996 $9.99 two-fold. First, to develop a test of wine expertise, both D-PN Pinot noir California 1996 $12.99 wine knowledge and sensory expertise, and second, to D-Zin Zinfandel California 1996 $12.77 examine whether subject expertise level in the two doE-PN Pinot noir North Coast 1995 $8.99 mains separately and combined affects wine preference. F-CS Cabernet Sauvignon Sonoma County 1994 $7.97 Sensory profiling of the wines was performed and the F-Zin Zinfandel Sonoma County 1995 $7.47 results were related to subject liking ratings using preferG-Sang 90% Sangiovese Tuscany, Italy 1996 $10.99 ence-mapping techniques [4,14]. H-Mont H-M/C
Materials and Methods Wines. Twelve inexpensive red wines made by eight different wineries were selected for the experiment. Varieties, region of origin, vintage, retail prices, and codes for the wines are provided in Table 1.
Abruzzo, Italy Veneto, Italy
1996 Nonvintage
$6.99 $8.99
Table 2 Composition of reference standards.
Subjects. To recruit subjects, posters with questionnaires and information about the study were displayed on campus at the University of California, Davis; 116 persons responded to the introductory questionnaire. A subset of 80 persons was selected; those persons consumed wine at least twice per month and preferred red to white or preferred both equally. Of the selected subjects, 57 (26 females and 31 males) completed the study. Average age and years of wine-drinking experience were 32.5 (SD = 10.0) and 14.5 (SD = 10.7), respectively. Eighteen of the subjects had professions related to wine or were students in enology. Descriptive analysis. Twelve students at the University of California (seven females, five males) were selected as panelists based on interest and availability. Five orientation and
Montepulciano 60% Merlot 40% Cabernet Sauvignon
Descriptor Aroma Berry Artificial fruit Canned vegetables Green olives Vanilla/oak Butter Leather Black pepper Spicy
Standard (375 mL) 2 strawberries, 10 frozen raspberries, 4 frozen cherriesa 50 g tropical fruit Koolaid™ (Kraft) 100 mL 1:1 mixture of canned asparagus and green bean brines 14 green olivesa 20 drops vanilla extract and 4 or 5 small oak chipsa 3 Tbsp. butter (approx. 60 g)a 3 or 4 small pieces of leather soaked 20 mina Approx. 100 mg ground black pepper 100 mg ground nutmeg, 2 whole cloves,a 1 drop anise extract, 150 mg ground cinnamon
Taste Bitter
2.50 mg (low) and 39 mg (high) quinine
Mouthfeel Astringency
0.19 g (low) and 0.47 g (high) aluminum sulfate
a
Removed before serving. Am. J. Enol. Vitic. 53:4 (2002)
Effect of Knowledge and Sensory Expertise — 277
sented to each subject at each of two sessions. A complete block design was used to assign the wines to each session and randomize the presentation to each subject within a session. Wines (30 mL) were presented to the subjects in coded, clear wine glasses for evaluation in semiprivate booths. Subjects were asked to rate their liking of the wines in the order listed on the scorecards and to rinse with water between samples. Each wine was scored on a separate piece of paper, and subjects were instructed not to look at their previous scores. Segmentation procedure. Two tests were administered to segment the subjects. After the first session, subjects were given a Wine Trivia Quiz (Figure 1) to determine their level of knowledge about wine. After the second session, a Smell Association Test (Figure 2) was conducted to assess their sensory expertise. The Wine Trivia Quiz was constructed using exam questions for an undergraduate-level class in wine. It consisted of 11 questions with varying levels of difficulty, with a maximum score of 42 points (the grading system for scoring the answers is shown in Figures 1 and 2). The Smell Association Test was developed to assess subject ability to discriminate wine aromas (question 1, maximum 3 points), to recognize aroma (cued recall, questions 2 and 3, maximum 6 points), and to describe wine aroma (free recall, questions 4 and 5, maximum 12 points). Questions 2 to 5 were graded on accuracy of identification, with more specific answers yielding higher scores. Because the three types of questions varied in level of difficulty, they were assigned different scores for completely correct answers. Moreover, there was a difference in degree of correctness of the answers in questions 2 to 5. For example, for question 2 the correct answer was D. Black pepper (6 points), but Cloves was given 3 points, as cloves are in the same second tier (spicy) in the Wine Aroma Wheel [13]. In the free recall questions (4 and 5), correct and specific terms such as those on the outside tier of the Wine Aroma Wheel received more points than more general terms on inner tiers of the wheel. Maximum number of points that could be awarded for the Smell Association Test was 39. Maximum scores of the two tests were weighted to be approximately equal. Data analysis. Wine liking ratings for all subjects were initially tested with analysis of variance (ANOVA) to analyze overall differences in preference for the wines. Main effects included “wine” and “subjects.” To analyze for differences in preferences among groups of subjects, subjects were segmented into groups based on their scores from the two tests (Smell Association Test and Wine Trivia Quiz), both separately and combined (overall expertise). Correlation coefficients between the scores in the two tests were calculated and tested with Spearman’s rank correlation test [5]. Test of normality of the distribution of scores for the two tests and the combined score was done with Kolmogorov-Smirnov test (K-S test) of goodness of fit [5]. For each segmentation combination, three groups were constructed by assigning approximately 25% of subjects with lowest scores to one group (low), approximately 50% to a second group (medium), and the highest scoring 25% to a third group (high). It proved impossible to achieve exactly the same group sizes in the three segmentation procedures, as some subjects
had equal scores. The chosen group sizes were the best compromise in the vicinity of lower and higher quartile borders. Differences in liking ratings between segments were tested using ANOVA. Main effects included wine and segment and the two-way interaction wine x segment. Where relevant, differences were tested using Fischer’s least significant difference at the 5% level (LSD 5%) and indicating either 95% confidence intervals (95% CI) or the size of LSD 5%. For descriptive analysis, each descriptor was analyzed by ANOVA using a mixed model, in which panelists were treated as a random factor. All differences between the wines (main effect) were tested against wine x panelist interaction. Multivariate analysis of the descriptive data was performed using partial least squares regression (PLSR [9,10,11]), with indicator variables for the wines as X-matrix and all significant sensory descriptors as Y-matrix. Sensory scores were averaged over panelists. PLSR of subjects’ liking rating (wines as X-matrix and liking as Y-matrix) was performed to obtain an internal preference map. Liking ratings were standardized (each subject’s preference scores were divided by their standard deviation). Relationships between subjects’ liking and sensory profiles of the wines were analyzed by PLSR with the descriptive analysis data as X-matrix and preference scores as Y-matrix. Again, liking ratings were standardized. For this analysis, descriptive data were averaged both over panelists and replicates. For PLSR analysis, full cross-validation was used [10]. All statistical analyses were performed using SPSS (version 9.0.0, Chicago, IL) and Unscrambler (version 7.6, Camo, Trondheim, Norway).
Results and Discussion Descriptive analysis. ANOVA results of descriptive data showed that of the 14 terms, all but four differed significantly across the wines. Mean intensity ratings, LSD %, and p-values for descriptors that varied significantly over wines are provided in Table 3. Although the LSD 5% level is not a very conservative estimate of significant differences, it was applied for this analysis. With the LSD 5% as significance criteria, only two pairs of wines did not differ in intensity of any of the descriptors: C-Zin and F-CS, H-M/C and H-Mont. The configuration of the wines resulting from PLSR of the descriptive analysis data is shown in Figure 3A and B. The first dimension, which explains 44% of total variation in sensory data (Y-matrix), contrasted wines high in Leather (H-M/C and HMont) and low in Butter and Berry, versus those high in Butter and Berry aromas and Fruity by mouth and low in Leather aroma (D-Zin, A-CS, and E-PN) (Figure 3A). The second dimension (23% of variation in sensory data) separated the wines high in Green olive and Canned vegetables aromas (D-PN and C-Mer) from those high in Sourness (G-Sang, H-M/C, and H-Mont). The third dimension (15% of variation in sensory data) separated wines high in Astringency and Black pepper aroma (CZin and F-CS) from D-PN, which is high in Butter aroma (Figure 3B). No clustering of these wines occurred by grape variety, although grouping by producer was observed for wineries F, C, and H. Wines from producer H (H-Mont and H-M/C) did
Am. J. Enol. Vitic. 53:4 (2002)
278 — Frøst and Noble
Question 1
Score Frequency
Question 2
Score Frequency
Question 3
Score Frequency
Question 4
Score Frequency
Question 5
Score Frequency
Question 6
Score Frequency
Question 7
Score Frequency
Question 8
Score Frequency
Question 9
Score Frequency
Question 10
Score Frequency
Question 11
Score Frequency
Most beers contain 3-5% alcohol. Completely fermented table wines contain (choose only one answer)? A. 3-5% alcohol by volume B. 5-9% alcohol by volume C. 9-14% alcohol by volume D. 17-21% alcohol by volume E. More than 21% alcohol by volume Basic question: C correct = 1 0 1 3 54
Which of the following grape varieties are used for red wines? A. Cabernet Sauvignon B. Riesling C. Merlot D. Pinot Noir Basic question: A, C, D each = 1, minus 2 for others 0 1 2 0 2 3
E. Chardonnay
3 52
Which of these grape varieties make red wine (indicate all that apply)? A. Müller-Thurgau B. Syrah C. Viognier D. Gamay E. Zinfandel Basic question: B, D, E, F each = 1, minus 2 for others 0 1 2 2 3 10
F. Gewurztraminer
3 12
F. Cabernet Franc
4 30
Which of the following areas are recognized as AVA (Approved Viticultural Area or American Viticultural Area)? A. Napa Valley B. Stag’s Leap District C. Sacramento Viticultural Area D. Lodi Appellation E. Santa Barbara Subtle question: A = 1, B, D, E each=2, minus 2 for C 0 1 2 5 12 1
3 17
4 1
5 18
6 0
Which of the following areas are in Napa Valley (indicate all that apply)? A. Stag’s Leap B. Howell Mountain C. Atlas Peak D. Guenoc Valley Basic question: A, B, C each = 1, minus 2 for D, E, F 0 1 2 34 9 8
7 3
E. Iron Horse Valley
F. Chalk Hill
3 6
Which of the following grape varieties are used for the famous wines from Bordeaux (indicate all that apply)? A. Pinot Noir B. Gamay C. Cabernet Sauvignon D. Grenache E. Sangiovese F. Merlot Harder question: C + F = 5, C alone = 2, C or F + all other = 1 0 1 2 5 12 18 10 17
Which two of the following grape varieties are used for the famous wines from Burgundy? A. Pinot Noir B. Gamay C. Cabernet Sauvignon D. Grenache E. Chardonnay
F. Hermitage
Harder question: A + E = 5, A alone = 2, A or E + all other = 1 0 1 2 5 15 24 1 17
Which one of the following is the major grape variety used for the wines from Chianti in Italy (choose only one answer)? A. Pinot Noir B. Nebbiolo C. Cabernet Sauvignon D. Grenache E. Sangiovese F. Mouvedre Harder question: E alone = 5, -1 for additional checks 0 4 5 21 6 30
Which of the following grape variety or combination of varieties is used for the red wines from Rioja in Spain (choose only one answer)? A. Garnacha, Monastrell, Pedro Ximénez, and Mazuelo B. Tempranillo, Viura, and Malvasia C. Macabeo D. Tempranillo, Garnacha, Mazuelo, and Graciano E. Malvasia and Tempranillo F. Tarragona and Graciano Harder question: D = 5, A = 1 0 1 17 16
5 24
Which of these varieties are used in German wines? A. Viognier B. White Riesling C. Muller-Thurgau Basic question, B + C = 3, B= 1, C=2, other minus 1 0 1 2 7 15 5
D. Semillon
E. Palomino
3 30
What variety is used to make white wine with the label Fumé Blanc? A. Chardonnay B. Sauvignon Blanc C. Cabernet Franc D. Semillon Basic question, B = 1, other minus 1 0 1 29 28
Figure 1 Contents of the Wine Trivia Quiz: Questions, list of possible answers, scores for answers, and frequency (out of 57) of scores to individual questions. Am. J. Enol. Vitic. 53:4 (2002)
E. Pinot Blanc
F. Pinot Gris
Effect of Knowledge and Sensory Expertise — 279
Question 1
The intensity of fruity/berry aroma is much higher in one of the two presented wines (A and B). Indicate which one is most intense: Correct: 3 points; Incorrect: 0 points (The wine was spiked with 2 fresh strawberries, 10 frozen raspberries, 4 frozen cherries /375 mL)
Score Frequency
Question 2
0 4
3 53
The wine is intense in one of the following aromas. Indicate which one you think it is: A. Cloves
B. Bell pepper
C. Banana
D. Black pepper
E. Soy
Correct: (D) 6 points; words in the same third tier of the wine aroma wheel [13]: 3 points (A, also an aroma in the spicy group); other words: 0 point (The wine was spiked with 200 mg ground black pepper/375 mL) Score Frequency
Question 3
0 4
3 2
6 51
The wine is intense in one of the following aromas. Indicate which one you think it is: A. Licorice/Anise
B. Vanilla
C. Cloves
D. Chocolate
E. Smokey aroma
Correct: (C) 6 points, words in the same third tier: 3 points (A, also an aroma in the spicy group), other words: 0 point (The wine was spiked with 5 whole cloves soaked for 20 min/375 mL) Score Frequency
Question 4
0 9
3 9
6 39
The wine is intense in a particular aroma. Describe the aroma of the wine with one or a few words. Correct: (asparagus) 12 points; words in the same third tier: 9 points (e.g. green beans, green olives etc.), words in the same middle tier: 6 points (e.g., canned or cooked), other words: 0 point (The wine was spiked with 35 mL brine from canned asparagus/375 mL)
Score Frequency
Question 5
0 20
3 0
6 9
9 5
12 23
The wine is intense in a particular aroma. Describe the aroma of the wine with one or a few words. Correct: (vanilla) 12 points; words in the same third tier: 9 points (e.g., oak or cedar); words in the same inner tier: 6 points (woody); Chocolate or cocoa: 3 points; other words: 0 point (The wine was spiked with 10 drops of vanilla extract/375 mL)
Score Frequency
0 9
3 19
6 1
9 0
12 28
Figure 2 Contents of the Smell Association Test: Questions, list of possible answers, scores for answers, and frequency (out of 57) of scores to individual questions. The base red wine used for the reference standards was used to prepare spiked samples.
not differ significantly in ratings for any descriptor. In contrast, wines from producer D were widely separated from each other.
wide variation in preferences. Two pairs of wines were not rated significantly different in any descriptors (C-Zin and F-CS, H-M/C and H-Mont, Table 3). Both pairs did not differ in degree of liking.
Liking by all subjects. Mean liking ratings and standard deviations are shown in Figure 4. Results from ANOVA, Table 3 Mean intensity ratings, 95% confidence interval (95% CI), and p-values for terms varying including model parameters significantly across wines (p