- Email: [email protected]
The application of check-all-that-apply (CATA) consumer profiling to preference mapping of vanilla ice cream and its comparison to classical external preference mapping
Food Quality and Preference 21 (2010) 394–401
Contents lists available at ScienceDirect
Food Quality and Preference journal homepage: www.elsevier.com/locate/foodqual
The application of check-all-that-apply (CATA) consumer profiling to preference mapping of vanilla ice cream and its comparison to classical external preference mapping Lauren Dooley, Young-seung Lee, Jean-François Meullenet * Department of Food Science, University of Arkansas, 2650 N. Young Avenue, Fayetteville, AR 72704, United States
a r t i c l e
i n f o
Article history: Received 19 January 2009 Received in revised form 27 August 2009 Accepted 11 October 2009 Available online 21 October 2009 Keywords: Check-all-that-apply Preference mapping Vanilla ice cream Multiple factor analysis
a b s t r a c t This study was conducted to evaluate the use and efficacy of check-all-that-apply (CATA) data for the creation of preference maps, and to compare these maps to classical external maps generated from traditional sensory profiles. Ten commercial vanilla ice cream products were presented to 80 consumers. Consumers answered an overall liking question using the 9-point hedonic scale and a CATA question with 13 attributes which described the sensory characteristics of vanilla ice cream. A trained descriptive panel of 17 individuals developed a profile of 23 attributes for the vanilla ice cream products. Preference maps created by CATA counts were compared to those by descriptive profiles via multiple factor analysis (MFA). The characterization of the products by both sensory methods showed very good agreement between the methods. The MFA of map configurations showed fair agreement between the techniques used to produce the preference maps, indicating that CATA data applied to preference mapping gave similar results to external preference mapping. Ó 2009 Elsevier Ltd. All rights reserved.
1. Introduction Check-all-that-apply (CATA) questions regarding consumerperceived product attributes have been used in consumer studies to determine what sensory attributes may be characteristic of a specific product (Lancaster & Foley, 2007). Some researchers already advocate the use of consumer sensory profiling to lead product development as an alternative to classical sensory profiling (Punter, 2008; Worch, Lê, & Punter, 2008). The format of the CATA question allows consumers to choose all potential attributes from the given lists to describe the test products. This is different from scaling in the sense that no intensities are given to the attributes. In addition, the descriptors are not constrained to product sensory attributes, but could also be related to product usage or concept fit. This type of methodology has the advantage of gathering information on perceived product attributes without requiring scaling, allowing for a slightly less contrived description of the main sensory properties of the product tested (depending on how the terms are created). The actual generation of CATA terms can be performed in many ways: the consumers can choose words to describe the product during the test (modified free choice profiling), terms can be given
* Corresponding author. Tel.: +1 7853418710. E-mail address: [email protected] (J.-F. Meullenet). 0950-3293/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.foodqual.2009.10.002
by a trained panel, or terms can be generated by consumers not testing the product (i.e. a focus group). Free choice profiling allows consumers to use as many or as few words as necessary to describe the product and evaluate the intensities of the chosen attributes, resulting in a less expensive and more accurate view of consumer perception and acceptance (Deliza, Macfie, & Hedderley, 2005; Gonzáles-Tomás & Costell, 2006; González Viñas, Garrido, & Wittig de Penna, 2001; Williams & Langron, 1984). However, if each consumer selects his/her own terms, the analysis becomes cumbersome since each term must be subjectively interpreted and combined with similar terms (Meilgaard, Civille, & Carr, 2007). Seo, Lee, and Hwang (2009) used consumers to describe sensory characteristics of coffee. Verification of the terms was then conducted by other consumers to confirm that the terms were appropriate and understandable. While this is an effective method, the time required is to complete the test is extensive. Terms generated by a trained panel have the benefit of being more comprehensive and better described, though they may be too complex for the average consumer to understand and could require simplification. Altering the terms in this manner is difficult to do while retaining the correct term description and definition. However, it has been shown that differences in sensory evaluations between trained and untrained (naïve consumers) are minimal (Benedito, Cárcel, & Mulet, 2001; Guerrero, Gou, & Arnau, 1997; Husson & Pagés, 2003; Lelievre, Chollet, Abdi, & Valentin, 2008), so using less obscure terms by a descriptive panel could be a
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401
beneficial tool for creating a CATA list. Ultimately, it is the researcher’s decision as to which method is most appropriate. The CATA method requires minimal instruction, is relatively easy to perform and is completed quickly (Lancaster & Foley, 2007). Furthermore, it could be a more practical approach than intensity scaling from the standpoint of consumer-led product development. Since CATA responses are directly linked to consumers’ perception of product characteristics, these responses could be utilized as supplemental data to maximize acceptance of the targeted products by consumers. CATA provides information on which attributes are detectable according to consumers and how that may relate to their overall liking and acceptance. Understanding sensory characteristics in the process of new product development is of great importance, as failure to obtain correct information about the sensory attributes may lead to fast disappearance of the new products from the marketplace (Stone & Sidel, 2007). To understand the relationship between consumer and sensory data, preference mapping is a useful method. Preference mapping is a widely used group of multivariate statistical techniques designed to optimize products by understanding the structure between consumer preference and sensory data to identify drivers of liking (Faye et al., 2006; Greenhoff & MacFie, 1999). Among the various product optimization mapping methods, the Euclidian Distance Ideal Point Mapping (EDIPM), an extension of Multidimensional Preference Mapping (MDPREF), is a new approach based on a density analysis of individual consumer ideal product placements in the product configuration space (Meullenet, Xiong, & Findlay, 2007). In this approach, the ideal point of individual consumers is the point which minimizes the correlation between Euclidian distances to the products and hedonic scores. Another optimization mapping technique, the Response Surface Model (RSM), proposed by Danzart, is based on external preference mapping (Danzart, Sieffermann, & Delarue, 2004). Multidimensional representation of sensory stimuli is first created by sensory (i.e. external) data. The consumer data for individual consumers is then regressed against the product coordinates in the sensory space to determine ideal points for both the individuals and the group (Meullenet, Lovely, Threlfall, Morris, & Striegler, 2008). To investigate the efficacy of CATA scales within the sensory environment, this study used ice cream as the testing medium. Ice cream is one of the most popular frozen desserts in the United States. The US ice cream market continues to grow and is expected to be valued at over $10 billion by 2012 (Datamonitor, 2007). Vanilla is the most popular ice cream flavor in the US and accounts for almost half of all ice cream sales (Bodyfelt, Tobias, & Trout, 1988). There are many companies producing and commercializing ice cream in the US. In order to compete in this highly competitive market, it is crucial for ice cream manufacturers to understand the strong and weak points of their products, and how consumers’ attitudes and preference patterns affect their products.
395
The objectives of this study were to (1) assess the use of CATA attribute responses for 10 commercial vanilla ice creams as an alternative to consumer attribute intensity ratings, and (2) compare CATA-generated preference maps to classical external maps generated from traditional sensory profiles.
2. Materials and methods 2.1. Samples and sample preparation Fifteen commercially-available ice creams were initially selected from local supermarkets for testing. Preliminary screening of texture and flavor attributes eliminated five samples due to brand replication and fat content, a popular indicator of ice cream quality, and the use of natural or artificial vanilla flavor. Ice creams were selected so that various combinations of these quality factors were represented in the study. The 10 remaining products, consisting of two high-fat products, six products with moderate fat content and two low-fat products, are detailed in Table 1. One scoop of each product was placed individually into a lidded white plastic container (45 mm diameter) coded with a three-digit random number. Samples were stored in a commercial-grade freezer (TS-49, True Manufacturing Co., St. Louis, MO, USA) at 18 °C for at least 24 h prior to testing to ensure sample consistency. All samples were tempered for 2 min at room temperature prior to serving for both descriptive analysis and consumer testing. The 2 min increment was determined to be the most appropriate tempering time by observing the condition of the ice cream as a function of time at room temperature. Samples were presented in a sequential monadic order to panelists according to a complete randomized block design, and the serving temperature ( 12 + 2 °C) was strictly monitored to maintain consistency (Bower & Baxter, 2003; Li, Marshall, Heymann, & Fernando, 1997). 2.2. Descriptive analysis The 10 vanilla ice creams were evaluated for taste, aromatic, flavor, and texture attributes by a descriptive panel of 17 individuals trained by the SpectrumÒ method (Sensory Spectrum Inc., Chatham, NJ, USA). Panelists have over 100 h of training and an average of 1000 h of testing experience. Two orientation sessions were conducted to familiarize the panelists with the samples. Flavor and texture lexicons were developed in four sessions, as described in Tables 2 and 3, respectively. The lexicons consisted of 23 total attributes specific to vanilla ice cream and definitions of each sensory attribute with associated references. Panelists quantified all attributes on a line scale from 0 to 15 (Meilgaard et al., 2007). Unsalted crackers and water were provided for panelists to clean and rinse their palate between samples, and a 10 min break helped prevent fatigue. The flavor attribute testing for all 10 products was
Table 1 A list of 10 commercial vanilla ice cream products. Brand
Code
Name/description
Fat content (%)
Flavor
Blue Bell Blue Bunny Ben and Jerry’s Best Choice Breyers Edy’s ‘‘Grand” Great Value Guilt Free Haagen-Dazs Yarnell’s
A B C D E F G H I J
Homemade vanilla Premium all natural vanilla Vanilla Vanilla Natural vanilla Rich and creamy vanilla Vanilla Vanilla Vanilla Homemade vanilla
13 10 24 11 12 5 11 4 28 15
Natural and Natural and Natural Artificial Natural Natural Artificial Natural and Natural Natural and
Manufacturer artificial artificial
artificial artificial
Blue Bell Creameries Wells’ Dairy, Inc. Ben and Jerry’s Homemade Holdings, Inc. (Unilever) Wal-Mart Stores Inc. Unilever Nestle Wal-Mart Stores Inc. Yarnell Ice Cream Co. Nestle Yarnell Ice Cream Co.
396
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401
Table 2 Flavor lexicon for vanilla ice cream. Term Basic taste Sweet
Salt
Reference
Intensity
The basic taste, perceived on the tongue, stimulated by sugars and high potency sweeteners
Solutions of sucrose in spring water
2–2.0%
5–5.0%
The basic taste, perceived on the tongue, stimulated by sodium salt, especially sodium chloride
Solutions of sodium chloride in spring water
10–10.0% 0.2–2.0%
16–15.0% 0.35–5.0%
0.5–8.5% 0.7–15.0% 0.05–2.0%
0.55–10.0%
0.15–10.0% 0.05–2.0%
0.20–15.0% 0.08–5.0%
0.15–10.0%
0.20–15.0%
Whole milk 1/2 and 1/2 Heavy whip cream
4.0 6.5 9.0
Alum –6.0
Tea –9.0
Sour
The basic taste, perceived on the tongue, stimulated by acids, such as citric acid
Solutions of citric acid in spring water
Bitter
The basic taste, perceived on the tongue, stimulated by substances such as quinine, caffeine, and certain other alkaloids
Solutions of caffeine in spring water
Aromatics Vanillin Cooked milk Milky Buttery/fat
Flavor Non fat dry milk Caramelized Oxidized Woody/stick Metallic Feeling factors Astringent
a
Definition
The sweet, vanilla-like aromatic characteristic of ethyl vanillin or imitation vanillas The aromatic associated with the flavor of milk, heated to the scalding point The aromatic associated with skim or whole milk products or milk derived products The aromatic associated with fresh butterfat; sweet cream
The aromatic associated with boxed, nonfat dry milk or milk reconstituted from dry milk solids, cardboardy A sweet aromatic characteristic of browned sugars and other carbohydrates The aromatics associated with slightly oxidized fats and oils The aromatic associated with dry fresh cut wood; balsamic or bark-like (1) The aromatic associated with metals, tinny or iron (2) a flat feeling factor stimulated on the tongue by metal The feeling factor on the tongue or other skin surfaces of the mouth described as puckering or drying
0.08–5.0%
Universal scalea
Heavy whipping Cream Cream cheese
Universal scalea
(1) Alum solution (0.01%) (2) tea in 1000 ml spring water for 5 min
Soda note (soda cracker 3.0), cooked apple note (applesauce 7.0), orange hit (orange juice 10.0), cooked grape note (grape juice 14.0), cinnamon hit (chewing gum 16.0).
conducted on two consecutive days, followed by two consecutive days for texture attribute testing.
The CATA counts were totaled for each product (Table 5) and the resulting contingency table was used in subsequent analyses.
2.3. Consumer testing
2.4. Statistical analyses
Eighty consumers were recruited to participate in the vanilla ice cream test at the University of Arkansas Sensory Service Center (Fayetteville, AR). Qualification criteria included adults over 18 years of age and vanilla ice cream product consumption at least one to two times per week. The test was sequential monadic and sample presentation orders were balanced in each serving position using the Williams Latin Square design (Williams, 1949). Each consumer tested five samples the first day and tested the remaining five the second day. The sessions were divided into 30 min increments, with most consumers finishing within 20 min. Consumers were asked to evaluate overall liking, appearance, flavor and texture attributes of each sample on a 9-point hedonic scale (1 = ‘‘dislike extremely”, 9 = ‘‘like extremely”). The attributes of scoopability, color, sweetness, vanilla flavor, creamy flavor, smoothness, melting in the mouth, melting in the bowl and hardness of each product were evaluated using the 5-point ‘‘Just About Right” (JAR) scale (data not shown or used in this study). Though not used in these analyses, since this JAR data was collected, the authors felt it necessary to mention. The final question of the survey listed common ice cream attributes and asked the consumers to check all attributes that applied to the given sample (Table 4).
External preference mapping was performed according to Danzart et al. (2004) using either the descriptive sensory profiles or CATA counts to determine a group ideal point (Meullenet et al., 2008). First, principal component analysis was performed on mean sensory profiles while correspondence analysis was performed on the CATA counts. To determine the area of the map maximizing the number of consumers satisfied, a quadratic model was constructed (i.e. regressing hedonic scores against the principal component scores), and the area of acceptability for each consumer was identified (i.e. area of the map where the hedonic score was predicted above the mean score for each consumer). The area of maximum density was regarded as the ideal point solution for this method. Both analyses are considered here as external because the product configurations were obtained from data other than liking. Since the descriptive profiles and CATA counts were scaled differently, the two data sets were standardized across all products prior to external preference mapping to minimize differences inherent to the scaling. To standardize the data, the data matrix had products in columns and attributes in rows. The data was standardized by columns so that the mean was zero and the variance was one.
397
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401 Table 3 Texture lexicon for vanilla ice cream. Term
Definition
Reference
Intensity
Scoopability/ manual firmness
The force required to cut the sample with your spoon. 2 and 4 min
Evaluate the force required to remove one/spoonful of sample from the cup (soft–hard)
Peanut butter
4.0
Cream cheese Metal spoon sample and references
8.0
Peanut butter
4.0
Cream cheese Plastic spoon
9.0
Marshmallow cream
2.5
Nougat Plastic spoon
4.0
Hardness-oral
Denseness
The force required to compress the sample between the tongue and mouth roof. 2 min
The amount of air or fluffiness perceived in the sample. 2 min
Compress through sample one time with tongue. Evaluate the force required to fully compress the sample (soft–hard)
Compress through sample one time with tongue. Evaluate the amount of air perceived in the sample (airy–compact)
Degree of ice
The amount of ice crystals felt in the mouth during the chew. 2 min
Compress 1/2 tsp of the sample. Evaluate the amount of ice crystals perceived in the sample (no ice–much ice)
Sherbet Plastic spoon
3.0
Smoothness
The amount of particles perceived in the sample during the chew. 2 min
Manipulate the sample three times with tongue. Evaluate the amount of particles perceived in the sample (not smooth– smooth)
Peanut butter
5.0
Cool whip Plastic spoon
14.0
Soy ice cream
5.0
Blue Bunny Plastic spoon
9.0
Saltine
0.5
Ritz Pringles Mayo
2.5 5.0 9.0
Frosting Three pulls Plastic Spoon
10.0
Rate of melt
Mouth coat
Elasticity
The rate in which the ice cream changes forms from a solid to a liquid. 2 min
The amount and degree of residue felt by the tongue when moved over the surface of the mouth. 2 min
The degree to which the samples appears to have an elastic/doughy impression. 4 min only
Place 1/2 tsp of sample in mouth and evaluate the rate in which the sample melts (slow–fast)
Expectorate the sample and feel the surface of the mouth with the tongue to evaluate (none–much)
Stick spoon in the sample and pull out. After three pulls evaluate the amount of sample pulled up by the spoon (none– much)
Table 4 An example of check-all-that-apply (CATA) question. Check all attributes that describe this sample: h Buttery h Sweet h Milk/dairy flavor h Custard/eggy flavor h Corn syrup h Artificial vanilla h Natural vanilla h Creamy flavor h Soft h Hard h Gummy h Icy h Creamy/smooth
Multiple factor analysis (MFA) was conducted using FactoMineR in R (v.2.6.2, 2008) to examine the similarities first between the descriptive profiles and CATA counts (Figs. 1 and 2) and second between the multivariate product configurations obtained from the three preference mapping techniques employed. For each preference mapping technique, the coordinates of the ideal points in the maps were estimated and used in the MFA analysis as illustrative data. MFA is a useful statistical technique to analyze the similarity of a set of observations explained by several different groups of variables on comparable or contradictory scales (Abdi & Valentin, 2007). MFA is able to balance the influence of each variable, can compare multiple data sets, and can demonstrate patterns of attribute correlation (Lê, Pagês, & Husson, 2008; Morand & Pagès, 2005; Nestrud & Lawless, 2008).
3. Results and discussion Internal preference mapping was also conducted as a point of comparison for descriptive analysis and CATA-based external preference mapping. The comparison seems appropriate since the CATA-based external preference can be considered as a hybrid method. Euclidian Distance Ideal Point Mapping (EDIPM, Meullenet et al., 2008) was used for internal preference mapping. The product configuration in the space was derived from principal component analysis of the centered overall liking data.
3.1. Comparison of product descriptions by CATA and descriptive analysis Individual product maps were created by MFA using descriptive sensory profiles and consumer CATA counts (Fig. 1). Overall, MFA comparing the characterization of the products by both profiles showed agreement between the two methods, although only 51%
398
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401
Table 5 Total counts of check-all-that-apply attributes for each product. Brand
Soft
Hard
Gummy
Icy
Creamy/ smooth
Buttery
Sweet
Milk/dairy flavor
Custard/eggy flavor
Corn syrup
Natural vanilla
Artificial vanilla
Creamy flavor
Blue Bell Blue Bunny Ben and Jerry’s Best Choice Breyers Edy’s ‘‘Grand” Great Value Guilt Free Haagen-Dazs Yarnell’s
42 37 21 29 8 25 51 34 17 43
20 20 31 29 43 38 11 22 32 16
5 7 7 4 7 4 12 3 7 4
37 16 22 28 61 20 8 29 14 14
31 45 34 37 10 35 59 35 40 51
44 46 29 43 17 39 38 30 30 21
58 53 53 61 57 61 46 52 61 60
60 58 50 62 52 59 58 57 52 54
45 35 28 32 25 26 21 22 35 22
10 13 16 13 11 16 12 23 21 15
25 28 27 34 37 28 27 19 19 24
39 33 29 33 37 43 31 45 49 42
45 51 44 54 24 53 52 35 43 47
Fig. 1. Multiple factor analysis individual product plots using descriptive sensory profiles (D) and CATA counts (C). Product codes are listed in Table 1.
of the variability was explained by the first two MFA dimensions. A and J showed the largest variance between the two methods, largely due to the disagreement between the descriptive and consumer (CATA) maps for these two products (not shown). As discussed by Pagès (2004), based on absolute scores of partial lines on each product for the first dimension, Fig. 1 indicates that products B, H, I, and J were more characterized for the first dimension by descriptive analysis profiles than CATA counts. If a line is drawn from the end of the descriptive and CATA vector for each of these products to the x-axis, descriptive analysis rates higher on the first dimension. Products E and G were more characterized by CATA counts for the first dimension. Fig. 2 shows the variable correlation circle obtained by MFA comparing descriptive analysis profiles and CATA counts. The vectors showed a strong correlation for descriptive (d) mouth coat, smoothness and rate of melt, with CATA counts (c) perceived soft, creamy flavor and creamy/smooth, between degree of ice (d) and icy (c), elasticity (d) and gummy (c), and between caramelized (d) and corn syrup (c). The opposite vector directions for some CATA and descriptive descriptors with opposite meanings also show agreement between the two methods. For example, hardness (c) was opposite to softness (c), and artificial vanilla opposite to
Fig. 2. Multiple factor analysis variable correlation circle obtained using descriptive analysis and CATA terms. Refer to Tables 2–4 for complete attribute descriptions. ‘CS_c’ and ‘Cflavor_c’ represent creamy/smooth and creamy flavor, respectively, in CATA terms.
natural vanilla flavor. No correlations were observed between hard (c) and icy (c), and gummy (c), or between milky (d) and butterfat (d). Descriptive sensory profiles did not show any correlation between bitter (d) and sweet (d) taste in the vanilla ice cream products, while bitter (d) had negative correlation with salt (d) despite its low loading. The most influential attributes (i.e. highest loadings) were found to be sweet, bitter, vanillin, degree of ice, elasticity, and smoothness for descriptive sensory profiles. Icy, natural vanilla, creamy/smooth, creamy flavor and artificial vanilla attributes for CATA consumer profiles played a relatively more important role in determining product locations in the map. 3.2. Preference mapping results and group ideal point locations Each of the three preference mapping techniques employed (external mapping on descriptive data and CATA and internal preference mapping) allow the identification of a group ideal product location in the maps. This point, in all three cases, is the location in the map maximizing the percentage of consumers who would be satisfied by a product placed at that location. Ideally, the three methods would give approximately the same answer and this is what we seek to assess here.
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401
399
Fig. 3. Results of external preference mapping using (a) descriptive analysis profiles (DD) and (b) CATA term counts (DC). The virtual product labeled ‘‘Opt” represents the location in the map maximizing the percentage of consumers satisfied (Danzart et al., 2004). Other product codes are as in Table 1. Values in parenthesis represent mean overall liking values.
Fig. 4. Results of internal preference mapping. The virtual product labeled ‘‘Opt” represents the location in the map maximizing the percentage of consumers satisfied (Euclidian Distance Ideal Point Mapping (EDIPM), Meullenet et al., 2007). Other product codes are listed in Table 1. Values in parenthesis represent mean overall liking values.
400
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401
The results of external preference mapping using descriptive sensory data (DD) and CATA counts (DC), and internal preference mapping are graphically shown in Figs. 3 and 4. The first two dimensions of the maps explained 50.2% and 59.0% of the variance in descriptive profiles and CATA counts, respectively. The average individual consumer fit was similar for the DD map (R2 = 0.59) and the DC map (R2 = 0.61) (data not shown). Overall, the products’ spatial representation on the first two dimensions seemed to differ. The product coordinates used to regress individual overall liking scores for the external maps were obtained from two different data sources and the disagreement is not too surprising. For the DD map (Fig. 3a), the ideal product was closest to product F (OL = 6.7), followed by product D (OL = 6.8). Product E (OL = 5.5), least liked by consumers, was furthest from the other products, including the optimal product. The DC map (Fig. 3b) placed the group ideal point near products A (OL = 6.3) and D (OL = 6.8), while product E was located on the far left of the map by itself, showing agreement with the DD map. Product D, which was most liked by consumers, was in closest proximity with the group ideal point for both maps. For internal preference mapping, EDIPM placed the ideal point closest to products A and D and furthest from product E (Fig. 4). This finding was similar to that of the DC map. To more finely compare the level of agreement/disagreement between the three maps, MFA was employed using the first two dimensions of the three maps created. Fig. 5 represents the locations of the 10 commercial vanilla ice cream products and the ideal product (i.e. individual factor map) determined for each of the three types of preference mapping on the first two MFA factors. As shown in Fig. 5, the optimal product was in close proximity with products A (OL = 6.3), D (OL = 6.8), and F (OL = 6.7), while E (OL = 5.5) was furthest from the remaining products and the ideal product. Overall, MFA of the three product maps showed fair agreement between the approaches employed. However, the DD map was contrary to the other two methods for products A and J, while the internal map showed dissimilarities to DD and DC, par-
Fig. 5. Multiple factor analysis individual product plots of the product configurations (first two dimensions) determined for external preference mapping using descriptive analysis data (DD) and CATA counts (DC) and internal preference mapping (EDIPM). The virtual product labeled ‘‘Opt” represents the optimal product derived from the three mapping methods employed. Other product codes are listed in Table 1.
ticularly for product F. This could be explained by the fact that product F was near the origin in both DD and DC maps, but was located in the far right upper quadrant in the EDIPM map (see Figs. 3 and 4). One key result of this analysis is that optimal product location showed little variation between the three maps. 3.3. Ideal product profiles Although the location of the group ideals in the three preference maps was reported to be fairly invariable, there is a need to determine the sensory characteristics that should be exhibited by the group ideal products. If the three methods are in agreement, we would expect the sensory profiles of the group ideals to be fairly similar. To determine the level of agreement between the preference mapping methods employed to determine the group ideals in the three maps, the CATA counts were regressed against the first two dimensions of the product spaces created using descriptive analysis data, CATA data and internal preference mapping. From these models, the ideal product CATA profiles were predicted. This is also known as reverse regression. Fig. 6a gives the fit of the CATA attributes for the three mapping methods employed. Overall, the CATA attributes were better fitted in the CATA space, particularly natural vanilla, creamy flavor, creamy/smooth, soft and hard attributes (R2 > 0.6). This is not surprising since this product space was derived from the CATA data. Overall, the CATA attributes were not as well fitted
Fig. 6. Ideal vanilla ice cream profiles according to descriptive, CATA and EDIPM data using (a) CATA attribute fit (R2) and (b) normalized ideal CATA counts.
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401
in the product/preference space derived from internal preference mapping while the quality of the fit was somewhere in between for the product space created from descriptive analysis data. This result is also not surprising since both the CATA and descriptive analysis product spaces are based on product sensory properties while the internal preference mapping product space is based on liking data. Fig. 6b represents the ideal sensory profiles in terms of normalized CATA counts for the three mapping methods employed. The normalization of these ideal profiles are easier to interpret than ideal raw CATA counts because they show if the ideal CATA count is above or below the mean of the observed counts. With normalized CATA counts, a positive value indicates that the ideal product was above the mean CATA count across products for a given attribute. These profiles allow the determination of the level of agreement between the three mapping methods employed. From Fig. 6b, we conclude that the three mapping methods provided similar optimal values for gummy, hard, soft, sweet, buttery, creamy/ smooth, icy, creamy flavor and custard/eggy flavor. However, there was disagreement for natural vanilla, artificial vanilla and corn syrup. For these attributes, the CATA methodology recommended high natural vanilla, low artificial vanilla and low corn syrup CATA counts.
4. Conclusions Overall, the characterization of the 10 commercial vanilla ice cream products shows good agreement between descriptive sensory profiles and consumer-perceived CATA profiles. Moreover, the CATA attribute data applied to preference mapping gave similar results to external preference mapping. The advantage of this technique is that the task asked of consumers is simple (i.e. when compared to intensity ratings), and that the responses may be more spontaneous than when intensities are rated. The limitation of this approach is that the optimal profile derived from the CATA maps is in terms of response counts and not intensities as given by a trained panel or consumers using attribute intensity scaling. Since it can be hypothesized that the three maps created could have been more similar for product configurations, including optimal points, with additional dimensions being considered, the use of additional PCs may be of interest in future studies. Although CATA questions seem to have some validity to characterize the sensory properties of products as perceived by consumers, many questions remain unanswered. Further studies may include the assessment of the effects of order and number of terms CATA questions on attribute selection.
References Abdi, H., & Valentin, D. (2007). Multiple factor analysis (MFA). In N. Salkind (Ed.), Encyclopedia of measurement and statistics. Thousand Oaks, CA: Sage Publications. Benedito, J., Cárcel, J. A., & Mulet, A. (2001). Cheese hardness by experts and untrained judges. Journal of Sensory Studies, 16, 277–285.
401
Bodyfelt, F. W., Tobias, J., & Trout, G. M. (1988). Sensory evaluation of ice cream and related products. In F. W. Bodyfelt, J. Tobias, & G. M. Trout (Eds.), The sensory evaluation of dairy products (pp. 166–226). New York: Van Nostrand Reinhold. Bower, J. A., & Baxter, I. A. (2003). Sensory properties and consumer perception of ‘home-made’ and commercial dairy ice cream. Journal of Sensory Studies, 18, 217–234. Danzart, M., Sieffermann, J.-M., & Delarue, J. (2004). New developments in preference mapping techniques: Finding out a consumer optimal product, its sensory profile and the key sensory attributes. Seventh sensometrics meeting, Davis, CA, USA. Datamonitor (2007). Ice cream in the United States: Industry profile. Available from www.datamonitor.com. Deliza, R., Macfie, H., & Hedderley, D. (2005). The consumer perception of passionfruit juice using free choice profiling. Journal of Sensory Studies, 20, 17–27. Faye, P., Brémaud, D., Teillet, E., Courcoux, P., Giboreau, A., & Nicod, H. (2006). An alternative to external preference mapping based on consumer perceptive mapping. Food Quality and Preference, 17, 604–614. Gonzáles-Tomás, L., & Costell, E. (2006). Sensory evaluation of vanilla-dairy desserts by repertory grid method and free choice profile. Journal of Sensory Studies, 21, 20–33. González Viñas, M. A., Garrido, N., & Wittig de Penna, E. (2001). Free choice profiling of chilean goat cheese. Journal of Sensory Studies, 16, 239–248. Greenhoff, K., & MacFie, H. J. H. (1999). Preference mapping in practice. In H. J. H. MacFie & D. M. H. Thomson (Eds.), Measurement of food preferences (pp. 137–166). New York: Aspen Publishers, Inc. Guerrero, L., Gou, P., & Arnau, J. (1997). Descriptive analysis of toasted almonds: A comparison between expert and semi-trained assessors. Journal of Sensory Studies, 12, 39–54. Husson, F., & Pagés, J. (2003). Comparison of sensory profiles done by trained and untrained juries: Methodology and results. Journal of Sensory Studies, 18, 453–464. Lancaster, B., & Foley, M. (2007). Determining statistical significance for choose-allthat-apply question responses. Seventh pangborn sensory science symposium, Minneapolis, USA. Lê, S., Pagês, J., & Husson, F. (2008). Methodology for the comparison of sensory profiles provided by several panels: Application to a cross-cultural study. Food Quality and Preference, 19, 179–184. Lelievre, M., Chollet, S., Abdi, H., & Valentin, D. (2008). What is the validity of the sorting task for describing beers? A study using trained and untrained assessors. Food Quality and Preference, 19, 697–703. Li, Z., Marshall, R., Heymann, H., & Fernando, L. (1997). Effect of milk fat content on flavor perception of vanilla ice cream. Journal of Dairy Science, 80, 3133–3141. Meilgaard, M., Civille, G. V., & Carr, B. T. (2007). Sensory evaluation, techniques (4th ed.). New York: CRC Press Inc. Meullenet, J.-F., Lovely, C., Threlfall, R., Morris, J. R., & Striegler, R. K. (2008). An ideal point density plot method for determining an optimal sensory profile for Muscadine grape juice. Food Quality and Preference, 19, 210–219. Meullenet, J.-F., Xiong, R., & Findlay, C. J. (2007). Multivariate: And probabilistic analyses of sensory science problems. Iowa, USA: IFT Press/Blackwell Publishing. Morand, E., & Pagès, J. (2005). Procrustes multiple factor analysis to analyze the overall perception of food products. Food Quality and Preference, 17, 36–42. Nestrud, M. A., & Lawless, H. T. (2008). Perceptual mapping of citrus juices using projective mapping and profiling data from culinary professionals and consumers. Food Quality and Preference, 19, 431–438. Pagès, J. (2004). Multiple factor analysis: Main features and application to sensory data. Revista Colombiana de Estadística, 27(1), 1–26. Punter, P. H. (2008). Bridging the gap between R&D and marketing: The ideal profile method. In Third European conference on sensory and consumer research, Hamburg, Germany. Seo, H.-S., Lee, S.-Y., & Hwang, I. (2009). Development of a sensory attribute pool of brewed coffee. Journal of Sensory Studies, 24, 111–132. Stone, H., & Sidel, J. L. (2007). Sensory research and consumer-led food product development. In H. J. H. MacFie (Ed.), Consumer-led food product development (pp. 307–320). Cambridge, England: Woodhead Publishing Limited. Williams, A. A., & Langron, S. P. (1984). The use of free choice profiling for evaluation of commercial ports. Journal of the Science of Food and Agriculture, 35, 558–568. Williams, E. J. (1949). Experimental designs balanced for the estimation of residual effects of treatments. Australian Journal of Scientific Research, Ser, A2, 149–168. Worch, T., Lê, S., & Punter, P. H. (2008). How reliable are consumers? Comparison of sensory profiles from consumers and experts. Ninth sensometrics meeting, St. Catherines, Ontario, Canada.
Contents lists available at ScienceDirect
Food Quality and Preference journal homepage: www.elsevier.com/locate/foodqual
The application of check-all-that-apply (CATA) consumer profiling to preference mapping of vanilla ice cream and its comparison to classical external preference mapping Lauren Dooley, Young-seung Lee, Jean-François Meullenet * Department of Food Science, University of Arkansas, 2650 N. Young Avenue, Fayetteville, AR 72704, United States
a r t i c l e
i n f o
Article history: Received 19 January 2009 Received in revised form 27 August 2009 Accepted 11 October 2009 Available online 21 October 2009 Keywords: Check-all-that-apply Preference mapping Vanilla ice cream Multiple factor analysis
a b s t r a c t This study was conducted to evaluate the use and efficacy of check-all-that-apply (CATA) data for the creation of preference maps, and to compare these maps to classical external maps generated from traditional sensory profiles. Ten commercial vanilla ice cream products were presented to 80 consumers. Consumers answered an overall liking question using the 9-point hedonic scale and a CATA question with 13 attributes which described the sensory characteristics of vanilla ice cream. A trained descriptive panel of 17 individuals developed a profile of 23 attributes for the vanilla ice cream products. Preference maps created by CATA counts were compared to those by descriptive profiles via multiple factor analysis (MFA). The characterization of the products by both sensory methods showed very good agreement between the methods. The MFA of map configurations showed fair agreement between the techniques used to produce the preference maps, indicating that CATA data applied to preference mapping gave similar results to external preference mapping. Ó 2009 Elsevier Ltd. All rights reserved.
1. Introduction Check-all-that-apply (CATA) questions regarding consumerperceived product attributes have been used in consumer studies to determine what sensory attributes may be characteristic of a specific product (Lancaster & Foley, 2007). Some researchers already advocate the use of consumer sensory profiling to lead product development as an alternative to classical sensory profiling (Punter, 2008; Worch, Lê, & Punter, 2008). The format of the CATA question allows consumers to choose all potential attributes from the given lists to describe the test products. This is different from scaling in the sense that no intensities are given to the attributes. In addition, the descriptors are not constrained to product sensory attributes, but could also be related to product usage or concept fit. This type of methodology has the advantage of gathering information on perceived product attributes without requiring scaling, allowing for a slightly less contrived description of the main sensory properties of the product tested (depending on how the terms are created). The actual generation of CATA terms can be performed in many ways: the consumers can choose words to describe the product during the test (modified free choice profiling), terms can be given
* Corresponding author. Tel.: +1 7853418710. E-mail address: [email protected] (J.-F. Meullenet). 0950-3293/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.foodqual.2009.10.002
by a trained panel, or terms can be generated by consumers not testing the product (i.e. a focus group). Free choice profiling allows consumers to use as many or as few words as necessary to describe the product and evaluate the intensities of the chosen attributes, resulting in a less expensive and more accurate view of consumer perception and acceptance (Deliza, Macfie, & Hedderley, 2005; Gonzáles-Tomás & Costell, 2006; González Viñas, Garrido, & Wittig de Penna, 2001; Williams & Langron, 1984). However, if each consumer selects his/her own terms, the analysis becomes cumbersome since each term must be subjectively interpreted and combined with similar terms (Meilgaard, Civille, & Carr, 2007). Seo, Lee, and Hwang (2009) used consumers to describe sensory characteristics of coffee. Verification of the terms was then conducted by other consumers to confirm that the terms were appropriate and understandable. While this is an effective method, the time required is to complete the test is extensive. Terms generated by a trained panel have the benefit of being more comprehensive and better described, though they may be too complex for the average consumer to understand and could require simplification. Altering the terms in this manner is difficult to do while retaining the correct term description and definition. However, it has been shown that differences in sensory evaluations between trained and untrained (naïve consumers) are minimal (Benedito, Cárcel, & Mulet, 2001; Guerrero, Gou, & Arnau, 1997; Husson & Pagés, 2003; Lelievre, Chollet, Abdi, & Valentin, 2008), so using less obscure terms by a descriptive panel could be a
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401
beneficial tool for creating a CATA list. Ultimately, it is the researcher’s decision as to which method is most appropriate. The CATA method requires minimal instruction, is relatively easy to perform and is completed quickly (Lancaster & Foley, 2007). Furthermore, it could be a more practical approach than intensity scaling from the standpoint of consumer-led product development. Since CATA responses are directly linked to consumers’ perception of product characteristics, these responses could be utilized as supplemental data to maximize acceptance of the targeted products by consumers. CATA provides information on which attributes are detectable according to consumers and how that may relate to their overall liking and acceptance. Understanding sensory characteristics in the process of new product development is of great importance, as failure to obtain correct information about the sensory attributes may lead to fast disappearance of the new products from the marketplace (Stone & Sidel, 2007). To understand the relationship between consumer and sensory data, preference mapping is a useful method. Preference mapping is a widely used group of multivariate statistical techniques designed to optimize products by understanding the structure between consumer preference and sensory data to identify drivers of liking (Faye et al., 2006; Greenhoff & MacFie, 1999). Among the various product optimization mapping methods, the Euclidian Distance Ideal Point Mapping (EDIPM), an extension of Multidimensional Preference Mapping (MDPREF), is a new approach based on a density analysis of individual consumer ideal product placements in the product configuration space (Meullenet, Xiong, & Findlay, 2007). In this approach, the ideal point of individual consumers is the point which minimizes the correlation between Euclidian distances to the products and hedonic scores. Another optimization mapping technique, the Response Surface Model (RSM), proposed by Danzart, is based on external preference mapping (Danzart, Sieffermann, & Delarue, 2004). Multidimensional representation of sensory stimuli is first created by sensory (i.e. external) data. The consumer data for individual consumers is then regressed against the product coordinates in the sensory space to determine ideal points for both the individuals and the group (Meullenet, Lovely, Threlfall, Morris, & Striegler, 2008). To investigate the efficacy of CATA scales within the sensory environment, this study used ice cream as the testing medium. Ice cream is one of the most popular frozen desserts in the United States. The US ice cream market continues to grow and is expected to be valued at over $10 billion by 2012 (Datamonitor, 2007). Vanilla is the most popular ice cream flavor in the US and accounts for almost half of all ice cream sales (Bodyfelt, Tobias, & Trout, 1988). There are many companies producing and commercializing ice cream in the US. In order to compete in this highly competitive market, it is crucial for ice cream manufacturers to understand the strong and weak points of their products, and how consumers’ attitudes and preference patterns affect their products.
395
The objectives of this study were to (1) assess the use of CATA attribute responses for 10 commercial vanilla ice creams as an alternative to consumer attribute intensity ratings, and (2) compare CATA-generated preference maps to classical external maps generated from traditional sensory profiles.
2. Materials and methods 2.1. Samples and sample preparation Fifteen commercially-available ice creams were initially selected from local supermarkets for testing. Preliminary screening of texture and flavor attributes eliminated five samples due to brand replication and fat content, a popular indicator of ice cream quality, and the use of natural or artificial vanilla flavor. Ice creams were selected so that various combinations of these quality factors were represented in the study. The 10 remaining products, consisting of two high-fat products, six products with moderate fat content and two low-fat products, are detailed in Table 1. One scoop of each product was placed individually into a lidded white plastic container (45 mm diameter) coded with a three-digit random number. Samples were stored in a commercial-grade freezer (TS-49, True Manufacturing Co., St. Louis, MO, USA) at 18 °C for at least 24 h prior to testing to ensure sample consistency. All samples were tempered for 2 min at room temperature prior to serving for both descriptive analysis and consumer testing. The 2 min increment was determined to be the most appropriate tempering time by observing the condition of the ice cream as a function of time at room temperature. Samples were presented in a sequential monadic order to panelists according to a complete randomized block design, and the serving temperature ( 12 + 2 °C) was strictly monitored to maintain consistency (Bower & Baxter, 2003; Li, Marshall, Heymann, & Fernando, 1997). 2.2. Descriptive analysis The 10 vanilla ice creams were evaluated for taste, aromatic, flavor, and texture attributes by a descriptive panel of 17 individuals trained by the SpectrumÒ method (Sensory Spectrum Inc., Chatham, NJ, USA). Panelists have over 100 h of training and an average of 1000 h of testing experience. Two orientation sessions were conducted to familiarize the panelists with the samples. Flavor and texture lexicons were developed in four sessions, as described in Tables 2 and 3, respectively. The lexicons consisted of 23 total attributes specific to vanilla ice cream and definitions of each sensory attribute with associated references. Panelists quantified all attributes on a line scale from 0 to 15 (Meilgaard et al., 2007). Unsalted crackers and water were provided for panelists to clean and rinse their palate between samples, and a 10 min break helped prevent fatigue. The flavor attribute testing for all 10 products was
Table 1 A list of 10 commercial vanilla ice cream products. Brand
Code
Name/description
Fat content (%)
Flavor
Blue Bell Blue Bunny Ben and Jerry’s Best Choice Breyers Edy’s ‘‘Grand” Great Value Guilt Free Haagen-Dazs Yarnell’s
A B C D E F G H I J
Homemade vanilla Premium all natural vanilla Vanilla Vanilla Natural vanilla Rich and creamy vanilla Vanilla Vanilla Vanilla Homemade vanilla
13 10 24 11 12 5 11 4 28 15
Natural and Natural and Natural Artificial Natural Natural Artificial Natural and Natural Natural and
Manufacturer artificial artificial
artificial artificial
Blue Bell Creameries Wells’ Dairy, Inc. Ben and Jerry’s Homemade Holdings, Inc. (Unilever) Wal-Mart Stores Inc. Unilever Nestle Wal-Mart Stores Inc. Yarnell Ice Cream Co. Nestle Yarnell Ice Cream Co.
396
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401
Table 2 Flavor lexicon for vanilla ice cream. Term Basic taste Sweet
Salt
Reference
Intensity
The basic taste, perceived on the tongue, stimulated by sugars and high potency sweeteners
Solutions of sucrose in spring water
2–2.0%
5–5.0%
The basic taste, perceived on the tongue, stimulated by sodium salt, especially sodium chloride
Solutions of sodium chloride in spring water
10–10.0% 0.2–2.0%
16–15.0% 0.35–5.0%
0.5–8.5% 0.7–15.0% 0.05–2.0%
0.55–10.0%
0.15–10.0% 0.05–2.0%
0.20–15.0% 0.08–5.0%
0.15–10.0%
0.20–15.0%
Whole milk 1/2 and 1/2 Heavy whip cream
4.0 6.5 9.0
Alum –6.0
Tea –9.0
Sour
The basic taste, perceived on the tongue, stimulated by acids, such as citric acid
Solutions of citric acid in spring water
Bitter
The basic taste, perceived on the tongue, stimulated by substances such as quinine, caffeine, and certain other alkaloids
Solutions of caffeine in spring water
Aromatics Vanillin Cooked milk Milky Buttery/fat
Flavor Non fat dry milk Caramelized Oxidized Woody/stick Metallic Feeling factors Astringent
a
Definition
The sweet, vanilla-like aromatic characteristic of ethyl vanillin or imitation vanillas The aromatic associated with the flavor of milk, heated to the scalding point The aromatic associated with skim or whole milk products or milk derived products The aromatic associated with fresh butterfat; sweet cream
The aromatic associated with boxed, nonfat dry milk or milk reconstituted from dry milk solids, cardboardy A sweet aromatic characteristic of browned sugars and other carbohydrates The aromatics associated with slightly oxidized fats and oils The aromatic associated with dry fresh cut wood; balsamic or bark-like (1) The aromatic associated with metals, tinny or iron (2) a flat feeling factor stimulated on the tongue by metal The feeling factor on the tongue or other skin surfaces of the mouth described as puckering or drying
0.08–5.0%
Universal scalea
Heavy whipping Cream Cream cheese
Universal scalea
(1) Alum solution (0.01%) (2) tea in 1000 ml spring water for 5 min
Soda note (soda cracker 3.0), cooked apple note (applesauce 7.0), orange hit (orange juice 10.0), cooked grape note (grape juice 14.0), cinnamon hit (chewing gum 16.0).
conducted on two consecutive days, followed by two consecutive days for texture attribute testing.
The CATA counts were totaled for each product (Table 5) and the resulting contingency table was used in subsequent analyses.
2.3. Consumer testing
2.4. Statistical analyses
Eighty consumers were recruited to participate in the vanilla ice cream test at the University of Arkansas Sensory Service Center (Fayetteville, AR). Qualification criteria included adults over 18 years of age and vanilla ice cream product consumption at least one to two times per week. The test was sequential monadic and sample presentation orders were balanced in each serving position using the Williams Latin Square design (Williams, 1949). Each consumer tested five samples the first day and tested the remaining five the second day. The sessions were divided into 30 min increments, with most consumers finishing within 20 min. Consumers were asked to evaluate overall liking, appearance, flavor and texture attributes of each sample on a 9-point hedonic scale (1 = ‘‘dislike extremely”, 9 = ‘‘like extremely”). The attributes of scoopability, color, sweetness, vanilla flavor, creamy flavor, smoothness, melting in the mouth, melting in the bowl and hardness of each product were evaluated using the 5-point ‘‘Just About Right” (JAR) scale (data not shown or used in this study). Though not used in these analyses, since this JAR data was collected, the authors felt it necessary to mention. The final question of the survey listed common ice cream attributes and asked the consumers to check all attributes that applied to the given sample (Table 4).
External preference mapping was performed according to Danzart et al. (2004) using either the descriptive sensory profiles or CATA counts to determine a group ideal point (Meullenet et al., 2008). First, principal component analysis was performed on mean sensory profiles while correspondence analysis was performed on the CATA counts. To determine the area of the map maximizing the number of consumers satisfied, a quadratic model was constructed (i.e. regressing hedonic scores against the principal component scores), and the area of acceptability for each consumer was identified (i.e. area of the map where the hedonic score was predicted above the mean score for each consumer). The area of maximum density was regarded as the ideal point solution for this method. Both analyses are considered here as external because the product configurations were obtained from data other than liking. Since the descriptive profiles and CATA counts were scaled differently, the two data sets were standardized across all products prior to external preference mapping to minimize differences inherent to the scaling. To standardize the data, the data matrix had products in columns and attributes in rows. The data was standardized by columns so that the mean was zero and the variance was one.
397
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401 Table 3 Texture lexicon for vanilla ice cream. Term
Definition
Reference
Intensity
Scoopability/ manual firmness
The force required to cut the sample with your spoon. 2 and 4 min
Evaluate the force required to remove one/spoonful of sample from the cup (soft–hard)
Peanut butter
4.0
Cream cheese Metal spoon sample and references
8.0
Peanut butter
4.0
Cream cheese Plastic spoon
9.0
Marshmallow cream
2.5
Nougat Plastic spoon
4.0
Hardness-oral
Denseness
The force required to compress the sample between the tongue and mouth roof. 2 min
The amount of air or fluffiness perceived in the sample. 2 min
Compress through sample one time with tongue. Evaluate the force required to fully compress the sample (soft–hard)
Compress through sample one time with tongue. Evaluate the amount of air perceived in the sample (airy–compact)
Degree of ice
The amount of ice crystals felt in the mouth during the chew. 2 min
Compress 1/2 tsp of the sample. Evaluate the amount of ice crystals perceived in the sample (no ice–much ice)
Sherbet Plastic spoon
3.0
Smoothness
The amount of particles perceived in the sample during the chew. 2 min
Manipulate the sample three times with tongue. Evaluate the amount of particles perceived in the sample (not smooth– smooth)
Peanut butter
5.0
Cool whip Plastic spoon
14.0
Soy ice cream
5.0
Blue Bunny Plastic spoon
9.0
Saltine
0.5
Ritz Pringles Mayo
2.5 5.0 9.0
Frosting Three pulls Plastic Spoon
10.0
Rate of melt
Mouth coat
Elasticity
The rate in which the ice cream changes forms from a solid to a liquid. 2 min
The amount and degree of residue felt by the tongue when moved over the surface of the mouth. 2 min
The degree to which the samples appears to have an elastic/doughy impression. 4 min only
Place 1/2 tsp of sample in mouth and evaluate the rate in which the sample melts (slow–fast)
Expectorate the sample and feel the surface of the mouth with the tongue to evaluate (none–much)
Stick spoon in the sample and pull out. After three pulls evaluate the amount of sample pulled up by the spoon (none– much)
Table 4 An example of check-all-that-apply (CATA) question. Check all attributes that describe this sample: h Buttery h Sweet h Milk/dairy flavor h Custard/eggy flavor h Corn syrup h Artificial vanilla h Natural vanilla h Creamy flavor h Soft h Hard h Gummy h Icy h Creamy/smooth
Multiple factor analysis (MFA) was conducted using FactoMineR in R (v.2.6.2, 2008) to examine the similarities first between the descriptive profiles and CATA counts (Figs. 1 and 2) and second between the multivariate product configurations obtained from the three preference mapping techniques employed. For each preference mapping technique, the coordinates of the ideal points in the maps were estimated and used in the MFA analysis as illustrative data. MFA is a useful statistical technique to analyze the similarity of a set of observations explained by several different groups of variables on comparable or contradictory scales (Abdi & Valentin, 2007). MFA is able to balance the influence of each variable, can compare multiple data sets, and can demonstrate patterns of attribute correlation (Lê, Pagês, & Husson, 2008; Morand & Pagès, 2005; Nestrud & Lawless, 2008).
3. Results and discussion Internal preference mapping was also conducted as a point of comparison for descriptive analysis and CATA-based external preference mapping. The comparison seems appropriate since the CATA-based external preference can be considered as a hybrid method. Euclidian Distance Ideal Point Mapping (EDIPM, Meullenet et al., 2008) was used for internal preference mapping. The product configuration in the space was derived from principal component analysis of the centered overall liking data.
3.1. Comparison of product descriptions by CATA and descriptive analysis Individual product maps were created by MFA using descriptive sensory profiles and consumer CATA counts (Fig. 1). Overall, MFA comparing the characterization of the products by both profiles showed agreement between the two methods, although only 51%
398
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401
Table 5 Total counts of check-all-that-apply attributes for each product. Brand
Soft
Hard
Gummy
Icy
Creamy/ smooth
Buttery
Sweet
Milk/dairy flavor
Custard/eggy flavor
Corn syrup
Natural vanilla
Artificial vanilla
Creamy flavor
Blue Bell Blue Bunny Ben and Jerry’s Best Choice Breyers Edy’s ‘‘Grand” Great Value Guilt Free Haagen-Dazs Yarnell’s
42 37 21 29 8 25 51 34 17 43
20 20 31 29 43 38 11 22 32 16
5 7 7 4 7 4 12 3 7 4
37 16 22 28 61 20 8 29 14 14
31 45 34 37 10 35 59 35 40 51
44 46 29 43 17 39 38 30 30 21
58 53 53 61 57 61 46 52 61 60
60 58 50 62 52 59 58 57 52 54
45 35 28 32 25 26 21 22 35 22
10 13 16 13 11 16 12 23 21 15
25 28 27 34 37 28 27 19 19 24
39 33 29 33 37 43 31 45 49 42
45 51 44 54 24 53 52 35 43 47
Fig. 1. Multiple factor analysis individual product plots using descriptive sensory profiles (D) and CATA counts (C). Product codes are listed in Table 1.
of the variability was explained by the first two MFA dimensions. A and J showed the largest variance between the two methods, largely due to the disagreement between the descriptive and consumer (CATA) maps for these two products (not shown). As discussed by Pagès (2004), based on absolute scores of partial lines on each product for the first dimension, Fig. 1 indicates that products B, H, I, and J were more characterized for the first dimension by descriptive analysis profiles than CATA counts. If a line is drawn from the end of the descriptive and CATA vector for each of these products to the x-axis, descriptive analysis rates higher on the first dimension. Products E and G were more characterized by CATA counts for the first dimension. Fig. 2 shows the variable correlation circle obtained by MFA comparing descriptive analysis profiles and CATA counts. The vectors showed a strong correlation for descriptive (d) mouth coat, smoothness and rate of melt, with CATA counts (c) perceived soft, creamy flavor and creamy/smooth, between degree of ice (d) and icy (c), elasticity (d) and gummy (c), and between caramelized (d) and corn syrup (c). The opposite vector directions for some CATA and descriptive descriptors with opposite meanings also show agreement between the two methods. For example, hardness (c) was opposite to softness (c), and artificial vanilla opposite to
Fig. 2. Multiple factor analysis variable correlation circle obtained using descriptive analysis and CATA terms. Refer to Tables 2–4 for complete attribute descriptions. ‘CS_c’ and ‘Cflavor_c’ represent creamy/smooth and creamy flavor, respectively, in CATA terms.
natural vanilla flavor. No correlations were observed between hard (c) and icy (c), and gummy (c), or between milky (d) and butterfat (d). Descriptive sensory profiles did not show any correlation between bitter (d) and sweet (d) taste in the vanilla ice cream products, while bitter (d) had negative correlation with salt (d) despite its low loading. The most influential attributes (i.e. highest loadings) were found to be sweet, bitter, vanillin, degree of ice, elasticity, and smoothness for descriptive sensory profiles. Icy, natural vanilla, creamy/smooth, creamy flavor and artificial vanilla attributes for CATA consumer profiles played a relatively more important role in determining product locations in the map. 3.2. Preference mapping results and group ideal point locations Each of the three preference mapping techniques employed (external mapping on descriptive data and CATA and internal preference mapping) allow the identification of a group ideal product location in the maps. This point, in all three cases, is the location in the map maximizing the percentage of consumers who would be satisfied by a product placed at that location. Ideally, the three methods would give approximately the same answer and this is what we seek to assess here.
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401
399
Fig. 3. Results of external preference mapping using (a) descriptive analysis profiles (DD) and (b) CATA term counts (DC). The virtual product labeled ‘‘Opt” represents the location in the map maximizing the percentage of consumers satisfied (Danzart et al., 2004). Other product codes are as in Table 1. Values in parenthesis represent mean overall liking values.
Fig. 4. Results of internal preference mapping. The virtual product labeled ‘‘Opt” represents the location in the map maximizing the percentage of consumers satisfied (Euclidian Distance Ideal Point Mapping (EDIPM), Meullenet et al., 2007). Other product codes are listed in Table 1. Values in parenthesis represent mean overall liking values.
400
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401
The results of external preference mapping using descriptive sensory data (DD) and CATA counts (DC), and internal preference mapping are graphically shown in Figs. 3 and 4. The first two dimensions of the maps explained 50.2% and 59.0% of the variance in descriptive profiles and CATA counts, respectively. The average individual consumer fit was similar for the DD map (R2 = 0.59) and the DC map (R2 = 0.61) (data not shown). Overall, the products’ spatial representation on the first two dimensions seemed to differ. The product coordinates used to regress individual overall liking scores for the external maps were obtained from two different data sources and the disagreement is not too surprising. For the DD map (Fig. 3a), the ideal product was closest to product F (OL = 6.7), followed by product D (OL = 6.8). Product E (OL = 5.5), least liked by consumers, was furthest from the other products, including the optimal product. The DC map (Fig. 3b) placed the group ideal point near products A (OL = 6.3) and D (OL = 6.8), while product E was located on the far left of the map by itself, showing agreement with the DD map. Product D, which was most liked by consumers, was in closest proximity with the group ideal point for both maps. For internal preference mapping, EDIPM placed the ideal point closest to products A and D and furthest from product E (Fig. 4). This finding was similar to that of the DC map. To more finely compare the level of agreement/disagreement between the three maps, MFA was employed using the first two dimensions of the three maps created. Fig. 5 represents the locations of the 10 commercial vanilla ice cream products and the ideal product (i.e. individual factor map) determined for each of the three types of preference mapping on the first two MFA factors. As shown in Fig. 5, the optimal product was in close proximity with products A (OL = 6.3), D (OL = 6.8), and F (OL = 6.7), while E (OL = 5.5) was furthest from the remaining products and the ideal product. Overall, MFA of the three product maps showed fair agreement between the approaches employed. However, the DD map was contrary to the other two methods for products A and J, while the internal map showed dissimilarities to DD and DC, par-
Fig. 5. Multiple factor analysis individual product plots of the product configurations (first two dimensions) determined for external preference mapping using descriptive analysis data (DD) and CATA counts (DC) and internal preference mapping (EDIPM). The virtual product labeled ‘‘Opt” represents the optimal product derived from the three mapping methods employed. Other product codes are listed in Table 1.
ticularly for product F. This could be explained by the fact that product F was near the origin in both DD and DC maps, but was located in the far right upper quadrant in the EDIPM map (see Figs. 3 and 4). One key result of this analysis is that optimal product location showed little variation between the three maps. 3.3. Ideal product profiles Although the location of the group ideals in the three preference maps was reported to be fairly invariable, there is a need to determine the sensory characteristics that should be exhibited by the group ideal products. If the three methods are in agreement, we would expect the sensory profiles of the group ideals to be fairly similar. To determine the level of agreement between the preference mapping methods employed to determine the group ideals in the three maps, the CATA counts were regressed against the first two dimensions of the product spaces created using descriptive analysis data, CATA data and internal preference mapping. From these models, the ideal product CATA profiles were predicted. This is also known as reverse regression. Fig. 6a gives the fit of the CATA attributes for the three mapping methods employed. Overall, the CATA attributes were better fitted in the CATA space, particularly natural vanilla, creamy flavor, creamy/smooth, soft and hard attributes (R2 > 0.6). This is not surprising since this product space was derived from the CATA data. Overall, the CATA attributes were not as well fitted
Fig. 6. Ideal vanilla ice cream profiles according to descriptive, CATA and EDIPM data using (a) CATA attribute fit (R2) and (b) normalized ideal CATA counts.
L. Dooley et al. / Food Quality and Preference 21 (2010) 394–401
in the product/preference space derived from internal preference mapping while the quality of the fit was somewhere in between for the product space created from descriptive analysis data. This result is also not surprising since both the CATA and descriptive analysis product spaces are based on product sensory properties while the internal preference mapping product space is based on liking data. Fig. 6b represents the ideal sensory profiles in terms of normalized CATA counts for the three mapping methods employed. The normalization of these ideal profiles are easier to interpret than ideal raw CATA counts because they show if the ideal CATA count is above or below the mean of the observed counts. With normalized CATA counts, a positive value indicates that the ideal product was above the mean CATA count across products for a given attribute. These profiles allow the determination of the level of agreement between the three mapping methods employed. From Fig. 6b, we conclude that the three mapping methods provided similar optimal values for gummy, hard, soft, sweet, buttery, creamy/ smooth, icy, creamy flavor and custard/eggy flavor. However, there was disagreement for natural vanilla, artificial vanilla and corn syrup. For these attributes, the CATA methodology recommended high natural vanilla, low artificial vanilla and low corn syrup CATA counts.
4. Conclusions Overall, the characterization of the 10 commercial vanilla ice cream products shows good agreement between descriptive sensory profiles and consumer-perceived CATA profiles. Moreover, the CATA attribute data applied to preference mapping gave similar results to external preference mapping. The advantage of this technique is that the task asked of consumers is simple (i.e. when compared to intensity ratings), and that the responses may be more spontaneous than when intensities are rated. The limitation of this approach is that the optimal profile derived from the CATA maps is in terms of response counts and not intensities as given by a trained panel or consumers using attribute intensity scaling. Since it can be hypothesized that the three maps created could have been more similar for product configurations, including optimal points, with additional dimensions being considered, the use of additional PCs may be of interest in future studies. Although CATA questions seem to have some validity to characterize the sensory properties of products as perceived by consumers, many questions remain unanswered. Further studies may include the assessment of the effects of order and number of terms CATA questions on attribute selection.
References Abdi, H., & Valentin, D. (2007). Multiple factor analysis (MFA). In N. Salkind (Ed.), Encyclopedia of measurement and statistics. Thousand Oaks, CA: Sage Publications. Benedito, J., Cárcel, J. A., & Mulet, A. (2001). Cheese hardness by experts and untrained judges. Journal of Sensory Studies, 16, 277–285.
401
Bodyfelt, F. W., Tobias, J., & Trout, G. M. (1988). Sensory evaluation of ice cream and related products. In F. W. Bodyfelt, J. Tobias, & G. M. Trout (Eds.), The sensory evaluation of dairy products (pp. 166–226). New York: Van Nostrand Reinhold. Bower, J. A., & Baxter, I. A. (2003). Sensory properties and consumer perception of ‘home-made’ and commercial dairy ice cream. Journal of Sensory Studies, 18, 217–234. Danzart, M., Sieffermann, J.-M., & Delarue, J. (2004). New developments in preference mapping techniques: Finding out a consumer optimal product, its sensory profile and the key sensory attributes. Seventh sensometrics meeting, Davis, CA, USA. Datamonitor (2007). Ice cream in the United States: Industry profile. Available from www.datamonitor.com. Deliza, R., Macfie, H., & Hedderley, D. (2005). The consumer perception of passionfruit juice using free choice profiling. Journal of Sensory Studies, 20, 17–27. Faye, P., Brémaud, D., Teillet, E., Courcoux, P., Giboreau, A., & Nicod, H. (2006). An alternative to external preference mapping based on consumer perceptive mapping. Food Quality and Preference, 17, 604–614. Gonzáles-Tomás, L., & Costell, E. (2006). Sensory evaluation of vanilla-dairy desserts by repertory grid method and free choice profile. Journal of Sensory Studies, 21, 20–33. González Viñas, M. A., Garrido, N., & Wittig de Penna, E. (2001). Free choice profiling of chilean goat cheese. Journal of Sensory Studies, 16, 239–248. Greenhoff, K., & MacFie, H. J. H. (1999). Preference mapping in practice. In H. J. H. MacFie & D. M. H. Thomson (Eds.), Measurement of food preferences (pp. 137–166). New York: Aspen Publishers, Inc. Guerrero, L., Gou, P., & Arnau, J. (1997). Descriptive analysis of toasted almonds: A comparison between expert and semi-trained assessors. Journal of Sensory Studies, 12, 39–54. Husson, F., & Pagés, J. (2003). Comparison of sensory profiles done by trained and untrained juries: Methodology and results. Journal of Sensory Studies, 18, 453–464. Lancaster, B., & Foley, M. (2007). Determining statistical significance for choose-allthat-apply question responses. Seventh pangborn sensory science symposium, Minneapolis, USA. Lê, S., Pagês, J., & Husson, F. (2008). Methodology for the comparison of sensory profiles provided by several panels: Application to a cross-cultural study. Food Quality and Preference, 19, 179–184. Lelievre, M., Chollet, S., Abdi, H., & Valentin, D. (2008). What is the validity of the sorting task for describing beers? A study using trained and untrained assessors. Food Quality and Preference, 19, 697–703. Li, Z., Marshall, R., Heymann, H., & Fernando, L. (1997). Effect of milk fat content on flavor perception of vanilla ice cream. Journal of Dairy Science, 80, 3133–3141. Meilgaard, M., Civille, G. V., & Carr, B. T. (2007). Sensory evaluation, techniques (4th ed.). New York: CRC Press Inc. Meullenet, J.-F., Lovely, C., Threlfall, R., Morris, J. R., & Striegler, R. K. (2008). An ideal point density plot method for determining an optimal sensory profile for Muscadine grape juice. Food Quality and Preference, 19, 210–219. Meullenet, J.-F., Xiong, R., & Findlay, C. J. (2007). Multivariate: And probabilistic analyses of sensory science problems. Iowa, USA: IFT Press/Blackwell Publishing. Morand, E., & Pagès, J. (2005). Procrustes multiple factor analysis to analyze the overall perception of food products. Food Quality and Preference, 17, 36–42. Nestrud, M. A., & Lawless, H. T. (2008). Perceptual mapping of citrus juices using projective mapping and profiling data from culinary professionals and consumers. Food Quality and Preference, 19, 431–438. Pagès, J. (2004). Multiple factor analysis: Main features and application to sensory data. Revista Colombiana de Estadística, 27(1), 1–26. Punter, P. H. (2008). Bridging the gap between R&D and marketing: The ideal profile method. In Third European conference on sensory and consumer research, Hamburg, Germany. Seo, H.-S., Lee, S.-Y., & Hwang, I. (2009). Development of a sensory attribute pool of brewed coffee. Journal of Sensory Studies, 24, 111–132. Stone, H., & Sidel, J. L. (2007). Sensory research and consumer-led food product development. In H. J. H. MacFie (Ed.), Consumer-led food product development (pp. 307–320). Cambridge, England: Woodhead Publishing Limited. Williams, A. A., & Langron, S. P. (1984). The use of free choice profiling for evaluation of commercial ports. Journal of the Science of Food and Agriculture, 35, 558–568. Williams, E. J. (1949). Experimental designs balanced for the estimation of residual effects of treatments. Australian Journal of Scientific Research, Ser, A2, 149–168. Worch, T., Lê, S., & Punter, P. H. (2008). How reliable are consumers? Comparison of sensory profiles from consumers and experts. Ninth sensometrics meeting, St. Catherines, Ontario, Canada.